Studies of cardiovascular disease risk prediction by machine learning algorithms often do not assess their ability to generalize to other populations and few of them include an analysis of the interpretability of individual predictions. This manuscript addresses the development and internal and external validation of predictive models for the assessment of risks of major adverse cardiovascular events. Global and local interpretability analyses of predictions were conducted towards improving model reliability and tailoring preventive interventions. The models were trained and validated in a retrospective cohort with the use of data from Hospital das Clínicas da Faculdade de Medicina de Ribeirão Preto, Brazil. Data from Beth Israel Deaconess Medical Center, USA, were used for external validation. Eight machine learning algorithms, namely Penalized Logistic Regression, Random Forest, XGBoost, Decision Tree, Support Vector Machine, k-Nearest Neighbors, Naive Bayes and Multi-Layer Perceptron were trained to predict a 5-year risk of major adverse cardiovascular events and their predictive performance was evaluated regarding accuracy, ROC curve (receiver operating characteristic), and AUC (area under the ROC curve). LIME and Shapley values methods interpreted individual predictions. Random Forest showed the best predictive performance in both internal validation (AUC = 0.87; Accuracy = 0.79) and external one (AUC = 0.79; Accuracy = 0.71). Compared to LIME, Shapley values provided explanations more consistent with exploratory analysis and importance of features. Among the machine learning algorithms evaluated, Random Forest showed the best generalization ability, both internally and externally, and Shapley values for local interpretability were more informative than LIME ones, which is in line with our exploratory analysis and global interpretation of the final model. Machine learning algorithms with good generalization and accompanied by interpretability analyses are recommended for assessments of individual risks of cardiovascular diseases and development of personalized preventive actions.

This paper proposes a multimodal retrieval system for Chest CT (called ChestFinder) that combines image and report embeddings’ into a filter-and-query strategy. ChestFinder is composed of three modules, namely (i) text transformation, (ii) feature extraction, and (iii) ranking aggregation. Text transformation is conducted by a fine-tuned Generative Pre-training Transformer model (GPT), and the image embeddings are extracted after (i) training a Residual Neural Network (ResNet-50), and (ii) reducing and scaling the encoded vectors. ChestFinder produces one list of similar images and another of related reports for each query input composed of a Chest CT with a radiology report. Then, the ChestFinder ranking aggregation module fuses those two lists to produce the final ordered set of retrieved objects, as in a top-k query. The aggregation is performed by a fine-tuned Threshold Algorithm (TA) whose weights are calculated by a Multi-Layer Perceptron (MLP) trained to label the reports. To examine the quality enhancement brought by this multimodal search, we constructed a dataset of Chest CTs from our University Hospital PACS/RIS systems by filtering distinct cases diagnosed with emphysema (one finding per case and with at least two radiologists agreeing on the diagnosis). A holdout experimental evaluation showed the ChestFinder search achieved higher Accuracy and Sensitivity than content-only top-k searches. Results also indicated quality gains drawn from the adjustments of ChestFinder modular components: (i) fine-tuned GPT achieved up to 0.89 F1-Score in data testing with a stable train/validation ratio for radiology reports, (ii) fine-tuned GPT significantly outperformed the zero-shot approach as well as a fine-tuned BERT, (iii) non-weighted ranking aggregation increased the search accuracy in up to 10%, and (iv) fine-tuned TA outperformed the baseline and non-weighted ranking aggregation in up to 52%.

