Despite the remarkable performance of deep models in medical imaging, they still require source data for training, which limits their potential in light of privacy concerns. Federated learning (FL), as a decentralized learning framework that trains a shared model with multiple hospitals (a.k.a., FL clients), provides a feasible solution. However, data heterogeneity and resource costs hinder the deployment of FL models, especially when using vision language models (VLM). To address these challenges, we propose a novel contrastive language-image pre-training (CLIP) based FL approach for medical image classification. Specifically, we introduce a masked feature adaptation module (FAM) as a communication module to reduce the communication load while freezing the CLIP encoders to reduce the computational overhead. Furthermore, we propose a masked multi-layer perceptron (MLP) as a private local classifier to adapt to the client tasks. Moreover, we design an adaptive Kullback-Leibler (KL) divergence-based distillation regularization method to enable mutual learning between FAM and MLP. Finally, we incorporate model compression to transmit the FAM parameters while using ensemble predictions for classification. Extensive experiments on four publicly available medical datasets demonstrate that our model provides feasible performance (e.g., 8% higher compared to second best baseline on ISIC2019) with reasonable resource cost (e.g., 120 times faster than FedAVG).

Federated learning (FL) in practical applications frequently encounters the challenge of limited labeled data availability on client devices, where data annotation is often expensive or requires domain expertise. This scarcity, especially under non-independent and non-identically distributed (Non-IID) settings, leads to error accumulation from pseudo-labels and under-utilization of unlabeled data. To overcome this limitation, we propose a federated semi-supervised learning (FSSL) approach that jointly uses labeled and unlabeled data distributed across both server and clients. Building upon existing FSSL frameworks, our method introduces two key innovations: (1) an optimized client consistency regularization mechanism using Kullback-Leibler (KL) divergence for low-confidence samples, and (2) an advanced pseudo-labeling strategy incorporating Entropy Meaning Loss (EML) for high-confidence samples, which are designed to maximize the utility of unlabeled data and enhance model generalization capabilities. Our classification experiments on common (CIFAR-10, CIFAR-100, MNIST, SVHN) and medical datasets (Brain Tumor, Nail Disease) show that the proposed method outperforms the existing baseline models. On CIFAR-10, with only 10% labeled data (IID), the proposed FSSL-UC achieves a 1.33% higher accuracy than FedMatch, while under the more challenging setting of 1% labeled data (Non-IID), the improvement reaches 6.33%. On the brain tumor dataset, with 1% labeled data, FSSL-UC outperforms FedMatch by 14.47% (IID) and 7.61% (Non-IID). In addition, FSSL-UC also showed stable advantages on MNIST, SVHN, and Nail Disease datasets, demonstrating its robustness and generalization ability.

With the rapid development of artificial intelligence (AI), especially in the medical field, the need for its explainability has grown. In medical image analysis, a high degree of transparency and model interpretability can help clinicians better understand and trust the decision-making process of AI models. In this study, we propose a Knowledge Distillation (KD) based approach that aims to enhance the transparency of the AI model in medical image analysis. The initial step is to use traditional CNN to obtain a teacher model and then use KD to simplify the CNN architecture, retain most of the features of the data set, and reduce the number of network layers. It also uses the feature map of the student model to perform hierarchical analysis to identify key features and decision-making processes. This leads to intuitive visual explanations. We selected three public medical data sets (brain tumor, eye disease, and Alzheimer's disease) to test our method. It shows that even when the number of layers is reduced, our model provides a remarkable result in the test set and reduces the time required for the interpretability analysis.

Electroencephalography (EEG) signals are widely used in brain-computer interfaces, clinical diagnostics, and neuroscience topics. However, these signals are often contaminated by physiological and non-physiological artifacts, such as eyerelated artifacts and power line interference. While various denoising techniques have been developed, most focus on removing physiological artifacts, with limited attention given to random noise caused by equipment and ambient conditions. This type of non-physiological noise typically exhibits statistical properties that approximate a normal distribution (Gaussian distribution). In this study, we simulate such artifacts by adding Gaussian noise (amplified by a factor of 50) to clean EEG signals from the EEGdenoiseNet dataset. We evaluated 14 classical and modern denoising methods using signal-to-noise ratio (SNR), mean squared error (MSE), and normalized cross-correlation (NCC). Experimental results show that the regression method (RM), implemented as ridge regression, achieves the lowest MSE and highest NCC, outperforming deep learning-based approaches such as convolutional neural networks (CNN) under these conditions. These findings demonstrate the effectiveness of regression-based techniques in mitigating simulated non-physiological artifacts and provide guidance for future EEG signal processing. The codes are available at https://github.com/AIPMLab/EEG-Denoising.

