This work introduces a lightweight framework for 3D object classification in point clouds, derived from the Attention-Based Point Cloud Edge Sampling (APES) model. The approach, termed APES-Soft, is based on a systematic ablation study aimed at reducing architectural complexity while preserving classification performance. Three ablation scenarios were investigated, each altering distinct network components to determine the most effective configuration. Scenario II emerged as the most favorable trade-off variant, reaching 93.8% Accuracy alongside 93.7% Precision, 93.8% Sensitivity, 93.7% F1-Score, 93.5% Matthews Correlation Coefficient (MCC), and 89.2% Jaccard index—within the range of reported ModelNet40 results. A supplementary evaluation on ScanObjectNN PB_T50_RS further showed that Scenario II retained competitive performance on a harder benchmark, reaching 80.88%. Furthermore, Scenario II reduced training time to 20.35 hours and memory usage by 21.89%, while using 0.817M parameters, 5.885 GMACs/sample, 11.770 GFLOPs/sample, and 18.141 ms/sample for single-sample inference. Statistical analyses, including ANOVA, Tukey’s HSD, Kruskal–Wallis, and Friedman tests, were interpreted conservatively and do not support claims of formal equivalence or statistically significant superiority. Instead, the results indicate that Scenario II maintains a competitive performance profile while reducing computational cost, highlighting APES-Soft as a reliable and efficient solution for 3D object classification in resource-limited environments.

We introduce a novel technique called ShapedNet to enhance body composition assessment. This method employs a deep neural network capable of estimating Body Fat Percentage (BFP), performing individual identification, and enabling localization using a single photograph. The accuracy of ShapedNet is validated through comprehensive comparisons against the gold standard method, Dual-Energy X-ray Absorptiometry (DXA), utilizing 1273 healthy adults spanning various ages, sexes, and BFP levels. The results demonstrate that ShapedNet outperforms in 19.5% state of the art computer vision-based approaches for body fat estimation, achieving a Mean Absolute Percentage Error (MAPE) of 4.91% and Mean Absolute Error (MAE) of 1.42. The study evaluates both gender-based and Gender-neutral approaches, with the latter showcasing superior performance. The method estimates BFP with 95% confidence within an error margin of 4.01% to 5.81%. This research advances multi-task learning and body composition assessment theory through ShapedNet.

Osteoporosis is a worldwide health issue marked by decreased bone density and degradation of bone tissue, which raises the risk of fractures. Early diagnosis of low bone mineral density (BMD) is crucial in reducing risks by providing appropriate treatment or prevention methods. However, the most common method of measuring BMD is the Dual-energy x-ray absorptiometry, which might not be affordable or accessible to many patients. This study proposes using machine learning methods to predict BMD through anthropometric measurements, anamnesis, age, and sex. A dataset containing 905 patients with their corresponding features and BMD values was also introduced. Different regression algorithms were evaluated, and the model predictions were interpreted using SHapley Additive exPlanations. The approach demonstrated good performance, with an average mean absolute error and mean absolute percentage error of 0.0771 g/cm2and 6.34%, respectively. As a result, this proposed method can potentially become a tool for healthcare professionals to predict BMD in a cost-effective and accessible manner.

Stroke is a leading cause of death globally, with higher chances of recovery when prompt and accurate diagnosis is followed by appropriate treatment. Various neuroimaging techniques, including computed tomography (CT), are used for stroke detection. Computer-aided diagnosis (CAD) systems can capture information imperceptible to the human eye, making them valuable tools in stroke diagnosis. This study proposes a novel approach for segmenting hemorrhagic stroke in CT scans using Deep Learning. Specifically, we evaluate the effectiveness of SSD, YOLO-v4, and YOLACT as prompts for the Segment Anything Model (SAM) in hemorrhagic stroke segmentation. Additionally, we compare YOLACT and SAM for segmentation performance. The methods showed promising results, with the proposed SAM and MobileSAM achieving an accuracy of 99.82%, while YOLACT attained 99.74%. The use of zeroshot and one-stage models demonstrated exceptional efficiency in addressing the segmentation challenges in medical images.

In this article, we propose a new approach to classify individuals with or without depression using electroencephalogram (EEG) signals. The methodology involves transforming EEG signals into the frequency domain, enabling the identification of spectral markers and their representation through directed graphs. The resulting spectral tensor is then employed to train the neural network model ResNet-101 with a fast processing architecture, which is ideal for embedded systems. Our method achieves high accuracy (91.03%), precision (91.81%), recall (91.03%), and F1-Score (90.96%) rates on the MODMA dataset, utilizing a subset of three channels (Fp1, Fpz, and Fp2). Our results demonstrate the method’s suitability for applications in embedded systems with a fast prediction time of 0.820 seconds per sample and a training time of 0.226 seconds per sample. These findings suggest promising prospects for mobile applications, emphasizing the method’s potential for deployment in diverse clinical settings.