Osteoporosis is a systemic disorder that reduces the bone mineral density, increasing the vertebrae's fragility and proneness to fracture. Although the bone densitometry index t-Score is a solid marker for the osteoporosis diagnosis, its measure alone is insufficient to predict the future development of fragility fractures. A complementary approach to address vertebral bone characterization is the analysis of magnetic resonance imaging (MRI) by radiomic features, which model vertebral bodies' morphological properties after color and texture. Radiomic features have been employed for detecting fragility fractures in related work, but, to the best of our knowledge, no study has been conducted on their suitability to recover similar, diagnosed cases that could hint at future fractures. We fulfill this gap by designing a Content-based Image Retrieval (CBIR) tool with embedded radiomic features, which uses past cases recovered from an annotated database to (i) identify an existing fragility fracture in a query vertebra and (ii) predict a fracture to a query vertebra from an aging patient. The proposed CBIR was evaluated on a reference database of 273 vertebral bodies from sagittal T2-weighted MRIs. The results indicate our fine-tuned approach spotted fragility fractures accurately$(\mathrm{F}1-\text{Score} =0.83,\ \text{Precision} =0.83,\ \text{AUC} =0.81,\ \text{CI} =95\%)$. We also investigated the CBIR potential to predict fractures in a case study regarding three patients from the reference database (confirmed osteoporosis, MRI in [2012–2017]). The system correctly inferred the prediction of future fractures for query vertebrae, which were confirmed a few years later (MRI in [2018–2021]). Such empirical findings suggest CBIR can support a differential diagnosis in the assessment of local fragility fractures.

Many patients suffer from chronic skin lesions, commonly known as ulcers. The size evolution of chronic wounds provides meaningful clues regarding the patient's clinical state for healthcare professionals and caretakers. Many studies have been proposed in recent years to support the treatment of skin ulcers. However, there is a lack of practical solutions, as existing studies are not targeted at immediate use in daily medical practice. In this work, we propose URule, an essentially practical framework for segmentation and measurement of skin ulcers. URule-App, a mobile instance of the framework, analyzes images taken by a common camera from a mobile device. The segmentation requires the user to manually outline the outsider region of both the wound and the measurement tool. URule-Seg segments the image and estimates the wound area. The user can further improve the estimated area by manually informing the span of a centimeter in the image. The experimental evaluation reveals that URule can accurately segment ulcer wounds semi-automatically, with an average F-Measure of 0.8 for segmentation, and processing measurement tools better than the manual process in three out of five tested rulers.

Tissue segmentation in photographs of lower limb chronic ulcers is a non-intrusive approach that supports dermatological analyses. This paper presents 2PLA, a method that combines supervised and unsupervised learning strategies for enhancing the segmentation of dermatological wounds. Given an ulcer photo captured according to a fixed protocol, 2PLA first phase performs a pixelwise classification of points of interest, whereas pre-processing filters are employed for the smoothing of image noise. The cleaned image is further sent to the 2PLA divide-and-conquer second phase. It builds upon SLIC superpixel construction algorithm for dividing the lower limb into regions of interest with well-defined borders, and clusters the superpixels by taking advantage of the similarity-based DBSCAN algorithm. We set up the phases of our method by using a real annotated set of dermatological wounds, and empirical evaluations on representative samples up to 100,000 points showed a compact Multi-Layer Perceptron with Levenberg-Marquardt training algorithm (Cohen-Kappa = .971, Sensitivity = .98, and Specificity = .98) outperformed other classifiers as 2PLA first phase. Additionally, experimental trials on DBSCAN with five distance functions (L1, L2, L∞, Canberra, and BrayCurtis) indicated L1function provided fewer groups in comparison to the competitors, and the number of clusters was an exponential decay to the similarity ratio. Accordingly, we used the elbow criterion for finding the L1-based DBSCAN threshold as 2PLA second phase parameterization. We evaluated the fine-tuned setting of our method over a labeled set of ulcer images, and wounded tissues were segmented within a .05 Mean Absolute Error ratio. These results illustrate the impact of learning parameters on 2PLA as well as the method efficacy for wound segmentation.