Electroencephalography (EEG) signals are susceptible to numerous sources of interference; therefore, it is important to employ appropriate EEG denoising techniques. This study investigates common EEG denoising techniques, aiming to identify the optimal strategy for contemporary EEG signal processing. Using the publicly available EEGDenoiseNet dataset, eight techniques are used, namely EEGIFNet (a Dual-Branch Interactive Fusion Network), EEGDnet (a 2D-Transfomer-based architecture), EEGDfus(a Dual-Branch Fusion Network with Transfomer attention), GCTNet(a Generative Adversarial Network based model), EEGDiR(a Retentive Network based model), 2×3R-CNN, CNN, and RNN-LSTM. These techniques are compared on the basis of the root mean square error (RMSE) and signal-to-noise ratio (SNR). Experimental results indicate that the “EEGDfus” model achieved the highest performance metrics with an RMSE of 20.1 and an SNR of 10.3 dB. The “GCTNet” model has the best balance with performance and efficiency. These findings highlight the potential of the interactive Fusion-CNN Network to enhance the precision of denoising, particularly in cases of mixed interference from various artifacts. Our code is available at https://github.com/AIPMLab/EEG-Denoising-Comparation/.

This study introduces the SHAP-integrated convolutional diagnostic network (SICDN), an interpretable feature selection method designed for limited datasets, to address the challenge posed by data privacy regulations that restrict access to medical datasets. The SICDN model was tested on classification tasks using pneumonia and breast cancer datasets, demonstrating over 97% accuracy and surpassing four popular CNN models. We also integrated a historical weighted moving average technique to enhance feature selection. The SICDN shows potential in medical image prediction, with the code available on https://github.com/AIPMLab/SICDN.

Despite the widespread application of neural networks in deep learning for image classification, challenges persist in accurately recognizing images, especially complex structures such as brain tumors. We proposed the Attention-Based Residual Multiscale Convolutional Neural Network (ABRMS-Net), which integrates four attention mechanisms to capture multi-scale features and subtle differences in images. We compared ABRMS-Net with eight popular convolutional neural networks (CNNs) across two datasets. Experimental results indicate that our model outperforms these eight CNNs, demonstrating exceptional performance, including achieving the best classification accuracy of 99.47 ± 0.39% in the brain tumor dataset and exceeding the baseline by over 21.58% in the evaluation metrics in the face expression dataset. Additionally, through heatmap visualization, this paper demonstrates the interpretability of ABRMS-net within neural networks and its effectiveness in using a small number of training samples for the development of few-shot learning. The code will be made publicly accessible at https://github.com/AIPMLab/ABRMS-Net.

Federated learning (FL) in real-world scenarios often faces the challenge of a lack of labeled data on the client side, while obtaining labeled data is costly or requires expertise. In this paper, we address this challenge through federated semi-supervised learning (FSSL), leveraging labeled and unlabeled data distributed on both the server and the client. Based on the existing FSSL framework, we propose an improved method that introduces a modified client consistency mechanism and an enhanced pseudo-label generation strategy to more efficiently use unlabeled data, thereby improving the model generalization. Our classification experiments on common datasets CIFAR-10, MNIST and SVHN show that the proposed method outperforms the existing baseline models. Specifically, on the CIFAR-10 dataset, the proposed method achieves an accuracy improvement from 78.68% to 80.01% under the IID setting and from 46.52% to 52.85% under the non-IID setting. For the MNIST dataset, the accuracy increases from 97.70% to 98.34% in the IID setting and from 95.11% to 95.60% in the non-IID case. On the SVHN dataset, the accuracy rises from 92.29% to 92.54% under IID conditions and improvement from 80.26% to 81.62% in the non-IID setting. These results enhance the generalizability of the FL model. Our code is available at https://github.com/AIPMLab/IJCNN25_FSSL.

Contrastive Language-Image Pretraining (CLIP) has emerged as a powerful paradigm for cross-modal learning, using image-text pairs to achieve remarkable zero-shot classification performance. However, its calibration on noisy data has been less explored, especially in out-of-distribution settings where overconfidence can lead to misclassification. In this work, we evaluate CLIP’s calibration ability under different noise conditions using 10 noise types from the ImageNet-C dataset, including natural noise, digital noise, weather noise, and blur noise. In addition, we propose test-time augmentation (TTA) to improve calibration by increasing prediction diversity and reducing overconfidence. We conduct ~ 600 simulations, and the experimental results show that ViT-B/32 achieves higher accuracy (ACC) and lower expected calibration error (ECE) than ResNet50 in in-distribution settings (e.g., 94.66% ACC vs. 62.37% ACC, 0.59% ECE vs. 1.03% ECE on brightness). However, ResNet50 outperforms ViT-B/32 in out-of-domain (OOD) situations (e.g., 3.14% ECE vs. 20.93% ECE on defocus, fine-tuned on glass). By applying TTA, we reduce ViT-B/32’s ECE to 13.07% on defocus (fine-tuned on glass), demonstrating its effectiveness. Our results highlight the importance of calibration in cross-modal learning and provide a simple yet effective solution for noisy and OOD calibration. Our code is available at: https://github.com/AIPMLab/CLIPSimulationsGao.

