COVID-19 prognosis using clinical tabular data faces significant challenges due to missing values and class imbalance issues. Existing methods often overlook the complex high-order interrelationship among clinical attributes and struggle with training stability on imbalanced datasets. We propose ACGM, an attribute-centric graph modeling network that simultaneously addresses missing data imputation and COVID-19 prognosis. ACGM consists of three key modules: an attributes preprocessing module (APM) for coarse-grained imputation initialization, a graph-enhanced attributes imputation module (GEAIM) that models high-order inter-attribute relationships through graph structures, and a graph-enhanced disease prognosis module (GEDPM) that leverages these complex attribute interactions for final prediction. GEAIM and GEDPM employ a mean-teacher strategy with attributes graph matching to preserve high-order relationships, enhance training stability, and maintain structural integrity of attribute interactions. Extensive experiments are conducted on four public COVID-19 tabular datasets, demonstrating the superiority of our ACGM over existing methods. Through comprehensive interpretability analysis, we identify that attributes such as LDH, Difficulty In Breathing, and SaO2 significantly impact COVID-19 prognosis, aligning well with clinical insights and radiologist assessments.

Using artificial intelligence (AI) offers opportunities to analyze medical images and to support early cancer detection. For instance, neural networks, in different implementation, can be used to analyze data image parts (i.e., voxels), defining a trained network useful for lung cancer nodules detection. We present our experience in designing and testing a transformerbased deep learning architecture, aiming to detect pulmonary cancer nodule candidates using 3D Computed Tomography (CT) images. The module also includes a preprocessing pipeline based on dynamic sampling of voxels extracted from images, to support data filtering and results explainability. The proposed architecture has been implemented, trained, and tested using the LUNA16 publicly available dataset. Experimental results proved both high effectiveness and competitive performance metrics across standard evaluations. Trained module can be used on a large CT dataset aiming to support clinicians in lung cancer early detection as well as to support in followup for lung cancer patients treatments. This work represent, indeed, results for preliminary applications in a research project (Advancing Lung Cancer Screening: Artificial Intelligence, Multimodal Imaging and Cutting-Edge Technologies for Early Detection and Characterization), conducted in collaboration with San Raffaele Hospital (Italy), Campus Biomedico University (Italy) and University Hospital of Salerno.

Accurate lung tumor segmentation is crucial for improving diagnosis, treatment planning, and patient outcomes in oncology. However, the complexity of tumor morphology, size, and location poses significant challenges for automated segmentation. This study presents a comprehensive benchmarking analysis of deep learning-based segmentation models, comparing traditional architectures such as U-Net and DeepLabV3, selfconfiguring models like nnUNet, and foundation models like MedSAM, and MedSAM 2. Evaluating performance across two lung tumor segmentation datasets, we assess segmentation accuracy and computational efficiency under various learning paradigms, including few-shot learning and fine-tuning. The results reveal that while traditional models struggle with tumor delineation, foundation models, particularly MedSAM 2, outperform them in both accuracy and computational efficiency. These findings underscore the potential of foundation models for lung tumor segmentation, highlighting their applicability in improving clinical workflows and patient outcomes.

The silent progression of cardiovascular disease (CVD) is a major problem particularly in CVD prevention. New techniques enabled by the rise of electronic health records may facilitate CVD prevention. Both public health research and big data applications, such as digital twins, are dependent on access to longitudinal and sensitive data; a challenge which may be facilitated by access to longitudinal synthetic data. In this study, we establish a fidelity benchmark for longitudinal synthetic data by extending a well-known method for cross-sectional synthetic data to a longitudinal application within CVD. We find that the univariate distributional difference between the real and the synthetic data is kept low and that pairwise relations are preserved in the synthetic data. Further, we see that the variablewise temporal trends are preserved, yet may be more extensively studied and have some room for improvement. The results of this study is important to enable future studies within public health prevention and cardiovascular digital twins.

Emphysema is a hallmark of Chronic Obstructive Pulmonary Disease and an independent risk factor for lung cancer. Computed Tomography (CT) is the main diagnostic platform for identifying emphysema. In clinical practice, the quantitative assessment identifies emphysema as low attenuation areas under a specific cut-off threshold set to -950 Hounsfield Unit. Despite its wide adoption, this method lacks consensus on an optimal cut-off threshold and is prone to measurement variation, asking for new solutions that encompass this limitation. We propose a hybrid deep learning approach for emphysema classification that combines convolutional neural networks for local feature extraction with MAMBA's capability to model long-range dependencies. This fusion ensures a complementary feature representation, capturing both fine-grained and global contextual information. Furthermore, we demonstrate the effectiveness of self-supervised pretraining in domain-specific data, refining the weight configuration of the model to better align with the target distribution and improve its performance during supervised training. The results show on the SCAPIS public cohort that our hybrid model not only outperforms the traditional LAV950 method for emphysema quantification but also surpasses two well-established deep learning architectures. The code is available at: https://github.com/TrainLaboratory/Emphysema.

Generating positron emission tomography (PET) images from computed tomography (CT) scans via deep learning offers a promising pathway to reduce radiation exposure and costs associated with PET imaging, improving patient care and accessibility to functional imaging. Whole-body image translation presents challenges due to anatomical heterogeneity, often limiting generalized models. We propose a framework that segments whole-body CT images into four regions, i.e., head, trunk, arms, and legs, and uses district-specific Generative Adversarial Networks (GANs) for tailored CT-to-PET translation. Synthetic PET images from each region are stitched together to reconstruct the whole-body scan. Comparisons with a baseline non-segmented GAN and experiments with Pix2Pix and CycleGAN architectures tested paired and unpaired scenarios. Quantitative evaluations at district, whole-body, and lesion levels demonstrated significant improvements with our district-specific GANs. Pix2Pix yielded superior metrics, ensuring precise, high-quality image synthesis. By addressing anatomical heterogeneity, this approach achieves state-of-the-art results in whole-body CT-to-PET translation. This methodology supports healthcare Digital Twins by enabling accurate virtual PET scans from CT data, creating virtual imaging representations to monitor, predict, and optimize health outcomes.

Effective management of pediatric diabetes remains a clinical challenge, particularly due to the onset of lipodystrophy resulting from repeated insulin injections and inadequate rotation of injection sites. These complications adversely affect subcutaneous tissue integrity and insulin pharmacokinetics, ultimately compromising glycemic control. In this study, we introduce diPen, a novel smart case designed to integrate with commercial insulin pens and equipped with a dual-sensor system: an optical module for non-invasive lipodystrophy detection and an inertial measurement unit for monitoring injection-site rotation. To evaluate the feasibility of the proposed approach, a clinical study was conducted involving a pediatric cohort. The system employs a personalized machine learning pipeline, leveraging a leave-one-acquisition-out validation strategy to replicate realworld deployment scenarios, wherein newly acquired data from the same subject are evaluated without prior exposure during training. The system demonstrated promising performance in both detection and classification tasks, suggesting that di-Pen may represent a viable tool for enhancing insulin therapy through personalized, data-driven injection guidance and tissue health monitoring.

