Multistatic imaging techniques, such as Synthetic Aperture Ultrasound (SAU) or Plane Wave Imaging (PWI), offer several advantages in terms of image quality for diagnostic ultrasound imaging. However, the vast amount of data generated by these methods can be challenging to process and store. To address this issue, various compression techniques have been developed. In this work, we propose a compression method based on a physical approach utilizing evanescent wave components in the radio frequency (RF) data. The basic idea behind our approach is to eliminate the higher frequencies in the data that are no longer necessary, due to the limited spatial sampling frequency. By doing so, we can reduce the amount of data without sacrificing noticeable amounts of image quality, as shown by simulation results. An additional advantage of our approach is that no decompression steps have to be conducted if the image reconstruction algorithm operates in the spatial frequency-domain.
Glaucoma is a condition that affects the optic nerve, with loss of retinal nerve fibers, increased excavation of the optic nerve, and a progressive decrease in the visual field. It is the leading cause of irreversible blindness in the world. Manual classification of glaucoma is a complex and time-consuming process that requires assessing a variety of ocular features by experienced clinicians. Automated detection can assist the specialist in early diagnosis and effective treatment of glaucoma and prevent vision loss. This study developed a deep learning model based on vision transformers, called ViT-BRSET, to detect patients with increased excavation of the optic nerve automatically. ViT-BRSET is a neural network architecture that is particularly effective for computer vision tasks. The results of this study were promising, with an accuracy of 0.94, an F1-score of 0.91, and a recall of 0.94. The model was trained on a new dataset called BRSET, which consists of 16,112 fundus images of patients with increased excavation of the optic nerve. The results of this study suggest that ViT-BRSET has the potential to improve early diagnosis through early detection of optic nerve excavation, one of the main signs of glaucomatous disease. ViT-BRSET can be used to mass-screen patients, identifying those who need further examination by a doctor.
The resting brain has been extensively investigated for low frequency synchrony between brain regions, namely Functional Connectivity. However the other main stream of the brain connectivity analysis that seeks causal interactions between brain regions, Effective Connectivity, has been still little explored. Inherent complexity of brain activities in resting-state, as observed in Blood Oxygenation-Level Dependant fluctuations, calls for exploratory methods for characterizing these causal networks [1]. To determine the structure of the network that causes this dynamics, it is developed a method of identification based on least squares, which assumes knowledge of the signals of brain activity in different regions. As there is no access to functional Magnetic Resonance Imaging, data it is developed a model to obtain the Blood Oxygenation Level Dependent signals and it is implemented a reverse hemo-dynamic function. To assess the performance of the created model Monte Carlo simulations have been used.
Colorectal cancer is a leading health concern worldwide, with late detection being a primary challenge due to its often-asymptomatic nature. Routine examinations like colonoscopies play a pivotal role in early detection. This study harnesses the potential of Deep Learning, specifically convolutional neural networks, in enhancing the accuracy of polyp detection from medical images. Three distinct models, YOLOv5, YOLOv7, and YOLOv8, were trained on the PICCOLO dataset, a comprehensive collection of polyp images. The comparative analysis revealed YOLOv5’s submodel S as the most efficient, achieving an accuracy of 92.2
This research addresses the significant challenge of automating the annotation of medical images, with a focus on capsule endoscopy videos. The study introduces a novel approach that synergistically combines Deep Learning and Content-Based Image Retrieval (CBIR) techniques to streamline the annotation process. Two pre-trained Convolutional Neural Networks (CNNs), MobileNet and VGG16, were employed to extract and compare visual features from medical images. The methodology underwent rigorous validation using various performance metrics such as accuracy, AUC, precision, and recall. The MobileNet model demonstrated exceptional performance with a test accuracy of 98.4
The average life expectancy has increased in the last decades, but it is still necessary to promote a healthy and active aging. Although older people need to have more caution when doing exercise, it is an important aspect to maintain a healthy and active life. Access to facilities by older people to do exercises in groups and having someone guiding them in the execution of the exercises usually poses challenges for them, due to transportation-related limitations. Our goal is to enable older people to do exercises in their homes, allowing them to have exercise plans that meet their needs. Group exercises with their friends is also an important feature to motivate them to exercise more often. In this paper, we propose a proof-of-concept of a smart gym at home system based on defined scenarios and its installation in a near-real scenario. The proof-of-concept provides a guide to a given exercise plan, with video demonstrations of each exercise. The interaction of the system was carefully thought, since older people are typically not comfortable with technologies. The main interaction method in this proof-of-concept is speech, since the users are distant from the system's devices while doing the exercises.