Deep learning models, including convolutional neural networks (CNNs) and transformer-based architectures, have achieved remarkable results in medical image segmentation tasks. However, the impact of optimization strategies and training settings on their performance is not explored enough. This study conducted extensive experiments with three CNN models (U-Net, UNet++, Attention-UNet) and four transformer-based models (TransUNet, HiFormer, Swin-UNet, TransDeepLab) using four public medical imaging datasets (DRIVE, Skin Lesion, Lung, Synapse). Three optimizers - SGD, Adam, and AdaGrad were evaluated to assess their influence on the stability and performance of the model. For example, in skin lesion segmentation, Dice scores ranged from 89.34% (Swin-UNet) to 93.40% (TransUNet(Pre)) with SGD, from 71.70% (TransDeepLab) to 89.22% (Attention-UNet) with Adam, and from 84.45% (Swin-UNet(Pre)) to 91.58% (TransUNet(Pre)) with AdaGrad. For the Synapse multi-organ segmentation task, Dice scores varied from 61.39% (Swin-UNet) to 77.58% (UNet++), indicating that CNNs tend to outperform transformer models in multi-organ segmentation. The findings highlight the critical impact of optimizer selection in improving segmentation performance and provide valuable information to advance medical image segmentation methods. Our codes are available at https://github.com/AIPMLab/SegmentationSimulationLu.

Domain adaptation (DA) is a quickly expanding area in machine learning that involves adjusting a model trained in one domain to perform well in another domain. While there have been notable progressions, the fundamental concept of numerous DA methodologies has persisted: aligning the data from various domains into a shared feature space. In this space, knowledge acquired from labeled source data can improve the model training on target data that lacks sufficient labels. In this study, we demonstrate the use of 10 deep learning models to simulate common DA techniques and explore their application in four medical image datasets. We have considered various situations such as multi-modality, noisy data, federated learning (FL), interpretability analysis, and classifier calibration. The experimental results indicate that using DA with ResNet34 in a brain tumor (BT) data set results in an enhancement of 4.7% in model performance. Similarly, the use of DA can reduce the impact of Gaussian noise, as it provides ∼3% accuracy increase using ResNet34 on a BT dataset. Furthermore, simply introducing DA into FL framework shows limited potential (e.g., ∼ 0.3% increase in performance) for skin cancer classification. In addition, the DA method can improve the interpretability of the models using the gradcam++ technique, which offers clinical values. Calibration analysis also demonstrates that using DA provides a lower expected calibration error (ECE) value ∼2% compared to CNN alone on a multi-modality dataset. The codes for our experiments are available at https://github.com/AIPMLab/Domain_Adaptation.

Functional near-infrared spectroscopy (fNIRS) is a non-invasive method for detecting brain activity; however, noise, including motion artifacts and systemic interference, significantly impacts signal quality. To identify an effective denoising technique, we evaluated seven popular methods: Savitzky-Golay (SG) filter, spline interpolation, multivariate disturbance filtering(MDF), traditional band-pass filtering, coefficient of variation (CV) analysis, correlation-based signal improvement (CBSI) and time derivative distribution repair (TDDR), using a public dataset. Experimental results indicate that the SG filter offers the highest signal-to-noise ratio (SNR) of 27.65, and excels at removing high-frequency noise like spikes through averaging techniques. However, the CV method provides the highest contrast enhancement, with a contrast-to-noise ratio (CNR) of 19.41. Importantly, spline interpolation balances signal contrast and primary signal extraction effectively, so that we considered it as our best method. These findings offer valuable guidance for selecting appropriate filters in denoising fNIRS signals. The code and results are available on https://github.com/AIPMLab/Signal_SMC.