Diabetic retinopathy (DR) is a complication resulting from diabetes mellitus, caused by damage to the blood vessels in the retina due to excess glucose in the blood. This condition is one of the leading causes of vision loss in adults with diabetes. Early detection and appropriate treatment are crucial to prevent the progression of the disease. The main objective of this study is to develop an advanced tool for classifying retinal image photographs. To achieve this, we used the Normalizer Free Neural Network (NFNet) as a feature extractor, combining it with machine learning techniques through the method of transfer learning. We evaluated the effectiveness of our model by comparing its results with those of various established and recognized convolutional neural network architectures in the literature. The results show that, using the NFNet, we achieved an accuracy of 99.83% and an F1-Score of 99.46% when combined with support vector machines using radial basis function kernel. Given the significance of these results, the next step is to explore the possibility of developing a diagnostic support tool using the developed methodology.

The advancement of machine learning techniques has brought significant progress to the classification of brain tumors, proving essential for early diagnoses and effective treatments. This study focuses on evaluating the performance of feature extractors and classifiers for the binary division of brain tumors. Five extractors were employed: Zernike Moments, DWT (Discrete Wavelet Transform), LBP (Local Binary Pattern), GLCM (Gray-Level Co-occurrence Matrix) and HU Moments. The HU Moments extractor stood out with the shortest average execution time, 13.56 microseconds. Subsequently, the Grid Search technique was employed to identify the best hyperparameters for five classifiers: Random Forest, K-Nearest Neighbors, Support Vector Machine, Multilayer Perceptron, and Naive Bayes. Four evaluation metrics were used, prioritizing precision and F1-Score. The results revealed that the combination of Zernike Moments and KNeighbors achieved a test precision of 100.00%, surpassing the results of previous studies.

Free Space Optics (FSO) has emerged as a crucial communication modality in recent years, especially with the implementation of techniques associated with the 5G network and its subsequent advancements. This study proposes the use of Machine Learning, including Convolutional Neural Networks (CNNs) and classifiers such as Random Forest, Naïve Bayes, Multilayer Perceptron, and Support Vector Machine, to analyze and classify the effects of rain in FSO systems. The methodology involves generating a database through simulations in OptiSystem, followed by preprocessing the images to retain only relevant eye diagrams. A CNN is then applied as a feature extractor, with its attributes used as input for the classifiers. After classification, it is possible to discern the type of rain associated with each entry in the database. The results highlight effective combinations, such as VGG16 and VGG19 with the Bayes classifier for distances of 500m and 1 km, and Random Forest with InceptionV3 for 1 km, achieving accuracies above 90% and 99%, respectively. This study offers a practical and effective approach to signal quality analysis in FSO systems, emphasizing the importance of Machine Learning, especially CNNs, in this context. These techniques allow for precise analysis adaptable to weather conditions, providing valuable insights for future enhancements and real-world implementations.

This study introduces an innovative methodology for Parkinson’s disease (PD) classification using electroencephalogram (EEG) signals. The approach involves transforming EEG signals into the frequency domain, identifying spectral markers through directed graphs, and consolidating multichannel data. The resulting spectral tensor is classified using diverse neural network architectures. The proposed method achieves high accuracy rates (Dataset I: 99%, Dataset II: 96.7%), precision (Dataset I: 99.92%, Dataset II: 97.22%), and F1-Score (Dataset I: 99.91%, Dataset II: 96.63%). The findings underscore the methodology’s efficacy in emphasizing relevant EEG signal characteristics, contributing to efficient PD diagnosis.

Anthropometric measurements are used in several fields of study and can be used in public health as an indicator of cardiovascular risk; they can also be used as a parameter for making tailored clothes or even for reconstructing the body for nutritional monitoring purposes. Thus, the automatic estimation of these measurements can improve anthropometric processes, even more so if it is based on 2D images, as this represents a gain due to the low cost of implementing this technology. Our work presents an approach to estimating these anthropometric measurements through images using Machine Learning. Furthermore, in this work, we propose a dataset containing 913 samples with images and measurements. Finally, we performed experiments to analyze the influence of patient information on the estimation of measurements, as well as a gender analysis to get insights from the dataset. In our approach, with specific configurations, we reached the mark of 0.778 ± 0.083 cm using MAE, 1.096 ± 0.292 cm using RMSE, and 2% using MAPE, differing on the experiment, gender, and classifier.