The capsule endoscopy examination is a common medical procedure used to diagnose and treat gastrointestinal tract diseases without the need for invasive procedures. Images captured during the examination can reveal a wide range of abnormalities, including lesions, inflammation, ulcers, bleeding, and tumors. However, interpreting these images can be a challenge for physicians since the videos contain a large number of frames (images) to be analyzed. To attempt to achieve an early diagnosis and reduce the lethality of gastrointestinal system pathologies, the use of artificial intelligence has been extensively studied to alleviate the workload of healthcare professionals, as the large number of images resulting from an examination makes manual categorization of each image challenging. This work studied the use of machine learning methods such as OneClassSVM and XGBoost based on features extracted from deep neural networks and compared them to traditional convolutional neural network methods, such as the ResNet152 network. The Kvasir-Capsule and ERS datasets were used to evaluate the proposed methods, focusing on classifying images as normal or abnormal. Among the evaluated methods, XGBoost showed the best results among others, with a weighted F1-score of 0.71 on the ERS dataset and 0.87 on the Kvasir-Capsule dataset. The class imbalance in both datasets proved to be a continuous challenge, adding to the challenge of the low quantity images in the ERS dataset.
Inherited retinal diseases such as Retinitis Pigmentosa and Stargardt's disease are genetic conditions that cause the photoreceptors in the retina to deteriorate over time. This can lead to vision symptoms such as tubular vision, loss of central vision, and nyctalopia (difficulty seeing in low light) or photophobia (high light). Timely healthcare intervention is critical, as most forms of these conditions are currently untreatable and usually focused on minimizing further vision loss. Machine learning (ML) algorithms can play a crucial role in the detection of retinal diseases, especially considering the recent advancements in retinal imaging devices and the limited availability of public datasets on these diseases. These algorithms have the potential to help researchers gain new insights into disease progression from previous classified eye scans and genetic profiles of patients. In this work, multi-class identification between the retinal diseases Retinitis Pigmentosa, Stargardt Disease, and Cone-Rod Dystrophy was performed using three pretrained models, ResNet101, ResNet50, and VGG19 as baseline models, after shown to be effective in our computer vision task. These models were trained and validated on two datasets of autofluorescent retinal images, the first containing raw data, and the second dataset was improved with cropping to obtain better results. The best results were achieved using the ResNet101 model on the improved dataset with an Accuracy (Acc) of 0.903, an Area under the ROC Curve (AUC) of 0.976, an F1-Score of 0.897, a Recall (REC) of 0.903, and a Precision (PRE) of 0.910. To further assess the reliability of these models for future data, an Explainable AI (XAI) analysis was conducted, employing Grad-Cam. Overall, the study showed promising capabilities of Deep Learning for the diagnosis of retinal diseases using medical imaging.
Cup and optic disc segmentation has become one of the main objects of study in the field of creating and improving machine learning-oriented models due to the importance of vision for human beings and the ability to assist physicians in diagnosing ocular problems. Within this context, this study presents a new method based on the proportional division of images concerning features extracted from the sample set. These samples go through a pre-processing step involving image resizing before going to deep feature extraction and K-means clustering, thus dividing the set for validation and training. Soon after, the amount of samples is increased through data augmentation before going on to the U-Net training. The proposed method has been evaluated on the public RIM-ONE and DRISHTI-GS datasets, and presented promising results in the segmentation of both structures, with emphasis on obtaining the value of 92.2
Mauritius has a very high prevalence of diabetes, hypertension and cardiovascular diseases, which are often linked to bad eating habits. The use of mHealth applications to bring about positive behavioural changes is practically non-existent in Mauritius since the local unique culinary scene makes it difficult to use such existing applications. The primary objective of this study is to create a customized mHealth application that caters for the peculiarities of the eclectic Mauritian cuisine. MauLifeStyle, a web-based application that caters for calorific content of different types of Mauritian foods and typical activities of the Mauritian population, has been developed and tested using recruited participants. The MauLifeStyle application has been evaluated based on different criteria of MoHTAM, which is a refined technology acceptance model for smart mobile phones. The clinical outcomes were not statistically significant following three months of monitoring. However, the participant feedback supports the demand for digitally mediated self-education for motivated individuals. The developed mHealth application comprises four main functionalities: Blood Glucose Monitoring, Food Intake, Health and Fitness, and Calorie Monitor. It is anticipated that over time, MauLifeStyle has the potential to reduce the prevalence of Diabetes in Mauritius and consequently alleviate the financial burden on the healthcare system through the application of preventive medicine.