Despite the remarkable performance of vision language models (VLMs), such as contrastive language image pretraining (CLIP), the large size of these models is a considerable obstacle to their use in federated learning (FL) systems where the parameters of local client models need to be transferred to a global server for aggregation. Another challenge in FL is the heterogeneity of data from different clients, which affects the generalization performance of the solution. In addition, natural pretrained VLMs exhibit poor generalization ability in the medical datasets, suggests there exists a domain gap. To solve these issues, we introduce a novel method for the federated adversarial adaptation (FAA) of CLIP. Our method, named FAA-CLIP, handles the large communication costs of CLIP using a lightweight feature adaptation module (FAM) for aggregation, effectively adapting this VLM to each client’s data while greatly reducing the number of parameters to transfer. By keeping CLIP frozen and only updating the FAM parameters, our method is also computationally efficient. Unlike existing approaches, our FAA-CLIP method directly addresses the problem of domain shifts across clients via a domain adaptation (DA) module. This module employs a domain classifier to predict if a given sample is from the local client or the global server, allowing the model to learn domain-invariant representations. Extensive experiments on six different datasets containing both natural and medical images demonstrate that FAA-CLIP can generalize well on both natural and medical datasets compared to recent FL approaches. Our codes are available athttps://github.com/AIPMLab/FAA-CLIP.

Radiomics combined with deep learning is an emerging field within biomedical engineering that aims to extract important characteristics from medical images to develop a predictive model that can support clinical decision-making. This method could be used in the realm of brain disorders, particularly autism spectrum disorder (ASD), to facilitate prompt identification. We propose a novel radiomic features [deep radiomic features (DTF)], involving the use of principal component analysis to encode convolutional neural network (CNN) features, thereby capturing distinctive features related to brain regions in subjects with ASD subjects and their age. Using these features in random forest (RF) models, we explore two scenarios, such as site-specific radiomic analysis and feature extraction from unaffected brain regions to alleviate site-related variations. Our experiments involved comparing the proposed method with standard radiomics (SR) and 2-D/3-D CNNs for the classification of ASD versus healthy control (HC) individuals and different age groups (below median and above median). When using the RF model with DTF, the analysis at individual sites revealed an area under the receiver operating characteristic (ROC) curve (AUC) range of 79%–85% for features, such as the leftlateral-ventricle,cerebellum-white-matter,andpallidum, as well as the rightchoroid-plexusandvessel. In the context of fivefold cross validation with the RF model, the combined features (DTF from 3-D CNN, ResNet50, DarketNet53, and NasNet_large with SR) achieved the highest AUC value of 76.67%. Furthermore, our method also showed notable AUC values for predicting age in subjects with ASD (80.91%) and HC (75.64%). The results indicate that DTFs consistently exhibit predictive value in classifying ASD from HC subjects and in predicting age.

Classifier performance enhancement remains a key research area in machine learning, particularly given the scarcity and high cost of labeled data. This study addresses the challenge of improving classifier performance using a novel reinforcement learning (RL) approach. Specifically, our proposed method uses RL to enhance classifier performance by representing the state as a data sample, the action as the standard deviation of Gaussian noise added to the data sample, and the reward as a composite of three components: 1) validation set, 2) noisy validation set, and 3) adversarial validation set. Through six commonly used convolution neural networks (CNNs), our Rl-based approach demonstrates higher performance metrics using breast cancer, COVID-19, and eye disease datasets. In particular, compared to these CNNs, the test accuracy for these classifications improved by 10%∼20% using RL. The implementation code is available at https://github.com/AIPMLab/Graduation-2024/tree/main/oujinajie.

Due to the fast advancement of artificial intelligence (AI), centralized-based models have become critical for healthcare tasks like in medical image analysis and human behavior recognition. Although these models exhibit suitable performance, they are frequently constrained by privacy concerns. To attenuate this, a centralized learning strategy cannot be used in cases where there is a risk of data privacy breach, particularly in healthcare centers. Federated learning (FL) is a technique that allows for training a global model without sharing data by training distributed local models and aggregating them. By implementing FL throughout the training process, we can obtain a model with comparable generalization abilities to centralized learning while maintaining data privacy. This survey provides an introduction to the fundamental concepts and categories of FL, highlights the limitations of the centralized healthcare model, and discusses how FL can address these constraints. We also provide a detailed overview of the healthcare applications using FL models, along with commonly used evaluation metrics and public data sets. In this context, we have implemented a case study to demonstrate how FL can be applied in the healthcare field. Furthermore, we outline the key challenges and future trends in FL.