Detecting alcoholism is challenging because unreli-able patient information impairs early diagnosis and treatment. However, EEG exams offer more reliable data. This paper introduces a new two-dimensional approach for the automatic diagnosis of alcoholism using EEG signals. The approach ana-lyzes changes in neural activity and highlights the impact of high and low-frequency signals by representing each patient as a node in a graph. It combines two-dimensional feature extraction from classical approaches to recent state-of-the-art Computer Vision techniques, such as Transfer Learning with Convolutional Neural Networks(CNN). The methodology to evaluate our proposal used 22 combinations of the traditional feature extraction methods and 34 combinations of recent CNN architectures used as feature extractors combined with Graph Neural Network and Graph Convolutional Network (GCN). The best results were achieved by combining GCN and GLCM-6 in Accuracy (99.35%), Fl-score (99.22 %), and Recall (98.62 %), outperforming state-of-the-art methods by 2%. This approach can support computer-aided diagnoses and early clinical detection of the disease.

A multitude of diseases can afflict the human body, each potentially causing a range of health problems. Among the most devastating are brain diseases that originate from tumors, affecting the well-being of countless people worldwide. Brain tumors are a pathology that results in numerous sequelae, significantly impacting public spending due to the costs of hospital clinical treatments, exams, and medicines for patients. These factors are linked to the problem, resulting in significant financial impacts on the health sector worldwide. The problem of detection and segmentation in medical images provides new solutions through different methods based on computer vision. This study proposes a fully automatic model based on the Internet of Things (IoT), capable of classifying, detecting, and segmenting magnetic resonance images of brain tumors. The proposed model can classify MR images, detect the tumor region, and segment it through deep extractors and classifiers, combined with deep learning using the Detectron2 network for brain tumor detection and fine-tuning for brain tumor segmentation. The model was trained and tested with the dataset (Brain MRI Segmentation - LGG Segmentation Dataset), obtaining excellent classification results using a transfer learning model of DenseNet201 + SVM RBF, achieving 93.10% accuracy. For the Detectron2 network, 99.38% accuracy was achieved, with 99.41% for brain tumor segmentation. The fully automatic IoT-based model (Health of Things) efficiently classified, detected, and segmented brain tumors in MR images, surpassing different reputable works in the literature.

Different pathologies can cause different public health problems and diseases that can cause the death of millions of people. Hemorrhagic stroke is a silent disease affecting millions of people worldwide, causing disorders and, in most cases, being fatal. CAD systems are systems capable of aiding medical diagnosis, and studies based on computed tomography images have been increasingly widespread in the literature and in intelligent applications aimed at medical diagnosis by images. The proposed study brings a new approach based on the use of deep learning for the detection of hemorrhagic stroke, in which it obtained an accuracy of 99.37% in the detection of stroke. A new stroke segmentation method called (Divisible Cell-Segmentation) was developed and used to delineate stroke borders by means of fine-tuning. The method is based on initialization within the hem-orrhagic stroke previously detected by the Detectron2 network using the concept of splittable cells for segmentation, where they multiply within the detected image, causing the segmentation of the edge of the hemorrhagic stroke when traveling through the stroke where they divide into the two-dimensional region. The New method obtained excellent results with 99.82% accuracy for the segmentation of CT images of cerebral hemorrhagic strokes, surpassing the state-of-the-art. The new method presented an optimization in processing time, surpassing works found in the literature, as well as greater efficiency and computational cost in the detection and segmentation of hemorrhagic stroke edges in computed tomography images.

Strokes are among the top three global causes of death. The diagnosis of stroke is commonly made based on the symptoms displayed and, specifically, on the the results of the neuroimaging tests. Currently, computed tomography (CT) is the fastest, most accessible, and most financially viable neuroimaging method. Thus, Computer-Aided Diagnosis (CAD) systems that can analyze the CT images are essential for obtaining diagnostic information. In order to aid medical diagnosis, this paper proposes a new automatic method of segmenting the areas affected by a hemorrhagic stroke in the CT images based on deep learning, using Mask R-CNN combined with Windowing of Parzen, Clustering and Region Growing methods through the techniques of fine-tuning. Our best model achieved an accuracy of 99.72% and the segmentation time of 6.49 s. Thus, we surpass methods already consolidated in the literature, with both manual and automatic initialization, and even Deep Learning techniques using fine-tuning. We validate the proposed model by comparing it to existing methods that segment images from the same dataset. We overcoming the state of art among the analyzed models, which proves the efficiency of our method for systems based on CAD.