Contrast-Enhanced Spectral Mammography (CESM) is a dual-energy mammographic technique that improves lesion visibility through the administration of an iodinated contrast agent. It acquires both a low-energy image, comparable to standard mammography, and a high-energy image, which are then combined to produce a dual-energy subtracted image highlighting lesion contrast enhancement. While CESM offers superior diagnostic accuracy compared to standard mammography, its use entails higher radiation exposure and potential side effects associated with the contrast medium. To address these limitations, we propose Seg-CycleGAN, a generative deep learning framework for Virtual Contrast Enhancement in CESM. The model synthesizes high-fidelity dual-energy subtracted images from low-energy images, leveraging lesion segmentation maps to guide the generative process and improve lesion reconstruction. Building upon the standard CycleGAN architecture, Seg-CycleGAN introduces localized loss terms focused on lesion areas, enhancing the synthesis of diagnostically relevant regions. Experiments on the CESM@UCBM dataset demonstrate that Seg-CycleGAN outperforms the baseline in terms of PSNR and SSIM, while maintaining competitive MSE and VIF. Qualitative evaluations further confirm improved lesion fidelity in the generated images. These results suggest that segmentation-aware generative models offer a viable pathway toward contrast-free CESM alternatives.

This study proposes a novel approach combining Multimodal Deep Learning with intrinsic eXplainable Artificial Intelligence techniques to predict pathological response in non-small cell lung cancer patients undergoing neoadjuvant therapy. Due to the limitations of existing radiomics and unimodal deep learning approaches, we introduce an intermediate fusion strategy that integrates imaging and clinical data, enabling efficient interaction between data modalities. The proposed Multimodal Doctor-in-the-Loop method further enhances clinical relevance by embedding clinicians’ domain knowledge directly into the training process, guiding the model’s focus gradually from broader lung regions to specific lesions. Results demonstrate improved predictive accuracy and explainability, providing insights into optimal data integration strategies for clinical applications.

Computed Tomography (CT) plays a pivotal role in medical diagnosis; however, variability across reconstruction kernels hinders data-driven approaches, such as deep learning models, from achieving reliable and generalized performance. To this end, CT data harmonization has emerged as a promising solution to minimize such non-biological variances. In this context, Generative Adversarial Networks (GANs) have proved as a powerful framework for harmonization, framing it as a style-transfer problem. However, GAN-based approaches still face limitations in capturing complex relationships within the images, which are essential for effective harmonization. In this work, we propose a novel texture-aware StarGAN for CT data harmonization, enabling one-to-many translations across different reconstruction kernels. Although the StarGAN model has been successfully applied in other domains, its potential for CT data harmonization remains unexplored. Furthermore, our approach introduces a multi-scale texture loss function that embeds texture information across different spatial and angular scales into the harmonization process, effectively addressing kernel-induced texture variations. We conducted extensive experimentation on a publicly available dataset accounting for xx chest CT scans distributed over three different reconstruction kernels, demonstrating the superiority of our method over the baseline StarGAN.

Multimodal deep learning harnesses diverse imaging modalities, such as MRI sequences, to enhance diagnostic accuracy in medical imaging. A key challenge is determining the optimal timing for integrating these modalities—specifically, identifying the network layers where fusion modules should be inserted. Current approaches often rely on manual tuning or exhaustive search, which are computationally expensive without any guarantee of converging to optimal results. We propose a sequential forward search algorithm that incrementally activates and evaluates candidate fusion modules at different layers of a multimodal network. At each step, the algorithm retrains from previously learned weights and compares validation loss to identify the best-performing configuration. This process systematically reduces the search space, enabling efficient identification of the optimal fusion timing without exhaustively testing all possible module placements. The approach is validated on two multimodal MRI datasets, each addressing different classification tasks. Our algorithm consistently identified configurations that outperformed unimodal baselines, late fusion, and a brute-force ensemble of all potential fusion placements. These architectures demonstrated superior accuracy, F-score, and specificity while maintaining competitive or improved AUC values. Furthermore, the sequential nature of the search significantly reduced computational overhead, making the optimization process more practical. By systematically determining the optimal timing to fuse imaging modalities, our method advances multimodal deep learning for medical imaging. It provides an efficient and robust framework for fusion optimization, paving the way for improved clinical decision-making and more adaptable, scalable architectures in medical AI applications.

Generative models have revolutionized Artificial Intelligence (AI), particularly in multimodal applications. However, adapting these models to the medical domain poses unique challenges due to the complexity of medical data and the stringent need for clinical accuracy. In this work, we introduce a framework specifically designed for multimodal medical data generation. By enabling the generation of multi-view chest X-rays and their associated clinical report, it bridges the gap between general-purpose vision-language models and the specialized requirements of healthcare. Leveraging the MIMIC-CXR dataset, the proposed framework shows superior performance in generating high-fidelity images and semantically coherent reports. Our quantitative evaluation reveals significant results in terms of FID and BLEU scores, showcasing the quality of the generated data. Notably, our framework achieves comparable or even superior performance compared to real data on downstream disease classification tasks, underlining its potential as a tool for medical research and diagnostics. This study highlights the importance of domain-specific adaptations in enhancing the relevance and utility of generative models for clinical applications, paving the way for future advancements in synthetic multimodal medical data generation.

The integration of artificial intelligence in healthcare has opened new horizons for improving medical diagnostics and patient care. However, challenges persist in developing systems capable of generating accurate and contextually relevant radiology reports, particularly in low-resource languages. In this study, we present a comprehensive benchmark to evaluate the performance of instruction-tuned Vision-Language Models (VLMs) in the specialized task of radiology report generation across three low-resource languages: Italian, German, and Spanish. Employing the LLaVA architectural framework, we conducted a systematic evaluation of pre-trained models utilizing general datasets, domain-specific datasets, and low-resource language-specific datasets. In light of the unavailability of models that possess prior knowledge of both the medical domain and low-resource languages, we analyzed various adaptations to determine the most effective approach for these contexts. The results revealed that language-specific models substantially outperformed both general and domain-specific models in generating radiology reports, emphasizing the critical role of linguistic adaptation. Additionally, models fine-tuned with medical terminology exhibited enhanced performance across all languages compared to models with generic knowledge, highlighting the importance of domain-specific training. We also explored the influence of the temperature parameter on the coherence of report generation, providing insights for optimal model settings. Our findings highlight the importance of tailored language and domain-specific training for improving the quality and accuracy of radiological reports in multilingual settings. This research not only advances our understanding of VLMs adaptability in healthcare but also points to significant avenues for future investigations into model tuning and language-specific adaptations.

Non-small cell lung cancer (NSCLC) remains a major global health challenge, with high post-surgical recurrence rates underscoring the need for accurate pathological response predictions to guide personalized treatments. Although artificial intelligence models show promise in this domain, their clinical adoption is limited by the lack of medically grounded guidance during training, often resulting in non-explainable intrinsic predictions. To address this, we propose Doctor-in-the-Loop , a novel framework that integrates expert-driven domain knowledge with explainable artificial intelligence techniques, directing the model toward clinically relevant anatomical regions and improving both interpretability and trustworthiness. Our approach employs a gradual multi-view strategy, progressively refining the model’s focus from broad contextual features to finer, lesion-specific details. By incorporating domain insights at every stage, we enhance predictive accuracy while ensuring that the model’s decision-making process aligns more closely with clinical reasoning. Evaluated on a dataset of NSCLC patients, Doctor-in-the-Loop delivers promising predictive performance and provides transparent, justifiable outputs, representing a significant step toward clinically explainable artificial intelligence in oncology.