Despite the progress made by convolutional neural networks (CNNs), optimization for training them remains an important problem. This paper reviews and evaluates existing acceleration methods for CNNs and their potential for domain-specific optimization and customization of FPGA accelerators. For example, the VGG11 network can achieve an average accuracy of 91. 48% in the CIFAR-10 data set while consuming only 1.56w of power, while FINN can achieve comparable performance (88.74%) with only 1.16 mv. This paper provides valuable contributions and prospects for the FPGA-accelerated implementation of CNNs, as well as offers guidance and ideas for research and development in hardware with the AI field.

The rapid advancement of generative artificial intelligence (AI) has initiated transformative changes in diverse sectors, the medical field being a prominent beneficiary. This article provides a comprehensive exploration of the impact of AI in medicine, delineating the range of opportunities and challenges that accompany its adoption. Many existing medical algorithms, while valuable, exhibit a propensity for singular-task orientation, potentially leading to suboptimal interactions and systemic disruption. This paper examines the potential of large language models (LLMs), such as ChatGPT, to improve medical algorithms. It discusses the basic principles of LLM and provides a detailed analysis of ChatGPT, highlighting its exceptional interactive capabilities. Additionally, the study evaluates the real-world applications of ChatGPT in the medical domain, comparing its performance with other leading LLMs through clinically relevant experiments and rigorous analysis. Although ChatGPT is still in its basic versions, this work emphasizes the need to explore and improve its capabilities for medical applications to improve interaction and reduce human costs in the future.

Ahstract- This work examines the impact of the medical metaverse on traditional medicine and the integration of various technologies that have greatly enhanced its capabilities. Technologies such as artificial intelligence (AI), blockchain, Internet of Things (IoT), augmented reality (AR), virtual reality (VR), 5G, big data, natural language processing, and digital twins have facilitated virtual healthcare delivery, allowing clinicians to diagnose patients regardless of distance and receive real-time data. The integration of these technologies has also improved healthcare outcomes and created new healthcare experiences. This paper highlights how the adoption of these technologies has improved healthcare by enhancing patient experiences and outcomes, especially in disease management and specialized care settings. However, the use of the medical metaverse is not without challenges, and this work offers solutions to address them. Despite the potential issues, the medical metaverse offers a new way of delivering healthcare services. Further development and optimization are necessary to realize the full potential of the medical metaverse, but it is a promising and exciting area of healthcare that is worth exploring.

This paper presents the use of federated learning (FL) in healthcare to improve the efficiency and accuracy of medical diagnosis while addressing privacy concerns related to medical data. FL allows data to remain local and trains models independently, with only model parameters communicated to the server. Creating FL models is a popular solution in healthcare systems now, particularly with the increasing use of Internet of Medical Things (IoMT) devices that enable the storage of large amounts of health data. This work provides a comprehensive analysis of the current FL models employed in various applications in healthcare. We applied the FL model to a skin cancer data set and achieved a remarkable result with a classification accuracy of 90% or higher, demonstrating the potential of FL in medical image classification tasks. In this context, we also discuss current bottlenecks and future research directions in FL healthcare.

Domain adaptation (DA) is a technique that uses the knowledge from similar data sets to enhance the generalizability of a model, which proves to be effective in addressing the issue of limited training data. However, due to the complexity of medical data, most advances have occurred in the natural domain and not in the medical field. Furthermore, the majority of advancements are derived from conventional DA datasets, which may introduce result bias. This article presents a comprehensive new analysis of four widely used DA algorithms, tested on more realistic medical data sets to assess the potential of these DA techniques. The examination covers an evaluation of their model performance, discrepancies in data distribution, and the interpretability of the models. For example, the Deep subdomain adaptation network (DSAN) achieves a high level of accuracy on the COVID-19 dataset (89.9%) using Resnet34. Furthermore, the interpretability of the DSAN’s prediction using the skin cancer dataset is implausible. In conclusion, we offer perspectives on the outcomes obtained. Our codes are available at https://github.com/AIPMLab/Domain_Adaptation.

Radiomics has emerged as a rapidly expanding field within precision medicine, offering promising applications for cancer diagnosis, classification, prognosis, prediction, and evaluation. However, the clinical viability of radiomics depends on the stability of its features and models. This paper provides a comprehensive analysis of recent advances in the assessment of the stability, repeatability, and reproducibility of radiomics. It explores various methodologies that aim to improve these performance metrics while addressing the challenges and limitations encountered in clinical applications. The survey emphasizes the critical importance of establishing unified standards that can harmonize data acquisition and analysis practices across multiple institutions. By promoting standardized approaches, this seeks to enable the delivery of more reliable and stable radiomic data, thus improving the overall stability of models and facilitating widespread adoption in clinical settings. Additionally, this paper explores the integration of artificial intelligence (AI) and medical imaging, recognizing it as a potential avenue for fostering secure data sharing while protecting patient privacy. Using AI techniques, healthcare professionals can collaborate and share insights more efficiently, accelerating the progress of radiomics research and its translation into clinical practice. By addressing these critical aspects, our work contributes to the continuous development and broader implementation of radiomics within the field of precision medicine.