The wireless capsule endoscopy is a non-invasive imaging method that allows observation of the inner lumen of the small intestine, but with the cost of a longer duration to process its resulting videos. Therefore, the scientific community has developed several machine learning strategies to help reduce that duration. Such strategies are typically trained and evaluated on small sets of images, ultimately not proving to be efficient when applied to full videos. Labelling full Capsule Endoscopy videos requires significant effort, leading to a lack of data on this medical area. Active learning strategies allow intelligent selection of datasets from a vast set of unlabelled data, maximizing learning and reducing annotation costs. In this experiment, we have explored active learning methods to reduce capsule endoscopy videos' annotation effort by compiling smaller datasets capable of representing their content.
This study focuses on investigating different CNN architectures and assessing their effectiveness in classifying Diabetic Retinopathy, a diabetes-associated disease that ranks among the primary causes of adult blindness. However, early detection can significantly prevent its debilitating consequences. While regular screening is advised for diabetic patients, limited access to specialized medical professionals can hinder its implementation. To address this challenge, deep learning techniques provide promising solutions, primarily through their application in the analysis of fundus retina images for diagnosis. Several CNN architectures, including MobileNetV2, VGG16, VGG19, InceptionV3, InceptionResNetV2, Xception, DenseNet121, ResNet50, ResNet50V2, and EfficientNet (ranging from EfficientNetB0 to EfficientNetB6), were implemented to assess and analyze their performance in classifying Diabetic Retinopathy. The dataset comprised 3662 Fundus retina images. Prior to training, the networks underwent pre-training using the ImageNet database, with a Gaussian filter applied to the images as a preprocessing step. As a result, the Efficient-Net stands out for achieving the best performance results with a good balance between model size and computational efficiency. By utilizing the EfficientNetB2 network, a model was trained with an accuracy of 85
The ability to recognize emotions through facial cues, in childhood, is helpful for social interactions. Image-based facial emotion recognition models need low computing power, but cannot accept sequential information from video data. Conversely, video-based facial emotion recognition models require high computational power, so it cannot be easily applied in a low computing environment. In this paper, we propose a method that classifies the emotion from facial expression video data by applying threshold using an image-based model. The proposed method improves the accuracy of 3.67
Prostate cancer is one of the most common types of cancer in men. The ISUP grade and Gleason Score are terms related to the classification of this cancer based on the histological characteristics of the tissues examined in a biopsy. This paper explains an approach that utilizes and evaluates pre-trained models such as ResNet-50, VGG19, and InceptionV3, regarding their ability to automatically classify prostate cancer and its severity based on images and masks annotated with ISUP grades and Gleason Scores. At the end of the training, the performance of each trained model is presented, as well as the comparison between the original and predicted data. This comparison aims to understand if this approach can indeed be used for a more automated classification of prostate cancer.