Abstract The use of computational techniques in the processing of histopathological images allows the study of the structural organization of tissues and their changes through diseases. This study aims to develop a tool for classifying histopathological images from breast lesions in the benign and malignant classes through magnification scales by an innovative way of using transfer learning techniques combined with machine learning methods and deep learning. The BreakHis dataset was used in the experiments, consisting of histopathological images of breast cancer with different tumor enlargement scales classified as Malignant or Benign. In this study, various combinations of Extractor‐Classifiers were performed, thus seeking to compare the best model. Among the results achieved, the best Extractor‐Classifier set formed was CNN DenseNet201, acting as an extractor, with the SVM RBF classifier, obtaining accuracy of 95.39% and precision of 95.43% for the 200X magnification factor. Different models were generated, compared to each other, and validated based on methods in the literature to validate the experiments, thus showing the effectiveness of the proposed model. The proposed method obtained satisfactory results, reaching results in the state‐of‐the‐art for the multi‐classification of subclasses from the different scale factors found in the BreakHis dataset and obtaining better results in the classification time.

Works that use point cloud avoid wasting time and cost of collection, using simulators and datasets available in the literature. In this way, there is access to an unlimited and organized amount of point clouds, an ideal setting for deep learning networks and Vehicular ad hoc networks (VANETs). However, models trained with synthetic data present problems when applied to real-world data.This work proposes the use of deep learning in the recognition of 3D objects captured with a Light Detection and Ranging (LIDAR), including a pre-processing stage. In addition, it is proposed two datasets, a real-world and a syntetic; each dataset includes three classes. A method of pre-processing is proposed to circumvent the distribution discrepancies of the proposed datasets and the existing datasets from literature, such as ModelNet. We use deep learning with the PointNet method, as it supports raw data from point clouds as input to the network. We performed three evaluation approaches: training and testing steps with the proposed datasets using(1)Lidar3DNetV1, which is a proposed network in this paper,(2)PointNet, and (3) classification of ModelNet datasets using Lidar3DNetV1. The proposed network achieved 98.33% of accuracy and a testing time of$88~\mu \text{s}$in the synthetic dataset, while in the real-world dataset, the network reached 98.48% and$145~\mu \text{s}$in accuracy and testing time, respectively.

Keeping a certain balance of body fat is essential for a healthy life, and proper nutrition is fundamental. One of the most worrying malnutrition problems is obesity, which plays a significant risk factor for chronic diseases like cardiovascular diseases, diabetes, and cancer. The Dual-energy X-ray absorp-tiometry (DXA) is the most accurate and automatic method that returns the body fat percentage; however, this method is expensive and not easily found at clinics. A lower-cost way of estimating the body fat percentage is through anthropometric measures. However, the literature has shown that estimating body fat percentage on women is challenging. In this work, we propose an approach specialized in gender to estimate body fat percentage using machine learning. Another contribution of this work is a dataset, BodyFat-163 (BF-163), containing the 12 anthropometric measures and the body fat percentage from DXA exams collected by a specialist. The dataset consists of 163 individuals (84 males and 79 females). Our experiments involved a variety of methods of regression, which includes Random Forest Regression, Extreme Gradient Boosting, Decision Tree, Support Vector Regression, Multilayer Perceptron Regression, and Least Square Support Vector Regression. The experiment results were evaluated with the metrics Mean Absolute Error (MAE), Root Mean Square Error, Mean Squared Logarithmic Error, and R2score. Our gender-based approach successfully estimates the body fat percentage achieving a MAE = 2.756, and R2 = 0.68 on the male set and MAE = 3.869, and R2 = 0.69 on the female set.

Modern computer vision techniques applied to radiographic studies are presented as an alternative to assist the specialist in screening and diagnosing the respiratory syndrome (SARS-CoV-2), assisting in clinically severe cases, such as acute pneumonia, acute respiratory failure, organ failure, and death. This work proposes a screening method based on the Internet of Medical Things (IoMT) based on deep learning techniques for the classification of COVID-19 from chest X-ray (CXR) exams. The proposed system called Computer-Aided Remote medical diagnostics System (CARMEDSys) applied to the diagnosis of COVID-19 consists of three main stages: 1) segmentation of the lung region in X-ray images, 2) deep extraction of attributes from the filtered pulmonary area and 3) Prediction patient status with machine learning assistance. The performance of CARMEDSys was evaluated considering twelve different deep neural networks, via the transfer of learning. Besides, the performance of this approach is evaluated against recent studies for the classification of healthy patients, with pneumonia, or with COVID-19. The evaluation methodology considered two different sets of radiographic images, reaching Sensitivity (99.97%), F1-Score (99.43%), and Accuracy (98.89%) promising to distinguish patients with pneumonia and COVID-19 combining DenseNet201 as attribute extractor with Support Vector Machine with radial basis function, exceeding up to 12.31 % sensitivity for prediction of COVID-19 recent related works.