Content-Based Image Retrieval (CBIR) has proven to be a suitable complement to traditional text-based searching. CBIR applications rely on two main steps, namely the representation of the images, and the similarity measuring between two represented images. Although modern segmentation and learning algorithms enable the accurate representation of local and global features within an image, how to properly compare the segmented objects is still an open issue. In this study, we propose a new comparison method called Counting-Labels Similarity Measure (CL-Measure). Our approach calculates the similarity between two images by comparing the labeled regions within these images and by balancing the influence of each label according to its predominance in both non-metric and metric fashion. The experiments on a real dataset of dermatological ulcers show that CL-Measure achieves a higher Precision for all values of Recall compared to its competitors in retrieval tasks.

This study presents an analysis of classification techniques for Computer-Aided Diagnosis (CAD) regarding ulcerated lesions. We focus on determining influence of both color and texture in the automated image classification and its implication. To do so, we assayed a dataset of dermatological ulcers containing five variations in terms of tissue composition of lesion skin: granulation (red), fibrin (yellow), callous (white), necrotic (black), and a mix of the previous variations (mixed). Every image was previously labelled by experts regarding this red-yellow-black-white-mixed model. We employed specially designed color and texture extractors to represent the dataset images, namely: Color Layout, Color Structure, Scalable Color, Edge Histogram, Haralick, and Texture-Spectrum. The first three are color feature extractors and the last three are texture extractors. Following, we employed the Symmetrica Uncert Attribute Eval method to determine the features suitable for image classification. We tested a set of classifiers that follows distinct paradigms over the selected features, achieving an accuracy ratio of up to 77% in terms of images correctly classified, with the area under the receiver operating characteristic (ROC) curve up to 0.84. The classification performance and the selected features enabled us to determine that texture features were more predominant than color in the entire classification process.

The use of computer systems for management and imaging analysis has brought many advantages to medical organizations. This paper comprises a survey regarding the Lyria PACS system implemented in 2011 at Hospital das Clinicas, Ribeirão Preto USP - Brazil. It presents the benefits of this system, the main challenges faced during its implementation, the heterogeneity of infrastructure resources and the monetary economy obtained after the installation of Lyria PACS in short and medium periods of time. The results are shown using statistics and growth of use.

This study presents some of the most important checkpoints found in the deployment or migration projects for PACS (Picture Archiving and Communication System) environments to the cloud. The checkpoints were mapped to the risk assessment table, proposed by ENISA (European Union Agency for Network and Information Security), for applications running from cloud computing services. Then a risk analysis model for PACS environments in the cloud was proposed and evaluated.

Mobile devices are emerging as an important technology to be incorporated into computerized health systems in order to enhance and support the health services provided to the patient. With the popularization of mobile and wireless technologies, the motivation and encouragement for the use of these technologies increases in support of the reliability and quality of data. An usability evaluation was conducted to identify problems and deficiencies presented in the mobile application. The combination of these two contexts creates a new term called mobile health, which has motivated much discussion of how greater access to mobile phone technology can be leveraged to mitigate the numerous pressures faced in the medical care of health systems. This paper presents the development of an electronic record system for tablets with a focus on pre-hospital patient care.

Our study aimed to develop a system for computer-aided diagnosis of vertebral compression fractures (VCFs) using magnetic resonance imaging (MRI), to help in the differentiation between malignant and benign VCFs. Lumbar spine MRI was used to acquire T1-weighted images in the sagittal plane. Images from 63 consecutive patients (38 women, 25 men, mean age 62.25 ± 14.13 years) with at least one VCF diagnosis were studied. Contrast and texture features were extracted from manually segmented images of 103 vertebral bodies with VCFs. The classification of malignant vs. benign VCFs was performed using the k-nearest neighbor (KNN) classifier with the Euclidean distance. Using a KNN classifier with k=3, feature selection, and 10-fold cross-validation, we obtained a value of the area under the receiver operating characteristic curve of 0.913.