Deep learning has revolutionized biomedical research by providing sophisticated methods to handle complex, high-dimensional data. Multimodal deep learning (MDL) further enhances this capability by integrating diverse data types such as imaging, textual data, and genetic information, leading to more robust and accurate predictive models. In MDL, differently from early and late fusion methods, intermediate fusion stands out for its ability to effectively combine modality-specific features during the learning process. This systematic review comprehensively analyzes and formalizes current intermediate fusion methods in biomedical applications, highlighting their effectiveness in improving predictive performance and capturing complex inter-modal relationships. We investigate the techniques employed, the challenges faced, and potential future directions for advancing intermediate fusion methods. Additionally, we introduce a novel structured notation that standardizes intermediate fusion architectures, enhancing understanding and facilitating implementation across various domains. Our findings provide actionable insights and practical guidelines intended to support researchers, healthcare professionals, and the broader deep learning community in developing more sophisticated and insightful multimodal models. Through this review, we aim to provide a foundational framework for future research and practical applications in the dynamic field of MDL.

Data Augmentation (DA) is a technique to increase the quantity and diversity of the training data, and by that alleviate overfitting and improve generalisation. However, standard DA produces synthetic data for augmentation with limited diversity. Generative Adversarial Networks (GANs) may unlock additional information in a dataset by generating synthetic samples having the appearance of real images. However, these models struggle to simultaneously address three key requirements: fidelity and high-quality samples; diversity and mode coverage; and fast sampling. Indeed, GANs generate high-quality samples rapidly, but have poor mode coverage, limiting their adoption in DA applications. We propose LatentAugment, a DA strategy that overcomes the low diversity of GANs, opening up for use in DA applications. Without external supervision, LatentAugment modifies latent vectors and moves them into latent space regions to maximise the synthetic images' diversity and fidelity. It is also agnostic to the dataset and the downstream task. A wide set of experiments shows that LatentAugment improves the generalisation of a deep model translating from MRI-to-CT beating both standard DA as well GAN-based sampling. We further demonstrate its effectiveness when translating from low-energy mammograms to dual-energy subtracted images in contrast-enhanced spectral mammography. Moreover, still in comparison with GAN-based sampling, LatentAugment synthetic samples show superior mode coverage and diversity.

Long-tailed data are a special type of multiclass imbalanced data with a very large amount of minority/tail classes that have a very significant combined influence. Long-tailed learning (LTL) aims to build high-performance models on datasets with long-tailed distributions that can identify all the classes with high accuracy, in particular the minority/tail classes. It is a cutting-edge research direction that has attracted a remarkable amount of research effort in the past few years. In this article, we present a comprehensive survey of the latest advances in long-tailed visual learning. We first propose a new taxonomy for LTL, which consists of eight different dimensions, including data balancing, neural architecture, feature enrichment, logits adjustment, loss function, bells and whistles, network optimization, and posthoc processing techniques. Based on our proposed taxonomy, we present a systematic review of LTL methods, discussing their commonalities and alignable differences. We also analyze the differences between imbalance learning and LTL. Finally, we discuss prospects and future directions in this field.

Accurate classification of non-small cell lung cancer (NSCLC) subtypes is crucial for implementing effective, personalized treatment strategies. This study introduces a novel 3D multimodal convolutional neural network (CNN) architecture for histological subtype classification of NSCLC, specifically distinguishing between lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC). In the context of multimodal deep learning, our approach employs an intermediate fusion technique to integrate both PET and CT imaging modalities, leveraging the complementary information provided by each. We utilized two public datasets alongside a private one, encompassing a total of 714 patients. Our multimodal method was compared against unimodal methods using either CT or PET images alone, achieving better performance. Interestingly, we found that integrating information from multiple imaging modalities can lead to more accurate and reliable NSCLC subtype classification also in case of skewed a priori sample distributions. This non-invasive method has the potential to enhance diagnostic accuracy, improve treatment decisions, and contribute to more personalized and effective lung cancer care strategies. The source code for the implementation described in this paper is available at https://github.com/aksufatih/multimodal-histology-classification

Alzheimer's Disease is the most common cause of dementia, whose progression spans in different stages, from very mild cognitive impairment to mild and severe conditions. In clinical trials, Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) are mostly used for the early diagnosis of neurodegenerative disorders since they provide volumetric and metabolic function information of the brain, respectively. In recent years, Deep Learning (DL) has been employed in medical imaging with promising results. Moreover, the use of the deep neural networks, especially Convolutional Neural Networks (CNNs), has also enabled the development of DL-based solutions in domains characterized by the need of leveraging information coming from multiple data sources, raising the Multimodal Deep Learning (MDL). In this paper, we conduct a systematic analysis of MDL approaches for dementia severity assessment exploiting MRI and PET scans. We propose a Multi Input-Multi Output 3D CNN whose training iterations change according to the characteristic of the input as it is able to handle incomplete acquisitions, in which one image modality is missed. Experiments performed on OASIS-3 dataset show the satisfactory results of the implemented network, which outperforms approaches exploiting both single image modality and different MDL fusion techniques.

We are witnessing a widespread adoption of artificial intelligence in healthcare. However, most of the advancements in deep learning in this area consider only unimodal data, neglecting other modalities. Their multimodal interpretation necessary for supporting diagnosis, prognosis and treatment decisions. In this work we present a deep architecture, which jointly learns modality reconstructions and sample classifications using tabular and imaging data. The explanation of the decision taken is computed by applying a latent shift that, simulates a counterfactual prediction revealing the features of each modality that contribute the most to the decision and a quantitative score indicating the modality importance. We validate our approach in the context of COVID-19 pandemic using the AIforCOVID dataset, which contains multimodal data for the early identification of patients at risk of severe outcome. The results show that the proposed method provides meaningful explanations without degrading the classification performance.

In our fast-paced world, timely access to information is essential. This urgency is highlighted in stress detection, where swift actions can mitigate harmful psycho-physiological effects. We introduce an early stress detection method using Deep Reinforcement Learning (DRL). This method utilizes DRL to efficiently analyze time series data segments, aiming for accurate and quick stress classification. We employ a dynamic observation window strategy, allowing the DRL agent to adjust based on data complexity. Our evaluations, performed on a public dataset using a Leave-One-Subject-Out (LOSO) method, emphasize DRL’s potential in stress detection. The related code is available at https://github.com/cosbidev/DRL-4-Early-Stress-Detection.