Systems developed to work with computational intelligence have become very efficient, and in some cases obtain more accurate results than evaluations by humans. Hence, this work proposes a new online approach based on deep learning tools according to the concept of transfer learning to generate a computational intelligence framework for use with the Internet of Health Things (IoHT) devices. This framework allows the user to add their images and perform platform training almost as easily as creating folders and placing files in regular cloud storage services. The trials carried out with the tool showed that even people with no programming and image processing knowledge were able to set up projects in a few minutes. The proposed approach is validated using three medical databases, which include cerebral vascular accident images for stroke type classification, lung nodule images for malignant classification, and skin images for the classification of melanocytic lesions. The results show the efficiency and reliability of the framework, which reached 91.6% Accuracy in the stroke images and lung nodules databases, and 92% Accuracy in the skin images databases. This prove the immense contribution that this work can bring to assist medical professionals in analyzing complex examinations quickly and accurately, allowing a large medical examination database through a consolidated collaborative IoT platform.

Diabetes is a major cause of blindness, kidney failure, heart attacks, stroke and lower limb amputation according to World Health Organization (WHO). Complications from poor diabetes management lead to the Diabetic Retinopathy (DR) which is a leading cause of acquired blindness in the working-age population worldwide. WHO estimated that DR accounts for ≈ 15-17% of all cases of total blindness in the US and Europe, 7% of all cases in China and Mongolia. In Brazil, according to the Ministry of Health, the disease affects 7.6% of the population. A cost saving intervention includes screening and treatment for retinopathy. Detecting the different lesions related to DR plays an important role towards the stage detection, prediction, and prevention. Our challenge here is to design a deep learning neural network able to fully detect such lesions in digital retinal fundus image to help building robust and scaled solutions to tackle this urgent diabetes scenario. These results show that the Diavision Portable device had a better performance using GLCM, LBP and SVM, presenting 100% for both evaluation metrics considered. Using Messidor dataset were obtained better performance with VGG16 and SVM, archiving 100% for all metrics. In terms of feature extraction time, the GLCM and VGG16 presented acceptable times, respectively, 17.63ms and 37.87ms.

This work combines the transfer learning technique with Convolutional Neural Networks (CNN) to classify the pathology within BI-RADS levels 3 and 4 for malignancy of breast masses. These BI-RADS levels represent the zone of the uncertainty of the degree of malignancy of the found mass, making it difficult for the human experts in classifying as malignant or benign. Eleven CNN architectures were used as feature extractors and combined with four traditional classification models: Bayes, Multilayer Perceptron (MLP), Support Vector Machines, and Random Forest. The combination DenseNet201-MLP achieved an accuracy higher than 63%, surpassing the performance of a human expert by 9.0%.

Recent statistics have shown that the main difficulty in detecting alcoholism is the unreliability of the information presented by patients with addiction; this hampers early diagnosis and reduces the effectiveness of treatment. However, electroencephalogram (EEG) exams can contribute with more reliable data for this analysis. This paper proposes a new approach for the automatic diagnosis of patients with alcoholism. It offers a method for examining the EEG signals from a two-dimensional perspective according to changes in the neural activity, highlighting the influence of high and low-frequency signals. This approach combines Transfer Learning and Con-volutional Neural Networks (CNN) to EEG signals analysis. The methodology to evaluate our proposal used 21 combinations of the classification traditional methods and 35 combinations of recent CNN architectures used as feature extractors combined with the following classical classifiers: Gaussian Naive Bayes, K-Nearest Neighbor (k-NN), Multilayer Perceptron (MLP), Random Forest (RF) and Support Vector Machine (SVM). CNN MobileNet combined with SVM achieved the best results in Accuracy (95.33%), Precision (95.68%), F1-Score (95.24%), and Recall (95.00%). This combination outperformed traditional methods by up to 8%. Thus, this approach is applicable as a classification stage for computer-aided diagnoses, useful for the triage of patients, and clinical support for the early diagnosis of this disease.

The identification of different textile fabrics is a task commonly learned in practice and, therefore, is considered a very strenuous and costly form of learning, causing annoyance to the individual who performs it. Based on this context, this paper proposes a new method for classifying textile fabrics, based on the development of a computer vision system using Convolutional Neural Network (CNN). CNN works as a feature extractor by incorporating the concept of Transfer Learning. Using Transfer Learning allows a pre-trained CNN model to be reused for a new problem. In order to highlight the high performance of CNN, an analysis is performed with feature extractors established in the literature. Parameters such as Accuracy, F1-Score, and processing time are considered to evaluate the efficiency of the proposed approach. For the classification were used Bayesian Classifier, Multi-layer Perceptron (MLP), k-Nearest Neighbor (kNN), Random Forest (RF), and Support Vector Machine (SVM). The results show that the best combination is the CNN architecture DenseNet201 with SVM (RBF), obtaining an accuracy of 94% and F1-Score of 94.2%.