Organ at Risk segmentation has an important role in the meticulous planning of radiotherapy for cancer treatment. Its primary objective is to safeguard the surrounding healthy tissues while precisely directing radiation to target cancer cells. Currently, this task needs manual intervention by physicians, a process that can be time-consuming and susceptible to errors. Consequently, the integration of automatic segmentation methods offers the potential to accelerate the delineation of organs during radiotherapy planning. In this study, we applied Eff-Unet, a fully convolutional neural network model, and trained it to perform the semantic segmentation of trachea in computed tomography images. This approach yielded a noteworthy 78.9
Home care of people with dementia (PwD) is mainly organized and carried out by non-professional family caregivers, who struggle to interpret the needs of PwD correctly and are confronted with the challenging behavior of their relatives. Although support services for family caregivers are widespread in Germany, they are rarely used due to the fact that information is poorly organized and relatives are faced with a flood of disorganized, outdated, and confusing content. Due to the technical development of chatbot technologies (ChatGPT), chatbots gain more and more relevance. Based on the new technological possibilities, we developed an online communication and service platform with an integrated chatbot within the eDEM-CONNECT project, with the aim of making structured and easily understandable information accessible for family caregivers. This work focuses on the development of a chatbot pipeline that has broad domain knowledge through a provided ontology on the topic of agitation of PwD. This allows the chatbot to provide relevant and peer-reviewed information to family members. In our approach, a patient history is first taken based on several diagnostic questions so that relevant information can be output in a later step. For this purpose, we demonstrate that agitations in natural language can be correctly recognized by the used BERT model and that our developed chatbot is able to select further diagnostic questions based on the predictions of a Markov logic network.
Growing obesity has been a worldwide issue for several years. This is the outcome of common nutritional disorders which results in obese individuals who are prone to many diseases. Managing diet while simultaneously dealing with the obligations of a working adult can be difficult. Today, people have a very fast-paced life and sometimes neglect food choices. In order to simplify the interpretation of the Nutri-score labeling this paper proposes a method capable of automatically reading food labels with this format. This method is intended to support users when choosing the products to buy based on the letter identification of the label. For this purpose, a dataset was created, and a prototype mobile application was developed using a deep learning network to recognize the Nutri-score information. Although the final solution is still in progress, the reading module, which includes the proposed method, achieved an encouraging and promising accuracy (above 90
Breast cancer is a neoplasm that mainly affects women above the age of 45. However, an increase in the incidence of this disease among young women has been observed. Although it is considered a cancer with a good prognosis when diagnosed early, early detection remains a challenge. In Brazil, the mortality rate due to breast cancer remains high, which is directly related to the late diagnosis of the disease. To contribute to the reduction of this rate, the development of effective early detection techniques is essential. These techniques can assist in diagnosing the disease at its initial stages, enabling quicker treatment and thereby increasing the chances of a cure. Computer-aided detection and diagnosis systems have been developed and improved in the field of computing. These systems base their accuracy and reasoning on data obtained through a combination of computer vision techniques, such as pattern recognition and machine learning. When applied, these techniques assist doctors and specialists in data analysis to provide diagnostic support and treatment planning. This significantly enhances a patient's chances of recovery. More recently, within the machine learning field, Deep Learning has become a prevalent focus of research due to its ability to automatically extract relevant features for the target task. In this work, the methodology proposed employs Convolutional Neural Networks for machine learning. While the results obtained are not superior to those in the literature, they are close and generally require fewer computational resources for training the selected networks after the selection process.
Before a dental professional performs any procedure or diagnosis, they need to know the patient's dental arch. For that, it is common for them to ask the patient to take a panoramic radiograph. The use of neural networks to assist this professional in this stage is not recent, and most studies use segmentation networks to solve the problem. However, the segmentation result does not make explicit the specific position of the tooth and its numbering according to the international system (FDI), presenting only more specific details. In this study, we aimed to use a powerful and efficient detection neural network called You Only Look Once v8 to perform automated tooth detection and numbering based on FDI, using a dataset that contains 166 anonymized and deidentified panoramic dental radiographs of patients from Noor Medical Imaging Center, Qom, Iran, and are public. Labels were created using an online tool for production in the YOLO standard. The metrics used to evaluate the trained model were precision, recall, and mAP50. The results of each were 0.95818, 0.95505, and 0.97384. The conclusion of the study uses the model training generated a weight to test the model in a real-world scenario.
This work focuses on the application of image processing techniques to segment and analyze images of brain sections with the aim of facilitating early diagnosis of brain tumors. The aim is to delineate specific regions of the brain, such as the cranial, intracranial, and encephalic regions, for subsequent geometric analysis. The process involves image pre-processing, conversion to polar coordinates, determination of contour points, Fourier Series approximation, and the use of the Least Square Method to obtain accurate representations of the regions. The proposed approach was tested on Magnetic Resonance Images of three different brains, showing its capability to accurately delineating the targeted regions. The results highlight the potential of signal processing techniques for analyzing brain images and provide insights for further research in this area.