The widespread adoption of electronic health records (EHRs) offers a valuable opportunity to support clinical research by containing crucial patient information, including diagnoses, symptoms, medications, lab tests, and more. Despite the success of deep learning for biomedical Named Entity Recognition (NER), the literature in this field still presents a gap regarding applications focused on lung cancer for the Italian language. Hence, this paper presents a transformer-based approach to extract named entities from Italian clinical notes related to Non-Small Cell Lung Cancer (NSCLC). We introduce a novel set of 25 clinical entities related to NSCLC building a corpus annotated for NER. We apply a state-of the-art model pre-trained on Italian biomedical texts to the manually annotated clinical reports of a cohort of 257 patients suffering from NSCLC, successfully dealing with class-imbalance problems and obtaining promising performance (average F1-score of 84.3%). We also compared our method with two other pre-trained state-of-the-art models showing that the domain specific knowledge offered by the proposed approach is necessary to achieve higher performance. These findings also showcase the feasibility of using transformers to extract biomedical information in the Italian language.

Lung cancer has the highest mortality rate among tumours and an accurate pathological assessment is crucial to deliver personalized treatments to patients. The gold standard for pathological assessment requires invasive procedures, which are not always possible and might cause clinical complications. Therefore, in the last years, efforts have been directed towards the development of machine and deep learning approaches for virtual biopsy, which leverage routinely collected CT scans. However, in many cases, the available datasets are limited in size, an issue that limits the training of any model. In this paper, we investigate if triplet networks can cope with this limitation: they are a class of neural networks that uses the same weights while working in tandem on three different input vectors to minimize the loss function. In particular, on a dataset including 87 CT scans collected from patients suffering from non-small cell lung cancer, we experimentally compare triplet networks against plain deep networks when performing histological subtype classification. The results show that the former outperforms the latter in almost all experiments.

The current deep learning approaches for low-dose CT denoising can be divided into paired and unpaired methods. The former involves the use of well-paired datasets, whilst the latter relaxes this constraint. The large availability of unpaired datasets has raised the interest in deepening unpaired denoising strategies that, in turn, need for robust evaluation techniques going beyond the qualitative evaluation. To this end, we can use quantitative image quality assessment scores that we divided into two categories, i.e., paired and unpaired measures. However, the interpretation of unpaired metrics is not straightforward, also because the consistency with paired metrics has not been fully investigated. To cope with this limitation, in this work we consider 15 paired and unpaired scores, which we applied to assess the performance of low-dose CT denoising. We perform an in-depth statistical analysis that not only studies the correlation between paired and unpaired metrics but also within each category. This brings out useful guidelines that can help researchers and practitioners select the right measure for their applications.

Nowadays diabetes still remains one of the leading causes of death worldwide and it has serious consequences if not properly treated. The advent of hybrid closed-loop systems, connection with consumer electronics and cloud-based data systems have hastened the advancement of diabetes technology. In the wake of this progress, we exploit information technology to make insulin pens smart so as to promote adherence to injection therapy and improve the socio-economic impact for the patient. In this respect, this work focuses on two main open issues, namely injection site rotation and lipodystrophies detection while the patient is taking the insulin. The first one is addressed collecting data with IMU sensor which are processed by a machine learning classifier to detect the injection site. The second one is tackled through a sensor equipped with two leds: features computed from such signals fed a one-class Support Vector Machine trained to recognise healthy tissue, so that samples different from those in the training set can be considered as lipodystrophies. The results obtained for the injection site recognition show an average accuracy larger than 0.957, whilst in the case of lipodystrophies detection we reach an accuracy greater than 0.95 using the IR led.

The involvement of axillary lymph node metastasis in breast cancer is one of the most important independent prognostic factors. While the metastasis of lymph node depends on primary tumour intrinsic behaviour, morphology and angioinvasivity, the involvement of the peritumoral tissue by the neoplastic cells also provides useful information for the potential tumour aggressiveness. The lymph node status is currently evaluated by histological invasive procedures with possible complications, asking for introducing safer approaches. Among different imaging techniques, the Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) highlights physiological and morphological characteristics, reflecting breast lesions behaviour and aggressiveness. In the recent years, deep learning (DL) approaches, such as Convolutional Neural Networks, gained increasing popularity for biomedical image processing. Thanks to their ability to autonomously learn from images the set of features for the specific task to solve, they allow finding non-invasive alternatives to the standard procedures used up to now. This paper aims to evaluate the applicability of DL approaches for the axillary lymph node metastasis prediction, considering primary tumour DCE-MRI sequence. Differently from other work in the literature, we include a detailed analysis of healthy tissue influence in lymph node tumour spread through the evaluation of different tumour bounding options. Promising results are reported on a dataset of 153 patients with 155 malignant lesions.

The year 2020 was characterized by the COVID-19 pandemic that has caused, by the end of March 2021, more than 2.5 million deaths worldwide. Since the beginning, besides the laboratory test, used as the gold standard, many applications have been applying deep learning algorithms to chest X-ray images to recognize COVID-19 infected patients. In this context, we found out that convolutional neural networks perform well on a single dataset but struggle to generalize to other data sources. To overcome this limitation, we propose a late fusion approach where we combine the outputs of several state-of-the-art CNNs, introducing a novel method that allows us to construct an optimum ensemble determining which and how many base learners should be aggregated. This choice is driven by a two-objective function that maximizes, on a validation set, the accuracy and the diversity of the ensemble itself. A wide set of experiments on several publicly available datasets, accounting for more than 92,000 images, shows that the proposed approach provides average recognition rates up to 93.54% when tested on external datasets.

The year 2020 was marked by the worldwide COVID-19 pandemic, which caused over 2.5 million deaths by the end of February 2021. Different methods have been established since the beginning to identify infected patients and restrict the spread of the virus. In addition to laboratory analysis, used as the gold standard, several applications have been developed to apply deep learning algorithms to chest X-ray (CXR) images to diagnose patients affected by COVID-19. The literature shows that convolutional neural networks (CNNs) perform well on a single image dataset, but fail to generalize to other sources of data. To overcome this limitation, we present a late fusion approach in which multiple CNNs collaborate to diagnose the CXR scan of a patient, improving the generalizability. Experiments on three datasets publicly available show that the ensemble of CNNs outperforms stand-alone networks, achieving promising performance not only in cross-validation, but also when external validation is used, with an average accuracy of 95.18%.

Recent years have witnessed the rise of pathomics as a mean to describe histopathological images with quantitative biomarkers for predictive and prognostic ends, combining digital pathology, omic science and artificial intelligence. This novel research branch is the counterpart of radiomics which pursues the same aims extracting knowledge from radiological images. In this paper, we present the design of a pathomic deep learning-based system to predict the treatment outcome in non-small cell lung cancer patients. We describe the system design and optimization under the condition of limited data and limited training, with corresponding tests. The experimental results show the feasibility of the proposed scalable architecture providing also a comparison between different transfer learning strategies.

The breakthrough of social media has boosted to an increase in the spread of misleading information, with a serious impact on society especially when related to health knowledge. Recently, researchers have been developing AI-based automatic systems to detect rumours in social microblogs. Nevertheless rumours detection at the level of single post, also referred to as micro-level, is still a major challenge since most of the efforts have been directed toward the macro-level, which means that the system considers as rumours news carried by a set of aggregated microblog posts. In this work, we provide two contributions: first, we compare two state-of-the-art representations to figure out which one better catches hidden information in the data. Second, we explore whether it is possible to exploit knowledge extracted on a topic to automatically recognise micro-level rumour in a different one. To this end, we experimentally investigate three transfer learning methods on two health-related datasets. The comparison with a baseline that does not use any knowledge transfer from the source and target domains reveals that negative transfer occurs.