The analysis of lung sound signals is an important approach for diagnosing lung diseases, as these signals cover vital physiological and pathological information. However, conventional methods face challenges in capturing weak physiological sound signals, which can affect diagnosis outcomes and heavily rely on the subjective experience and expertise of the auscultating physician. This paper presents a comprehensive analysis of lung sound signals, covering various aspects such as data acquisition, preprocessing, feature analysis, and classification. It then describes the preprocessing and noise reduction processes applied to lung sound signals, highlighting the classical methods used for feature engineering and real-time lung sound classification. Furthermore, the paper explores more advanced machine learning methods in this context and discusses and compares the relevant literature. So far, this paper has contributed to the advancement of lung sound signal analysis and its potential impact on healthcare outcomes.

Recent breakthroughs in artificial intelligence (AI) and deep learning have supported the emergence of innovative applications, among them OpenAI ChatGPT, a chatbot built on natural language processing. ChatGPT swiftly gained worldwide recognition thanks to its exceptional conversational prowess. This paper presents an in-depth examination of the inception and development of the chatbot and highlights its potential applicability within the medical field. Based on the ramifications of implementing ChatGPT in healthcare, the study provides novel solutions that guide future trends. The comprehensive analysis and insights presented in this research contribute to a nuanced understanding of the capabilities and prospective benefits offered by AI-powered chatbots in the medical field. Furthermore, this research emphasizes the importance of maintaining a human-centric approach in healthcare. Although AI chatbots can provide valuable assistance, they should not replace the role of medical professionals. Instead, they should be designed as complementary tools to support and enhance human decision-making processes. The findings facilitate informed decision-making, provide a basis for further investigation, and foster advancements in this promising area of research. Given the transformative potential of Chat-GPT, it is essential for medical professionals and researchers to understand the underlying concepts, challenges, and possibilities associated with the integration of AI chatbots into healthcare systems.

Domain adaptation (DA) techniques have emerged in machine learning, aiming to alleviate distribution differences between training and test sets using information from related domains. However, due to the intricacies of medical data, most advances have been made in the natural rather than medical field. In this paper, we provide a case study on how DA can be applied to the medical domain using a brain tumor data set. We present a full analysis of four widely used DA algorithms in the natural domain, with a focus on their application to the medical domain. The experiments provide detailed results and insightful observations that highlight the performance and medical applicability of these techniques.

Radiomics is the process of extracting valuable features from medical images, which can be further analyzed to improve diagnosis, prognosis, and clinical decision support in oncology, ultimately leading to the advancement of precision medicine. Radiomics is based on the extraction and modeling of imaging features for analysis, serving as an extension of computer-aided diagnosis (CAD) applications. The emergence of Artificial Intelligence (AI) has pushed radiomics by providing new insights and significantly advancing its implementation in clinical practice. Therefore, this paper aims to analyze and demonstrate the clinical aspects of AI-based radiomics, with a specific focus on glioblastoma (GBM), a malignant brain tumor. Furthermore, this paper highlights the current state of radiomics research pertaining to GBM, with potential future challenges and issues in the field.

The rapid development of artificial intelligence (AI) has been made possible by the swift advancement of technology. Recently, there has been a significant increase in interest in the concept of the Metaverse, which is viewed as the ultimate manifestation of the internet. However, building a vast Metaverse requires a substantial allocation of resources to maintain it. Such circumstances pose several challenges, including the disclosure of sensitive data and the problem of non-independent and homogeneous data distribution. In this study, we will focus on explaining federated learning (FL), a methodology that addresses these issues by training localized models and then consolidating them on a global scale to create the ultimate model without sharing the original data. Additionally, we will also discuss the current challenges and potential developments that should be considered in the upcoming years.

Federated learning (FL) has emerged as a promising approach for training machine learning models on distributed data while preserving privacy specifically in the field of medical diagnosis. This paper provides a review of the applications of FL in healthcare, presents the standard FL training process, and suggests future research directions. Our analysis indicates that while FL has shown great potential, more work is needed to optimize its implementation in healthcare settings and ensure the reliability of FL models. Further investigation is necessary to fully realize the potential of FL to improve medical diagnosis.