Content-based image retrieval systems are employed in several areas. One of the most prominent area is the medical field, due to the huge volume of digital images daily generated in healthcare institutions employed for decision making. There are several works applying CBIR techniques over medical images. However, the great majority of them do not verify whether the systems are actually considered by the specialists as a pontential aid in a real environment. In order to fill this research void in the literature, this work explores user experiments in a CBIR system involving resident physicians and radiologists. To do so, we developed a CBIR system according to requirements provided by the specialists and employed a methodology to analyze the effectiveness of the system for supporting them in clinical routine. The methodology aims at evaluating the system's impact in the user's decision, inquiring the specialists about the image classification and their degree of certainty in different situations using the system. By analyzing the obtained results we can argue that the proposed methodology joined with our medical CBIR system presented a high acceptance and viability rate regarding the radiologists interests in the clinical practice domain, providing a novel approach to analyze CBIR systems under realistic conditions.

We present color image processing methods for the analysis of images of dermatological lesions. The focus of the present work is on the application of feature extraction and selection methods for classification and analysis of the tissue composition of skin lesions or ulcers, in terms of granulation (red), fibrin (yellow), necrotic (black), callous (white), and mixed tissue composition. The images were analyzed and classified by an expert dermatologist into the classes mentioned above. Indexing of the images was performed based on statistical texture features derived from cooccurrence matrices of the RGB (Red, Green, and Blue), HSI (Hue, Saturation, and Intensity), L*a*b*, and L*u*v* color components. Feature selection methods were applied using the Wrapper algorithm with different classifiers. The performance of classification was measured in terms of the percentage of correctly classified images and the area under the receiver operating characteristic curve, with values of up to 73.8% and 0.82, respectively.

A challenge in content-based retrieval of image exams is to provide a timely answer that complies to the specialist's expectation. In many situations, when a specialist gets a new image to analyze, having information and knowledge from similar cases can be very helpful. However, the semantic gap between low-level image features and their high level semantics may impair the system acceptability. In this paper we propose a new method where we gather from the physicians the visual patterns they use to recognize anomalies in images and apply this knowledge not only in the preprocessing of the images, but also on building feature extractors based on these visual patterns. Moreover, our approach generates feature vectors with lower dimensionality diminishing the ldquodimensionality curserdquo problem. Experiments using computed tomography lung images show that the proposed method improves the precision of the query results up to 75%, and generates feature vectors up to 94% smaller than traditional feature extraction techniques while keeping the same representative power. This work shows that perception-based feature extraction combined with the image context can be successfully employed to perform similarity queries in medical image databases.

The research on Content-Based Image Retrieval (CBIR) is growing in relevance at a fast pace. Algorithms and tools for CBIR can help decision-making processes, for example allowing the specialist to retrieve cases similar to the one under evaluation. However, the main reservation about using CBIR is the semantic gap, which is the divergence among automatic results and what the user is expecting. We propose the ldquoperceptual parameterrdquo, which allows changing the relationship between the feature extraction algorithms and the distance functions, aimed at finding the best integration of both from the specialist's point of view. This work integrates the three main elements of similarity queries: the extracted features from the images, the distance function employed to quantify the similarity and the similarity perception from the user. These three elements allowed to build the "similarity operators". The experiments performed show that the new perceptual parameter can narrow the semantic gap between what the system retrieves and what the specialist expects.

In this paper we present a new method, called IDEA, which employs association rules to assist in medical image diagnosis. IDEA mines association rules, relating visual features with the knowledge gotten from specialists, and employs the associations to suggest possible diagnoses for a given medical image. IDEA incorporates two new algorithms called Omega and ACE. Omega performs simultaneously feature selection and data discretization very efficiently with linear cost on the number of feature values. ACE is a new associative classifier, which has the particular ability of suggesting multiple keywords to compose the diagnosis for a given medical image. The IDEA method has an important characteristic that makes it different from other CAD methods: it suggests multiple diagnosis hypotheses for an image and ranks them based on a measure of quality. The IDEA method was implemented in a prototype (IDEA system) for radiologists evaluate it. The radiologists showed enormous interest in employing the system to aid them in their daily work. The IDEA system was applied to real datasets and the results presented high accuracy (up to 96.7%). The results testify that association rules are well-suited to support the diagnosing task.