BACKGROUND: Biological phenomena usually evolves over time and recent advances in high-throughput microscopy have made possible to collect multiple 3D images over time, generating [Formula: see text] (or 4D) datasets. To extract useful information there is the need to extract spatial and temporal data on the particles that are in the images, but particle tracking and feature extraction need some kind of assistance. RESULTS: This manuscript introduces our new freely downloadable toolbox, the Visual4DTracker. It is a MATLAB package implementing several useful functionalities to navigate, analyse and proof-read the track of each particle detected in any [Formula: see text] stack. Furthermore, it allows users to proof-read and to evaluate the traces with respect to a given gold standard. The Visual4DTracker toolbox permits the users to visualize and save all the generated results through a user-friendly graphical user interface. This tool has been successfully used in three applicative examples. The first processes synthetic data to show all the software functionalities. The second shows how to process a 4D image stack showing the time-lapse growth of Drosophila cells in an embryo. The third example presents the quantitative analysis of insulin granules in living beta-cells, showing that such particles have two main dynamics that coexist inside the cells. CONCLUSIONS: Visual4DTracker is a software package for MATLAB to visualize, handle and manually track [Formula: see text] stacks of microscopy images containing objects such cells, granules, etc.. With its unique set of functions, it remarkably permits the user to analyze and proof-read 4D data in a friendly 3D fashion. The tool is freely available at https://drive.google.com/drive/folders/19AEn0TqP-2B8Z10kOavEAopTUxsKUV73?usp=sharing.

Recent epidemiological data report that worldwide more than 53 million people have been infected by SARS-CoV-2, resulting in 1.3 million deaths. The disease has been spreading very rapidly and few months after the identification of the first infected, shortage of hospital resources quickly became a problem. In this work we investigate whether artificial intelligence working with chest X-ray (CXR) scans and clinical data can be used as a possible tool for the early identification of patients at risk of severe outcome, like intensive care or death. Indeed, further to induce lower radiation dose than computed tomography (CT), CXR is a simpler and faster radiological technique, being also more widespread. In this respect, we present three approaches that use features extracted from CXR images, either handcrafted or automatically learnt by convolutional neuronal networks, which are then integrated with the clinical data. As a further contribution, this work introduces a repository that collects data from 820 patients enrolled in six Italian hospitals in spring 2020 during the first COVID-19 emergency. The dataset includes CXR images, several clinical attributes and clinical outcomes. Exhaustive evaluation shows promising performance both in 10-fold and leave-one-centre-out cross-validation, suggesting that clinical data and images have the potential to provide useful information for the management of patients and hospital resources.

The rising capabilities of storing and registering data has increased the number of temporal datasets, boosting the attention on time series classification and forecasting. In case of multivariate time series, symbolic methods that try to predict phenomena transform the data into a more compact format to produce a representation of the time series easy to be handled in a machine learning framework. However, up to now these representations do not grasp information on both inter-attribute variability and temporal variability. In this work we present an approach that, taking into account the relationships between attributes and their periodicity, reduces the multivariate time series to a collection of symbols, whose distribution is represented by histograms. The approach has been successfully tested on a publicly available dataset, the Telecom Italia Big Data Challenge 2014 dataset, reporting also the results attained by other methods available in the literature.

Breast cancer is the most common tumour in women and it is characterized by a huge variety of clinical and histological scenarios and imaging pattern. The axillary lymph node metastases presence or absence is one of the most important prognostic factors affecting the loco-regional recurrence and the overall survival. The lymph node status is usually determined by an histological exam, an invasive procedure that could result in complications. This work aims to provide a safer and non-invasive prognostic approach by introducing a radiomics-based method that predicts axillary lymph node metastasis. It combines primary tumor histological features and patients clinical data with quantitative measures extracted from the MR images. To compute these latter quantities we determine the convex hull of the ROIs and we introduce the Three Orthogonal Planes-Local Binary Pattern (TOP-LBP). On 99 samples the approach achieves a promising AUC equal to 85.6%.

The COVID-19 disease caused by the coronavirus SARS-nCoV2 is currently a global public health threat and Italy is one of the countries mostly suffering from this epidemic. It is therefore important to analyze epidemic data, considering also that the government deployed laws limiting the societal activities. We model COVID-19 dynamics with a SIQR (susceptible - infectious - quarantined - recovered) model, where we take into account the temporal variability of its parameters. Particle Swarm Optimization is used to find out the best parameters in the case of Italy and of Italian regions where the epidemic has the greatest impact. The basic reproductive number is estimated by a novel approach that averages out different PSO fits computed considering different temporal time-windows and reducing possible noise in the data. The results on data collected from February 24 to April 24 show that our approach is able to fit the data with low errors and that the basic reproductive number is characterized by a descending trend in time from 3.5 to a value below 1.

In many real world medical datasets learning under class imbalance can limit the performance of most supervised algorithms, resulting in a low recognition rate for samples belonging to the minority class. Within the wide literature on this issue the class of techniques that goes under the name of ensemble methods have been less investigated, despite their effectiveness. In this panorama, we present here an original strategy to construct the training set of each base classifier in the ensemble: it exploits information in the feature space that can give rise to unreliable classifications, which are determined by a novel algorithm here introduced. Our proposal is compared against multiple standard ensemble approaches on 18 publicly available biological datasets, showing promising results.

Class imbalance limits the performance of most learning algorithms, resulting in a low recognition rate for samples belonging to the minority class. Although there are different strategies to address this problem, methods that generate ensemble of classifiers have proven to be effective in several applications. This paper presents a new strategy to construct the training set of each classifier in the ensemble by exploiting information in the feature space that can give rise to unreliable classifications, which are determined by a novel algorithm here introduced. The performance of our proposal is compared against multiple standard ensemble approaches on 25 publicly available datasets, showing promising results.

Prostate cancer is the most common form of cancer in Western countries and there is the need to develop clinical decision support systems able to support physicians in the diagnosis of clinical relevant prostate cancer and avoid useless invasive prostate biopsies. In this respect, this paper introduces a radiomic approach that classifies the prostate cancer aggressiveness by combining Three Orthogonal Planes-Local Binary Pattern (TOP - LBP) with other texture measures. Furthermore, to combat the skewed nature of class priors, our proposal employs a data augmentation technique. The results achieved on 99 samples are up-and-coming, they favorably compare against conventional PI-RADS-based approach, and they show also the benefit given by the introduction of TOP-LBP in the radiomic signature.

The hospital is a center of healthcare services that, nowadays, can be considered as a highly technological corporation. In this work we introduce the Hospital 4.0 frame-work, discussing the innovation it would bring both at the methodological and technological level. Transforming the hospital organization from a multi-functional center where the patient is treated by different therapeutic units to an integrated center able to provide the patient with a personal care service, involving the patient as an active subject, represents the innovative challenge for the forthcoming years.