With the parameters set of the industry 4.0 and the growth of new intelligent and interconnected systems, those concepts have enabled innovative advances in several areas, among them, solutions for different renewable sources of energy efficiency. This study has as its objective the detection of short-circuit faults in wind turbines utilizing an analysis of vibration images. Using the Internet of things (IoT) context, we created a methodology to check the operating condition of a machine. The proposed method obtained excellent results, presenting a new interconnected approach to the industry 4.0 for short-circuit detection of induction generator squirrel-cage model, a widely used and growing model in the renewable energy market. Using the Random Forest-LBP combination, we achieved 87.9% accuracy with no false positives between the normal class and the failure classes.

Huntington's disease is a hereditary condition in which brain nerve cells rupture over time. This work proposes a new method for the detection of Huntington's disease using digitised voice signals by diseased and healthy volunteers while they were reading Lithuanian poems. In this approach, the produced features by voice signals suffer a dimensionality reduction to optimize the prediction stage. The performance evaluation regarded 186 speech exams and 24 volunteers, combining twelve audio signal feature extractors with classification models. The results indicate an excellent performance, reaching precision and accuracy over 99 percent with prediction time below 1 second. This approach shows promising results indicating its usability to improve the medical diagnosis via computer-aided diagnosis.

For any Automated Fingerprint Identification System, the quality of its images is vital to ensure the proper accuracy of the whole system. When the quality of an image is not satisfactory, enhancement processes may be applied to help the extraction of the fingerprint features. There are several enhancement techniques, and their suitability depends on the features of the original fingerprint image. Choosing the best enhancement method is crucial because these procedures do not always improve the image quality, and may even worsen it. This work addresses this topic and presents a classifier based on Convolutional Neural Networks (CNNs) that automatically chooses the most suitable enhancement method for a specific image and applies it, but only if necessary. Our solution avoids an excessive human effort to select the best enhancement process and also requires no further training. We evaluated our proposal using FVC's datasets, and results show the benefits of CNN-based feature extractors and that our solution was able to improve the quality of digital printing through the adaptive application of enhancement filters.

The World Health Organization (WHO) registered around 3 million deaths caused by Chronic Obstructive Pulmonary Disease (COPD), representing 5% of all deaths registered in 2015. Computed tomography (CT) is among the main exam for clinical diagnosis of lung diseases. However, the first challenge experienced by the radiology specialist is to define the region of interest. Thus, the identification of diseases using systems of computer-aided diagnosis (CAD) medical via image processing techniques offers more accuracy and agility for diagnosis. In this paper, we propose a new automatic segmentation of lungs in CT images. Our method uses a deep learning technique called Mask Region-Based Convolutional Neural Network (Mask R-CNN), combined with an adaptive active contour method called Fast Morphological Geodesic Active Contour (FGAC). The proposed method was evaluated with 72 lung images, consisting of 24 images of healthy volunteers and 48 of unhealthy patients. Our approach achieved promising results with Accuracy of 98.93%, Matthews Correlation Coefficient of 95.84%, Hausdorff Distance of 5.48, DICE of 96.47%, and Jaccard of 93.24%. Thus, our method surpasses a recent classic approach that also uses FGAC as a segmentation method.

Nutrition is a significant determinant of health, the resolution of many nutritional issues, initially requires an anthropometry examination. Body measures provide data for studying the relationship between diet, nutritional status, and health. Manual and automatic methods can perform body measurements. The manual method usually uses an anthropometric tape. However, the automatic process uses the equipment of Dual-energy X-ray absorptiometry (DXA). Our work presents a new approach to calculate body measures using 2D Camera Images, applying Digital Image Processing, Convolution Neural Networks, and Machine Learning techniques. The dataset used contains 38 exams, for each exam, has four digital images and the dimensions of body measurements, performed by a specialist. The methods used in this work for segmentation were Dense Human Pose Estimation - CNN with the Bayesian, K-Nearest Neighbors, Support Vector Machine, Decision Threes, Adaptive Boosting, Random Forest, Multilayer Perceptron and Expectation-Maximization classifiers. The approach with Dense Human Pose Estimation and Expectation-Maximization reached the best results, with mean squared error (MSE) always bellow 4.606 ± 3.412 cm when compared with specialist measures.