Artificial intelligence (AI) is revolutionizing the medical field in various ways. However, despite the successful implementation of various AI models in the healthcare field, the complexity, high dimensionality, and nonlinearity of AI hinder its widespread application due to the opacity of the models. The role of eXplainable artificial intelligence (XAI) is basically the same, providing clinicians and experts with predictive, diagnostic, and explanatory information about clinical decision-making patterns, and providing patients with explanations of the reasoning process and results generated by AI, thereby improving their acceptance and trust in such explanations.

Domain adaptation (DA) is a critical technique in machine learning, designed to alleviate distribution differences between training and test sets by leveraging information from similar datasets. This paper presents a practical simulation of both widely-used and recent DA techniques, with a specific focus on unsupervised learning scenarios where labels are only available in the source domain. We thoroughly investigate the impact of these methods on various publicly available datasets, providing detailed explanations of our findings. Furthermore, we conduct an ablation study and elaborate extensively on the simulation results. Our study offers valuable information on the resilience of selected DA algorithms, highlighting the importance of appropriate datasets and neural network architectures with training methodologies to achieve optimal performance in DA tasks.

Coronavirus disease 2019 (COVID-19) has been challenged specifically with the new variant. The number of patients seeking treatment has increased significantly, putting tremendous pressure on hospitals and healthcare systems. With the potential of artificial intelligence (AI) to leverage clinicians to improve personalized medicine for COVID-19, we propose a deep learning model based on 1-D and 3-D convolutional neural networks (CNNs) to predict the survival outcome of COVID-19 patients. Our model consists of two CNN channels that operate with CT scans and the corresponding clinical variables. Specifically, each patient data set consists of CT images and the corresponding 44 clinical variables used in the 3-D CNN and 1-D CNN input, respectively. This model aims to combine imaging and clinical features to predict short-term from long-term survival. Our models demonstrate higher performance metrics compared to state-of-the-art models with area under the receiver operator characteristic curve of 91.44%–91.60% versus 84.36%–88.10% and Accuracy of 83.39%–84.47% versus 79.06%–81.94% in predicting the survival groups of patients with COVID-19. Based on the findings, the combined clinical and imaging features in the deep CNN model can be used as a prognostic tool and help to distinguish censored and uncensored cases of COVID-19.

This article proposes to encode the distribution of features learned from a convolutional neural network (CNN) using a Gaussian mixture model (GMM). These parametric features, called GMM-CNN, are derived from chest computed tomography (CT) and X-ray scans of patients with coronavirus disease 2019 (COVID-19). We use the proposed GMM-CNN features as input to a robust classifier based on random forests (RFs) to differentiate between COVID-19 and other pneumonia cases. Our experiments assess the advantage of GMM-CNN features compared with standard CNN classification on test images. Using an RF classifier (80% samples for training; 20% samples for testing), GMM-CNN features encoded with two mixture components provided a significantly better performance than standard CNN classification ($p < 0.05$). Specifically, our method achieved an accuracy in the range of 96.00%–96.70% and an area under the receiver operator characteristic (ROC) curve in the range of 99.29%–99.45%, with the best performance obtained by combining GMM-CNN features from both CT and X-ray images. Our results suggest that the proposed GMM-CNN features could improve the prediction of COVID-19 in chest CT and X-ray scans.

) compared to 64.0% (p = 0.01) for the 3D CNN model output, 66.8% (p = 0.01) for standard radiomic features, 64.2% (p = 0.003) for CENT, and 57.6% (p = 0.3) for clinical variables. Our results suggest that the proposed GMM-CNN features used with a RF classifier can significantly improve the capacity to prognosticate PDAC patients prior to surgery via routinely-acquired imaging data.

This paper presents a novel set of image texture features generalizing standard grey-level co-occurrence matrices (GLCM) to multimodal image data through joint intensity matrices (JIMs). These are used to predict the survival of glioblastoma multiforme (GBM) patients from multimodal MRI data. The scans of 73 GBM patients from the Cancer Imaging Archive are used in our study. Necrosis, active tumor, and edema/invasion subregions of GBM phenotypes are segmented using the coregistration of contrast-enhanced T1-weighted (CE-T1) images and its corresponding fluid-attenuated inversion recovery (FLAIR) images. Texture features are then computed from the JIM of these GBM subregions and a random forest model is employed to classify patients into short or long survival groups. Our survival analysis identified JIM features in necrotic (e.g., entropy and inverse-variance) and edema (e.g., entropy and contrast) subregions that are moderately correlated with survival time (i.e., Spearman rank correlation of 0.35). Moreover, nine features were found to be associated with GBM survival with a Hazard-ratio range of 0.38-2.1 and a significance level of p < 0.05 following Holm-Bonferroni correction. These features also led to the highest accuracy in a univariate analysis for predicting the survival group of patients, with AUC values in the range of 68-70%. Considering multiple features for this task, JIM features led to significantly higher AUC values than those based on standard GLCMs and gene expression. Furthermore, an AUC of 77.56% with p = 0.003 was achieved when combining JIM, GLCM, and gene expression features into a single radiogenomic signature. In summary, our study demonstrated the usefulness of modeling the joint intensity characteristics of CE-T1 and FLAIR images for predicting the prognosis of patients with GBM.