The possibility of planning a therapy minimizing side effects and optimizing efficacy of cancer treatments is one of the main challenges tackled by precision medicine research in oncology. In this context, radiomics is revealing itself as a promising path for better understanding the correct approach to personalized cures. Its primary aim is to go beyond basic medical images analysis, which only leverages on direct measurements on the tumor mass, i.e. dimension and shape. On the contrary, radiomics approach is oriented to the extraction of heterogeneous and quantitative data from the images to characterize the disease from a wider perspective, in order to provide the physician a valid support for the therapy decision and survival prediction. This manuscript presents an application of radiomics to Non-Small Cell Lung Cancer, dealing with the novel task of predicting the possibility to carry out an adaptive therapy. We achieved promising performance, reporting a radiomics signature for predicting tumor reduction during therapy.

In the last years social networks have emerged as a critical mean for information spreading. In spite of all the positive consequences this phenomenon brings, unverified and instrumentally relevant information statements in circulation, named as rumours, are becoming a potential threat to the society. Recently, there have been several studies on topic-independent rumour detection on Twitter. In this paper we present a novel rumour detection system which focuses on a specific topic, that is health-related rumours on Twitter. To this aim, we constructed a new subset of features including influence potential and network characteristics features. We tested our approach on a real dataset observing promising results, as it is able to correctly detect about 89% of rumours, with acceptable levels of precision.

Class imbalance exists in many applications of bioinformatics and biomedicine, while dimension reduction in the feature space is often needed when building prediction models on a dataset. When the above two issues need to be considered simultaneously for skewed/imbalanced datasets, practitioners and researchers in machine learning may raise the following question: should feature selection be conducted before or after the resampling methods for combating the skewness of a dataset? While feature selection and class imbalance learning have been widely studied in the literature, little study has jointly investigated them. This paper presents a first empirical study on the performance of the two opposing pipelines for binary imbalance learning, i.e., first feature selection then resampling, or first resampling then feature selection. We carry out the study on 35 publicly available datasets belonging to the biological field, using 9 feature selection methods, 6 resampling approaches for class imbalance learning, and 3 well-known classifiers. Our experiments reveal that, there is no constant winner between the two pipelines, practitioners should test both pipelines in order to derive the best classification model for imbalance learning, in particular, the resampling before feature selection pipeline should not be neglected; but we also show that, the feature selection before resampling pipeline outperforms the other in more cases than not.

Chronic Obstructive Pulmonary Disease (COPD) is a preventable, treatable, and slowly progressive disease, whose course is aggravated by a periodic worsening of symptoms and lung function lasting for several days. The development of home telemonitoring systems has made possible to collect symptoms and physiological data in electronic records, boosting the development of decision support systems (DSSs). Current DSSs work with physiological measurements collected by means of several measuring and communication devices as well as with symptoms gathered by questionnaires submitted to COPD subjects. However, this contrasts with the advices provided by the World Health Organization and the Global initiative for chronic Obstructive Lung Disease that recommend to avoid invasive or complex daily measurements. For these reasons this manuscript presents a DSS detecting the onset of worrisome events in COPD subjects. It uses the hearth rate and the oxygen saturation, which can be collected via a pulse oximeter. The DSS consists in a binary finite state machine, whose training stage allows a subject specific personalization of the predictive model, triggering warnings, and alarms as the health status evolves over time. The experiments on data collected from 22 COPD patients tele-monitored at home for six months show that the system recognition performance is better than the one achieved by medical experts. Furthermore, the support offered by the system in the decision-making process allows to increase the agreement between the specialists, largely impacting the recognition of the worrisome events.

Neuroscience has been interested in cellular neuroanatomy since the time of Ramón y Cajal. Although the manual reconstruction of neuron morphologies is still widely used, the recent availability of large image data asks for automatic or semi-automatic tools. In this paper we present an automatic method to trace neurites in 3D volumes based on two main steps. The first detects neurites connected with a given seed that satisfy a conservative membership rule, the second detects weak neurite chunks allowing a local growth of the arbor on the basis of local intensity features. The local step also accommodates to the nonuniform illumination and to the noise of the sample. We tested our proposal on the Olfactory Projection dataset belonging to the well-known DIADEM challenge, comparing its performance against those achieved by other state-of-the-art methods available within the BigNeuron Project.

Investigation of membrane fluidity by two photon fluorescence microscopy opens up a new and important area of translational research, being a useful and sensitive method for disease monitoring and treatment. In this paper we investigate if biomembranes in human red blood cells (RBC) and peripheral mononuclear cells (PMC) could be used as markers for type 1 diabetes mellitus (T1DM) diagnosis, leading to the development of a method for monitoring T1DM progression that nowadays is lacking, as clinical exams cannot pursue this task with enough reliability. To this aim, we present a set of features computed from PMC and RBC images that are given to a multi-experts system leveraging on multi-spectral information for positive/negative classifications. The experiments are carried out on a dataset of 800 blood cell images belonging to 18 subjects adopting the leave-one-person-out approach.

The development of computer-aided diagnosis (CAD) systems for antinuclear autoantibodies tests in indirect immunofluorescence using HEp-2 cells has attracted growing research efforts in the last years. Although in this field many CAD solutions extract information from objects detected within the images, cell segmentation is an issue far from being solved. This work introduces a segmentation pipeline based on an active contour method, as none in the literature on HEp-2 cell segmentation does. Our proposal can detect objects whose boundaries are not necessarily defined by gradient, and this choice plays a relevant role for HEp-2 images with different fluorescence intensities and different staining patterns. The performances of the approach is tested not only on a public benchmark dataset with 18 images, but also on other 24 images that we made publicly available. Furthermore, its performances are compared with those provided by other state-of-the-art segmentation methods.

Chronic obstructive pulmonary disease (COPD) is a slowly progressive disease characterized by airway obstruction. Patients suffering from COPD could report exacerbations, which are deteriorations in respiratory health that worsen the course of the disease. The prompt recognition of an exacerbation from daily variations and its early treatment reduces the healing time and the risk of hospitalization. Using a pulse oximeter connected to a mobile phone, we remotely collect the values of heart rate and of the oxygen saturation on a cohort of seven elderly patients affected by severe COPD. We analyze if a score given by the weighted composition of these signals permits to detect the worrisome events prefiguring an exacerbation onset. A cross validation-based evaluation allows us to assess how the results generalize to an independent data set. We found that the tested score does not provide satisfactory results in terms of sensitivity and specificity, suggesting that it is not able to disambiguate between exacerbation onset and other COPD related events.

Computer-based medical systems play a very important role in medical applications because they can strongly support the physicians in the decision making process. The large amount of data nowadays available, although collected from high quality sources, usually contain irrelevant, redundant, or noisy information, suggesting that not all the training instances are useful for the classification task. To address this issue, we present here an instance selection method that, different from the existing approaches, selects in ``real-time" a subset of instances from the original training set on the basis of the information derived from each test instance to be classified. We apply our method to seven public benchmark datasets, achieving larger performances than a baseline classifier.