Among illnesses, heart diseases are accounted for as one of the most responsible for deaths. Precise and fast diagnoses increase the patient's chances to receive treatment time. A non-invasive and low-cost way to diagnose it is by using Electrocardiogram (ECG). In this paper, we propose a way to diagnosis two types of heart arrhythmia, by using the ECG record as an image. To access the performance of our system, five feature extraction methods well-known in literature are used along with five different classifiers are tested. We were able to identify heart disorders with over 96.00% of accuracy, using a vanilla neural-network, Multilayer Perceptron (MLP), and Local Binary Patterns (LBP) from ECG images. This investigation has shown promising results from a medical point-of-view.

The papers in this special section examines the use of interactive virtual environments in the field of neuroscience. Virtual environments is a technology able to establish a relationship between the user and the environment created, enabling real-time integration with controlled virtual objects. A virtual environment can be explored through visual and haptic devices, without real restrictions. The iteration derives from the communication between human actions and the outcome of these actions, processed by the computer generating a response inside the virtual environment. The interaction can be passive, such as watching television, or active, for instance in the case of users manipulating their body movements or a particular object inside a virtual scenario.

Around 6 to 8 thousand people die annually in the world due to the fact of having taken a pill erroneously. Some works have already proposed pill recognition systems commonly using attributes related to shape, color, and others. In this work, we propose a pill feature extractor to classify them based on shape and color (CoforDes). The proposed method was compared with the descriptors GLCM, SCM, LBP, Tamura, Fourier and the Zernick, Central, Statistical and Hu Moments. Three classifiers (KNN, SVM, and Bayes) were used to evaluate the feature extractors. The attributes were extracted in 0.01006 seconds in the PILL BR dataset and 0.00810 seconds in the NIH NLM PIR dataset using CorforDes, obtaining an accuracy of 99.85% in the PILL BR dataset and 99.82% in the NIH NLM PIR dataset. The specificity was 99.82% in the PILL BR dataset and 99.91% in the NIH NLM PIR base. The results show that CoforDes is an excellent feature extractor for the extraction of drug pill images and they can be embedded in real-time applications due to their rapid processing.

Lung cancer corresponds to 26% of all deaths due to cancer in 2017, accounting more than 1.5 million deaths globally. Considering this challenging situation, several computeraided diagnosis systems have been developed to detect lung cancer at early stages, which increases the patients' survival rate. Motivated by the success of deep learning in natural and medical image classification tasks, the proposed approach aims to explore the performance of deep transfer learning for lung nodules malignancy classification. For this, convolutional neural networks (CNN), such as VGG16, VGG19, MobileNet, Xception, InceptionV3, ResNet50, Inception-ResNet-V2, DenseNet169, DenseNet201, NASNetMobile and NASNetLarge, were used as features extractors to process the Lung Image Database Consortium and Image Database Resource Initiative (LIDC/IDRI). Next, the deep features returned were classified using Naive Bayes, MultiLayer Perceptron (MLP), Support Vector Machine (SVM), Near Neighbors (KNN) and Random Forest (RF) classifiers. Additionally, to compare the classifiers performance with themselves and with other ones in literature, the evaluation metrics Accuracy (ACC), Area Under the Curve (AUC), True Positive Rate (TPR), Precision (PPV), and F1-Score were computed. Finally, the best combination of deep extractor and classifier was CNN-ResNet50 with SVM-RBF, which achieved ACC of 88.41% and AUC of 93.19%. These results are equivalent to related works, even just using a CNN pre-trained on non-medical images. For this reason, deep transfer learning proved to be a relevant strategy to extract representative imaging biomarkers for lung nodule malignancy classification in chest CT images.

Stroke is an injury that affects the brain tissue, mainly caused by changes in the blood supply to a particular region of the brain. As consequence, some specific functions related to that affected region can be reduced, decreasing the quality of life of the patient. In this work, we deal with the problem of stroke detection in Computed Tomography (CT) images using Convolutional Neural Networks (CNN) optimized by Particle Swarm optimization (PSO). We considered two different kinds of strokes, ischemic and hemorrhagic, as well as making available a public dataset to foster the research related to stroke detection in the human brain. The dataset comprises three different types of images for each case, i.e., the original CT image, one with the segmented cranium and an additional one with the radiological density's map. The results evidenced that CNN's are suitable to deal with stroke detection, obtaining promising results.