In this paper, we propose a method based on association rule-mining to enhance the diagnosis of medical images (mammograms). It combines low-level features automatically extracted from images and high-level knowledge from specialists to search for patterns. Our method analyzes medical images and automatically generates suggestions of diagnoses employing mining of association rules. The suggestions of diagnosis are used to accelerate the image analysis performed by specialists as well as to provide them an alternative to work on. The proposed method uses two new algorithms, PreSAGe and HiCARe. The PreSAGe algorithm combines, in a single step, feature selection and discretization, and reduces the mining complexity. Experiments performed on PreSAGe show that this algorithm is highly suitable to perform feature selection and discretization in medical images. HiCARe is a new associative classifier. The HiCARe algorithm has an important property that makes it unique: it assigns multiple keywords per image to suggest a diagnosis with high values of accuracy. Our method was applied to real datasets, and the results show high sensitivity (up to 95%) and accuracy (up to 92%), allowing us to claim that the use of association rules is a powerful means to assist in the diagnosing task.

In this paper we present HEAD, the Human Encephalon Automatic Delimiter, a new and efficient method for skull-stripping in T1-weighted MRI that combines an unique histogram analysis with binary mathematical morphology. In our experiments we use real images with highly variable noise ratios and intensity non-uniformity. We evaluate our results based on manually generated true masks and the well known Jaccard metric, achieving accuracy close to 99%. We compare our method with the popular Brain Extractor Surface algorithm (BSE), which in the same experiments achieved less than 95% of accuracy.

In this paper we present a framework based on association-rules to help diagnosis of mammogram abnormalities. Our framework - SuGAR - combines low-level features automatically extracted from images with high-level knowledge gotten from specialists to mine association rules, suggesting possible diagnoses. Our framework is optimized, in the sense that it combines, in a single step, feature selection and discretization, reducing the mining complexity. The framework was applied to real datasets and the results show high sensitivity (up to 95%) and accuracy (up to 92%), allowing us to claim that association rules can effectively aid in the diagnosing task.

The content-based image retrieval (CBIR) has great interest of the medical community, because it is capable of retrieval similar images stored in servers that have known pathologies. However, an efficient and reliable CBIR solution has not been achieved yet, due to the complexity of the medical image and the great volume they represent. This work proposes a new methodology based on higher processing provided by Grid Computing technology to achieve the CBIR using the registration algorithms. The registration procedure use two metrics, square difference metric (SDM) and cross correlation (CC). Both metrics showed higher efficiency, SDM obtained precision average of 0.83% (breast image) and 0.94% (head image), the CC showed precision of 0.81% (breast) and 0.52% (head). The higher computational cost related to the registration algorithms was amortized by Grid Computing, that was capable of ensure data secure and represent a low cost solution to small clinics and public hospitals. Grid technologies open new opportunities to investigate the contribution on applying the registration algorithms to CBIR and new advances are expected.

This paper proposes the use of fractal analysis as a means to discriminate textured segmented regions of medical images. We show that the use of the fractals can boost the representation level of traditional image features allowing high rates of precision when answering similarity queries over images employing a variance weighted Manhattan distance. The cost to compute the fractal measurements is linear on the image size, what makes their use a suitable choice for large sets of images.

This paper presents a new picture archiving and communication system (PACS), called cbPACS (content-based PACS), which has content-based image retrieval resources. cbPACS answers similarity (range and nearest-neighbor) queries, taking advantage of a metric access method embedded into the image database manager. The images are compared via their features, which are extracted by an image processing system module. The system works on features based on the color distribution of the images through normalized histograms as well as metric histograms. Metric histograms are invariant with regard to scale, translation and rotation of images and also to brightness transformations. cbPACS is prepared to integrate new image features, based on the texture and shape of the main objects in the image.