MOTIVATION: Recently, confocal light sheet microscopy has enabled high-throughput acquisition of whole mouse brain 3D images at the micron scale resolution. This poses the unprecedented challenge of creating accurate digital maps of the whole set of cells in a brain. RESULTS: We introduce a fast and scalable algorithm for fully automated cell identification. We obtained the whole digital map of Purkinje cells in mouse cerebellum consisting of a set of 3D cell center coordinates. The method is accurate and we estimated an F1 measure of 0.96 using 56 representative volumes, totaling 1.09 GVoxel and containing 4138 manually annotated soma centers. AVAILABILITY AND IMPLEMENTATION: Source code and its documentation are available at http://bcfind.dinfo.unifi.it/. The whole pipeline of methods is implemented in Python and makes use of Pylearn2 and modified parts of Scikit-learn. Brain images are available on request. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Human action recognition is an active research area with applications in several domains such as visual surveillance, video retrieval and human–computer interaction. Current approaches assign action labels to video streams considering the whole video as a single sequence but, in some cases, the large variability between frames may lead to misclassifications. The authors propose a multiple subsequence combination (MSC) method that divides the video into several consecutive subsequences. It applies part‐based and bag of visual words approaches to classify each subsequence. Then, it combines subsequence labels to assign an action label to the video. The proposed approach was tested on the KTH, UCF sports, Youtube and Robo‐Kitchen datasets, which have large differences in terms of video length, object appearance and pose, object scale, viewpoint, background, as well as number, type and complexity of actions performed. Two main results were achieved. First, the MSC approach shows better performances compared to classify the video as a whole, even when few subsequences are used. Second, the approach is robust and stable since, for each dataset, its performances are comparable to the part‐based approach at the state‐of‐the‐art.

Several medical and biological applications face with multiclass recognition problems. Such polychotomies can be addressed by decomposition techniques, which reduce the polychotomy into a series of dichotomies and then provide the final multiclass label using a reconstruction rule. Within this framework, we present a reconstruction rule based on softmax regression, where the features of the new classification task are the crisp labels and the reliabilities of dichotomizers' classifications. The approach has been tested on six medical and biological datasets, decomposing the polychotomies via the Error-Correcting Output Code. Its performances favorably compare with those provided by other two well-known reconstruction rules both in terms of global accuracy and accuracy per class.

In this paper, we report on the first edition of the HEp-2 Cells Classification contest, held at the 2012 edition of the International Conference on Pattern Recognition, and focused on indirect immunofluorescence (IIF) image analysis. The IIF methodology is used to detect autoimmune diseases by searching for antibodies in the patient serum but, unfortunately, it is still a subjective method that depends too heavily on the experience and expertise of the physician. This has been the motivation behind the recent initial developments of computer aided diagnosis systems in this field. The contest aimed to bring together researchers interested in the performance evaluation of algorithms for IIF image analysis: 28 different recognition systems able to automatically recognize the staining pattern of cells within IIF images were tested on the same undisclosed dataset. In particular, the dataset takes into account the six staining patterns that occur most frequently in the daily diagnostic practice: centromere, nucleolar, homogeneous, fine speckled, coarse speckled, and cytoplasmic. In the paper, we briefly describe all the submitted methods, analyze the obtained results, and discuss the design choices conditioning the performance of each method.

Antinuclear autoantibodies (ANAs) are important markers to diagnose autoimmune diseases, very serious and also invalidating illnesses. The benchmark procedure for ANAs diagnosis is the indirect immunofluorescence (IIF) assay performed on the HEp-2 substrate. Medical doctors first determine the fluorescence intensity exhibited by HEp-2 cells, and then report the staining pattern for positive wells only. With reference to staining pattern recognition, in the literature we found works recognizing five main patterns characterized by well-defined cell edges. These approaches are based on cell segmentation, a task that should be harder than the classification itself. We present here a method extending the panel of detectable HEp-2 staining patterns, introducing the centromere and cytoplasmic patterns, which do not show well-defined cell edges, and where a segmentation-based classification may fail. We apply a local approach which extracts SIFT descriptors and then classifies an image through the bag of visual words approach. This permits to represent complex image contents without applying the segmentation procedure. We test our approach on a dataset of HEp-2 images with large variability in both fluorescence intensity and staining patterns. Despite the large skew of the a-priori class distribution, our system correctly recognizes the 98.3% of samples, with a F-measure equal to 92.3%, 95.2% and 99.0%, for each class.

In this paper we propose a new method for cells segmentation in HEp-2 images addressing and overcoming the main limitations of the existing approaches. The proposed method adopts image reconstruction for a preliminary image segmentation and, then, it employs a sort of classifier-controlled dilation for better determining the structure of the cells, where the classifier is trained using data of the image at hand. We compare the performance of the proposed method with the most representative approaches from the scientific literature on a common and publicly available dataset of HEp-2 images.

Indirect immunofluorescence (IIF) is the recommended technique to detect rheumatic diseases through the analysis of images exhibiting a fluorescence in the green band. Since the request of such tests has recently increased, researches efforts have been directed towards the development of computer-aided diagnosis (CAD) tools supporting the specialists and improving the standardization of the method. Technological advances have made available color cameras for IIF image acquisition, but their use requires to determine which color to greyscale conversion method provides most useful information for the needs of CAD development. In this respect, we experimentally compare four different methods converting a color image into a greyscale one, analyzing wide features sets for each conversion method and applying four classification paradigms. Experiments have been carried out on an annotated dataset of HEp-2 cells, finding out a subset of features which is independent from the color model used and showing that a greyscale representation based on the HSI model better exploits information for IIF images analysis.

Photobleaching effect is a major issue in developing autofocus algorithms for indirect immunofluorescence (IIF) assay since it quickly flattens the fluorescence intensity of the samples. To overcome this limitation we propose an autofocus algorithm, suited for IIF, aiming at minimizing the number of images required to focus the sample. We test our algorithm with an heterogeneous set of IIF images containing the substrates most widely used in clinical practice, comparing popular focus functions in order to find out the most appropriate for our procedure. Experimental results prove that our algorithm focus the sample using less images than other popular algorithms proposed in the literature.

Detection of antibodies via indirect immunofluorescence (IIF) is a common marker in patients with suspected connective tissue diseases. IIF readings are affected by several issues limiting their reproducibility and reliability: hence, the development of a computer-aided diagnosis (CAD) tool supporting IIF diagnostic procedure would be beneficial in many respects. Although some works in the literature use greyscale cooled cameras for IIF image acquisition, recent research as well as commercial CAD solution use colour cameras. Indeed, colour cameras are cheaper than greyscale cameras and have adequate performance for the needs of IIF image acquisition. However, their application asks for studying how to extract useful information from colour images for CAD development. This paper presents an experimental comparison between four different methods converting a colour image into a greyscale one. This analysis has been carried out testing different popular classification paradigms on an annotated IIF image dataset, and also performing pre-clinical tests. Results show that a conversion method based on information derived from RGB primary components outperforms the others relying on different colour models.