Goat leather is responsible for a large part of the income generated by the most diverse clothing products, but the lack of modernization of some stages of leather production is still very evident, and may lead to divergent opinions regarding the quality of leather among tanning industries and finishing industries. In this paper it is proposed a new approach to aid goat leather qualification specialists based on the position of the found failures in the goat leather. There are two steps in the proposed approach. The first one is find the failure regions and the last one is extract some feature from the failure map founded. In this paper the authors use Gray Level Co-occurrence Matrix (GLCM), Local Binary Patterns (LBP) and Structural Co-occurrence Matrix (SCM) as feature extractors from leather images. The classification task is executed by k-Nearest Neighbors (KNN), Multi Layer Perceptron (MLP) and Support Vector Machine (SVM). The combination of the LBP attribute extractor with the MLP classifier has 90% accuracy rates for the classification of regions with failure, but for quality classification of the leather, the SVM classifier has the best results (86% of accuracy rate), also using LBP. The results show that the proposed approach can be used to aid the specialists to classify the quality of the goat leather.

In this paper, we propose a new localization and navigation approach for mobile robots using topological maps and classification with reject option applying convolutional neural networks (CNN) for feature extraction in omnidirectional images. The use of CNN as feature extractor is based on the concept of Transfer Learning. Reject option is used to improve the task of the classifiers, querying information from the topological map. With the objective of evidencing the high performance of the technique considered, an analysis is made between several feature extractors and classifiers, established in the literature. Parameters such as processing time and accuracy are calculated to prove the credibility and effectiveness of the approach, since these properties are fundamental in the analysis of embedded systems. Considering the proposed approach, CNN stands out among the other feature extractors, as it generated the best results in extraction time and accuracy. It obtained an average accuracy of 99.86% and an extraction time of 0.1517s, proving to be a relevant method for the localization and navigation activities.

The use of energy sources boosts technological development over time, but the continuous and unrelenting usage of fossil fuel energy has resulted in harmful impacts to the ecosystem. Then came an alert in the scientific community, generating discourses and research in the field of renewable energy, for example, wind, solar and biofuels. Among various fields of study in wind power, this work is concerned in the study of electric generators, specifically in the detection of faults, using magnetic axial flux acquired from a squirrel cage induction generator (SCIG). We compare the performance of classifiers Bayes, Support Vector Machines, Optimum Path Forest and K- Nearest Neighbor in three different feature extraction methods, High Order Statistics (HOS), Fourier Transform and Structural Co-occurrence Matrix (SCM). We were able to correctly identify 99.4% of the Normal conditions using a Bayesian classifier alongside with as well identify incipient short-circuits with 95.6% of hit rate. SVM had a lowers false positive, 0.6% and 2% in HOS and SCM, respectively, while identifying 98% of the Normal conditions, so is more suitable for binary classification between fault and non-fault short-circuits.

Computed tomography (CT) stands out among the exams used by computer-aided diagnosis in medical imaging as it provides the visualization of internal organs such as the lungs and their structures. This paper focuses on the segmentation of lungs using a three-dimensional region growing (3D RG) method and the registration toolkit ITK library. To evaluate the proposed segmentation method, we used 30 exams from a dataset built in previous work with different types of images and diagnosis. The results were compared against two segmentation tools from OsiriX software, one by toolbox and the other by plugin MIA, and a third, by a medical specialist, which was considered as the gold standard. The results are given in a three-dimensional model with the VTK library and show that the lung segmentation, based on the 3D RG method presented here, produce an average segmentation correctness of 98.75%. Therefore, this method could be part of a computer-aided medical diagnosis system in Pulmonology.

Cardiovascular diseases (CVD) are the leading cause of death worldwide, and every year more people die of these diseases. Aiming to assist medical diagnoses through Computerized Tomography (CT) scans, this work proposes a new approach to segment CT images of the brain damaged by stroke. The proposed method takes into account two improvements of the level set method based on the likelihood of Normal distribution. The first improvement is to handle the grayscale image input according to a range analysis of the image intensity scale, adopting 80 HU for the window width and 40 HU for the center level. In addition, we propose an optimal level set initialization, where the zero level set is determined by analyzing the brain density. These improvements to the level set method generate efficient stroke segmentation in CT images of the brain. The results of the proposed method are compared against those of the level set algorithm based on the coherent propagation method, and also those from the Watershed and Region Growing algorithms using a ground truth built by a specialist. The experimental results show that the proposed method presents superior performance, and that it is a promising tool to assist medical diagnoses.

This work presents a new approach to segment hemorrhagic stroke based on an active contour method called Optimum Path Snakes (OPS). The Analysis of Human Tissue Densities (AHTD) was introduced as the evolution of the Pulmonary Density Analysis feature extractor. The results of this approach were compared with the Region Growing, Watershed and Level Set based on the coherent propagation methods to segment the stroke region. Accuracy, Matthews Correlation Coefficient, Dice Coefficient, Hausdorff Distance and Harmonic Means metrics were used to verify the efficacy of OPS over the other methods. The OPS method along with the AHTD extractor presented the best results, which demonstrated the potential for this approach to be used in medical diagnosis systems.