This paper proposes to use texture features extracted from multispectral microscopic images to detect histopathological abnormalities related to colorectal cancer (CRC): stroma (ST), benign hyperplasia (BH), intraepithelial neoplasia (IN) and carcinoma (Ca). Texture features, based on gray-level co-occurrence matrices (GLCM) and discrete wavelets (DW), are obtained from colon biopsy images, captured using 16 different bands of the visible spectrum. A random forest classifier is used to evaluate the usefulness of these texture features, for each spectral band, on the task of discriminating between the four types of abnormal tissue. Preliminary results on the data of 39 CRC patients show that such features, in particular those based on GLCM and Symlet wavelets, can accurately predict the type of CRC tissue (94% accuracy, 88% sensibility and 100% specificity for Symlet features in the 16thspectral band). These results also reveal important differences in the textural information captured in each band, which could be used to develop more efficient procedures for the diagnosis of CRC.

Sparse multivariate regression techniques like Lasso and Elastic Net are among the most popular approaches for the identification of biomarkers related to brain diseases like Alzheimer's. Because they use L1norm to enforce sparsity, these approaches are often sensitive to differences in voxel intensities within the same scan or across subjects. Also, when few samples are available, such approaches can select voxels that are only correlated by chance, leading to disconnected features that do not correspond to any significant brain structure. To address these challenges, we propose a novel sparse regression method that uses the L0norm for sparse regularization, and imposes spatial consistency constraints on the selected features without requiring an atlas of pre-defined regions. This method uses an efficient optimization strategy based on the Alternating Direction Method of Multipliers (ADMM), that can scale to large data matrices. The performance of the proposed method is evaluated using synthetic data and 3429 T1-weighted (MP-RAGE) images from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Experimental results show our method to outperform Lasso and Elastic Net regression in the recovery of spatially consistent features corresponding to known neuroimaging biomarkers.

This paper concerns the design and the implementation of a transimpedance amplifier (TIA) dedicated to detector of Near Infrared spectroscopy (NIRS). To reduce the effect of the input capacitance on the bandwidth, a bias circuit with low input impedance is connected to input stage. A single ended common source common gate input stage based on a cascode structure is used to get a higher gain bandwidth closed loop transimpedance amplifier. In addition, a higher open loop gain is got by adding second active load. To increase noise circuit performance, a feedback single transistor technique is considered. The TIA is implemented in 0.18 μm CMOS process. Simulation results show a transimpedance gain of 104.2 dBΩ, -3dB bandwidth of 19 MHz and an equivalent input noise current spectral density of 446 fA/√Hz. A comparative study confirms the feasibility of our proposal.

A quantitative texture analysis for discriminating GBM phenotypes in brain magnetic resonance (MR) images is proposed. GBM phenotypes captured using semi-automatic segmentation based on 3D Slicer Scripts. Segmentation was applied on the registered images considered the T1-Weighted and FLAIR sequence. Texture feature has been extracted from the gray level co-occurrence matrix (GLCM) based on GBM phenotypes. Feature vectors are then used in training a minimum distance classifier based on Mahalanobis distance metric. Simulation results for 13 patients show the highest accuracy of 67% based on the feature extraction from GLCM with offset =1 and 8 phases. Preliminary texture analysis demonstrated that the texture feature based on the GLCM is promising to distinguish GBM phenotypes.

Segmentation of image is used from a long time in medical image applications and its study is increased for enhanced the medical diagnosis. This paper concerns a deformable segmentation method for abnormal cells detection by using an improved Level set model which is solved several problems and disadvantages of others segmentation technique. Our approach employed by using real data of carcinoma cells obtained from optical microscopy. Preliminary simulation results showed high performance metrics of the proposed model. Comparative study with manual segmentation demonstrated and confirmed that the level set can be a promise model of abnormal cells detection and in a particularly an irregular shape like carcinoma cells type.

This paper concerns a preliminary study of the relationship between survival time of both overall and progression free survival, and multiple imaging features of patients with glioblastoma. Simulation results showed that specific imaging features were found to have significant prognostic value to predict survival time in glioblastoma patients.