Indirect immunofluorescence (IIF) imaging is the recommended laboratory technique to detect autoantibodies in patient serum, but it suffers from several issues limiting its reliability and reproducibility. IIF slides are observed by specialists at the fluorescence microscope, reporting fluorescence intensity and staining pattern and looking for mitotic cells. Indeed, the presence of such cells is a key factor to assess the correctness of slide preparation process and the reported staining pattern. Therefore, the ability to detect mitotic cells is needed to develop a complete computer-aided-diagnosis system in IIF, which can support the specialists from image acquisition up to image classification. Although recent research in IIF has been directed to image acquisition, image segmentation, fluorescence intensity classification and staining pattern recognition, no works presented methods suited to classify such cells. Hence, this paper presents an heterogeneous set of features used to describe the peculiarities of mitotic cells and then tests five classifiers, belonging to different classification paradigms. The approach has been evaluated on an annotated dataset of mitotic cells. The measured performances are promising, achieving a classification accuracy of 86.5 %.

TBY-2 cells are widely used in several applications and in particular in the study of programmed cell-death (PCD). However, automatic or semi-automatic computer-based tools supporting the specialist during the measurement process are still missing. In this paper we propose and test a semi-automatic tool for the segmentation of TBY-2 cell and the measurement of cytosol area. The algorithm has been designed in order to be easy to use and to have low complexity. Results obtained on a database of TBY-2 cells confirm that the algorithm can be effectively used by specialists in their daily work.

Decomposition methods are multiclass classification schemes where the polychotomy is reduced into several dichotomies. Each dichotomy is addressed by a classifier trained on a training set derived from the original one on the basis of the decomposition rule adopted. These new training sets may present a disproportion between the classes, harming the global recognition accuracy. Indeed, traditional learning algorithms are biased towards the majority class, resulting in poor predictive accuracy over the minority one. This paper investigates if the application of learning methods specifically tailored for imbalanced training set introduces any performance improvement when used by dichotomizers of decomposition methods. The results on five public datasets show that the application of these learning methods improves the global performance of decomposition schemes.

Learning under imbalanced dataset can be difficult since traditional algorithms are biased towards the majority class, providing low predictive accuracy over the minority one. Among the several methods proposed in the literature to overcome such a limitation, the most recent uses multi-experts system (MES) composed of balanced classifiers, whose decisions are aggregated according to a combination rule. Each classifier of the MES is trained with a balanced subset of the original training set, which can be determined applying different division methods. This paper explores how different MES combination rules perform with imbalanced TS, experimentally comparing fusion and selection combination criteria. Furthermore, two methods have been used to divide the original TS, namely random selection and clustering. The results confirm and extend previous findings reported in the literature showing that, on the one side, combination rules belonging to selection framework outperform the others and, on the other side, dividing the original training set via random selection rather than clustering permits to attain better performance.

Indirect immunofluorescence is currently the recommended method for the detection of antinuclear autoantibodies (ANA). The diagnosis consists of both estimating the fluorescence intensity and reporting the staining pattern for positive wells only. Since resources and adequately trained personnel are not always available for these tasks, an evident medical demand is the development of computer-aided diagnosis (CAD) tools that can support the physician decisions. In this paper, we present a system that classifies the staining pattern of positive wells on the strength of the recognition of their cells. The core of the CAD is a multiple expert system (MES) based on the one-per-class approach devised to label the pattern of single cells. It employs a hybrid approach since each composing binary module is constituted by an ensemble of classifiers combined by a fusion rule. Each expert uses a set of stable and effective features selected from a wide pool of statistical and spectral measurements. In this framework, we present a novel parameter that measures the reliability of the final classification provided by the MES. This feature is used to introduce a reject option that allows to reduce the error rate in the recognition of the staining pattern of the whole well. The approach has been evaluated on 37 wells, for a total of 573 cells. The measured performance shows a low overall error rate ( 2.7%-5.8%), which is below the observed intralaboratory variability.

The performance of a classification system is sometimes unsatisfactory for the needs of real applications. In these cases, the measure of classification reliability should be useful since it takes into account the many issues that influence the achievement of satisfactory results. The most common choice for confidence evaluation consists in using the confusion matrix estimated during the learning phase. As a consequence, the same reliability value is associated with every decision attributing a sample to the same class. In this respect, this paper proposes and compares three different reliability estimators of each classification act of classification systems that belong to the one-per-class framework. They are based on the reliabilities provided by each dichotomizer and are independent of the binary module design. Their performance have been assessed and ranked on private and public medical datasets, showing that one of the estimators outperforms the others.

Applications of robotics and mechatronics to neurorehabilitation are getting more and more consensus in the clinical community thanks to early encouraging results. They enable an objective assessment of patient's motor recovery and the administration of rehabilitation treatments specific for each patient. In particular, isometric force/torque measurements in post-stroke patients were recently used in clinical trials for the functional assessment, with encouraging results. A challenging issue in the processing of such measurements is to detect the initiation of the voluntary contraction of the patient (i.e., onset time). The onset detection is crucial to obtain clinically relevant data. In previous works, different deterministic methods for onset detection were presented. Each of those methods is signal-structure dependant, causing drop of performance when applied to different kind of signals. In this paper, we introduce an innovative technique for the automatic selection of the best onset detection method. To this aim, we adopt a supervised pattern recognition approach that dynamically selects, from a pool of deterministic methods, the one that is best suited for each signal according to the signal structure. The method has been tested on annotated force and torque datasets, showing that such a method improves not only the performance achieved by the single deterministic techniques, but also those attained by a group of clinical experts.

In autoimmune diseases, indirect immunofluorescence (IIF) represents the recommended method for detection of antinuclear autoantibodies (ANA). IIF diagnosis requires both the estimation of fluorescence intensity and the description of staining pattern, demanding for highly specialized personnel, who are not always available. In this respect, computer-aided diagnosis (CAD) tools can support physicians' decision. In this paper we report experiences in the staining pattern recognition of IIF wells. Since several cells constitute each well, we have developed a multiple expert system (MES) devised to classify the pattern of individual cells. The whole well staining pattern is computed on the strength of the recognition of its cells, testing two aggregation rules. The experimental results shows the feasibility of a CAD dedicated to the classification of staining pattern in the IIF field.

Indirect immunofluorescence is the recommended method for antinuclear autoantibodies (ANA) detection. IIF diagnosis requires estimating fluorescent intensity and pattern description, but resources and adequately trained personnel are not always available for these tasks. In this respect, an evident medical demand is the development of computer aided diagnosis tools that can offer a support to physician decision. In this paper we propose a system to classify the fluorescent intensity: initially we discuss two classifiers based on artificial neural networks that can recognize intrinsically dubious samples and whose error tolerance can be flexibly set according to a given rule. Since such classifiers complement one other, we adopt a multiple expert system that aggregates the two experts. The final decision of the system results from the combination of the outputs of the single experts. Measured performance shows error rates less than 1%, which candidates the method to be used in daily medical practice

In this paper, we report our experience in the development of a system for automatic acquisition of Immuno-Fluorescence Assay (IFA) images. We focus on two basic issues. Firstly, we determine an autofocus function that can deal with photobleaching, a physical phenomenon affecting automatic acquisition of IFA images, and present a set of experiments on real images that confirm its effectiveness. Secondly, we discuss if the physicians may reliably use digital IFA images in place of direct microscope observations to carry out the diagnosis. In this respect, we present the results of a preliminary experiment where physicians perform the diagnosis on a set of images both by looking directly to them at the fluorescence microscope and by looking at digital images on the screen of a workstation.