1st IFMBE Latin American Conference on Digital Health

Medical images play a crucial role in diagnosis and anatomical studies. Convolutional Neural Networks (CNNs), especially when utilizing transfer learning, have become essential for image classification in medical applications. However, the interpretability of CNNs remains a significant challenge. This paper proposes a methodology based on feature visualization and heatmaps to interpret the aspects of input images influencing label assignments. It includes ad-hoc CNNs and those based on the ImageNet dataset, including VGG19, ResNet, Inceptionv3, and MobileNet. Various classification schemes, training approaches, and data augmentation methods are explored. Feature visualization is performed on pretrained CNNs, so input patterns maximizing each output neuron response are identified. Heatmaps, which identify the regions of the image responsible for determining the classification, are generated. The proposed approach is tested on diverse medical image datasets, and a Python toolbox based on Keras is available on GitHub. Results demonstrate the reliability of the proposed approach in interpreting patterns and associating them with labels in medical images.

Cancer, a global health problem, claims many lives annually, with breast cancer (BC) being the most prevalent, accounting for 11.7% of the 19.3 million cases reported in GLOBOCAN2020. Early detection of breast tumors is critical yet challenging. Digital X-ray mammography aids in detection but interpreting mammograms can overlook abnormalities due to the volume of images. To improve detection, computer-assisted-diagnosis (CAD) methods have been proposed. This study aims to classify BC using image-based descriptors and Machine Learning (ML) Hard Voting Ensemble technique, combining algorithms like Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM). The process involves preprocessing, feature selection using the Boruta algorithm, and training RF, LR, and SVM. A Hard Voting Ensemble model is then created from their predictions. Validation includes five metrics and statistical analysis, including Receiver Operating Characteristic Curve (ROC) analysis for the Area Under the Curve (AUC). The SVM model performed best with AUC = 0.9647 and an accuracy (ACC) equals to 0.9286, followed by RF with AUC = 0.9624 and LR with AUC = 0.9553. The ensemble model achieved AUC = 0.9471 and ACC = 0.9286.

This study investigates the application of an advanced deep learning technique in the segmentation of medical images, focusing on a specific case study: the segmentation of the bladder in Cone Beam Computed Tomography images. The main objective is to develop an accurate and practical approach to delimiting the region of interest, contributing to advances in medical image processing. The methodology adopted uses a set of computed tomography images made available by the radiotherapy sector of the Hospital Universitário de Brasília (HUB), which served to train, test and evaluate the model. The database consists of 1,932 CT images (256 \(\times \) 256) and 1,932 corresponding masks. The chosen architecture, U-Net, was trained using data augmentation strategies to improve its generalization. The results demonstrated the viability of the proposed approach for bladder segmentation, with a Dice Coefficient of 81%. Furthermore, the development of an application integrated with the pretrained model that could provide a practical and accessible tool for Radiotherapy specialists. The qualitative analysis of the segmentations, reinforced by visual examples, highlights the model’s accuracy in locating and contouring the bladder anatomy in Cone Beam - CT images.

The structure and function of the cornea are significantly influenced by its biomechanical characteristics, and abnormalities in this property can result in a number of pathological conditions, including keratoconus. A new generation of diagnostic and therapeutic approaches has been created as a result of significant developments in the understanding and analysis of corneal biomechanics in recent years. In this paper, we give an overview of recent studies on corneal biomechanics, including biomimetic eye modeling, deep neuromuscular oculomotor control, AI-based diagnostic models, and ultrasound elasticity imaging. We also discuss the potential application of neural networks and genetic algorithms for precise intraocular pressure prediction, as well as microanatomy research for better understanding of corneal structure. We evaluated the effectiveness of various computational analyses of corneal biomechanics for the diagnosis of keratoconus and presented a potential screening index for corneal biomechanics in both healthy and keratoconus patients. The significance of corneal biomechanics in prognosis studies for patients with corneal external mechanical stress mode is covered in the final section. Our review focuses on the promising developments in corneal biomechanics analysis and their potential to enhance the identification and management of corneal pathologies.

This study presents an innovative project for the quantification and morphological characterization of PC12 cell cultures, offering a comparative analysis of two methods: ImageJ software and a newly developed Python algorithm. Utilizing images captured at specific intervals on days 1, 3, and 5, both methods involved preprocessing steps such as brightness and contrast adjustment, segmentation, and morphological operations, followed by particle analysis for quantification. The results showed no significant differences between the methods in measuring the area occupied by the cells and cluster sizes on the same day. However, both methods revealed statistically significant increases in cell proliferation between days. The Python algorithm, built with libraries such as OpenCV and Numpy, offers additional advantages, including automation, a graphical user interface (GUI) created in Visual Studio Code and easy data export to Excel. These features make it a feasible and effective tool for real-time cell analysis, complementing and potentially enhancing traditional methods. The study concludes that the proposed system is as accurate and efficient as ImageJ, with added operational flexibility, providing a robust alternative for quantitative cell culture analysis. This advancement holds significant promise for applications in tissue engineering and regenerative medicine.

Image segmentation is an extremely important process in the medical context. If the method used not only addresses segmentation but also allows for the generation of interpretations expressed in natural language, it can lead to significant contributions to the study and resolution of certain medical problems. This paper presents a method for interpreting magnetic resonance images (MRIs) based on the automatic generation of interval-valued membership functions (IVMFs) and their interpretation through specific measures applied to these functions. The analysis conducted in previous work is substantially expanded. A set of new measures on the IVMFs is proposed, allowing for the interpretation of MRI images in terms of pixel intensities across different sequences. An exhaustive analysis is carried out on brain MRIs in T1, T2, and PD sequences. Global transformations on the images are considered, including contrast stretching, brightness increase and decrease, histogram equalization, and noise addition. It is demonstrated that the generated IVMFs and the proposed measures reflect and explain the expected changes. It is concluded that the proposed method, based on fuzzy logic, is suitable for the interpretation of MRIs, and its extension to other types of images or its application to other types of data is proposed as future work.

This study explores the predictive power of genetic signatures in forecasting patient responses to checkpoint inhibitor immunotherapy. By conducting an evaluative analysis of various genetic markers cited in the literature, this research aims to identify which patients are likely to benefit from immunotherapy. Despite the transformative potential of immunotherapy, a significant proportion of patients fail to respond favorably, underscoring the urgent need for precise predictive tools. Utilizing advanced machine learning techniques, the study generates predictive models that analyze accuracy and area under the curve (AUC) metrics to determine the most effective genetic signatures. These findings enhance our understanding of gene behavior in response to treatment and improve the selection process for suitable candidates for immunotherapy, thereby optimizing treatment outcomes.

Alzheimer’s disease, though not curable, can be effectively managed, particularly when identified early. Timely intervention significantly influences its progression, enhancing the patient’s well-being and relieving caregiver burden. The Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset offers valuable information for machine learning analysis. However, incomplete data is a common challenge due to the high cost of tests like electroencephalograms and magnetic resonance imaging. Our proposal involves employing machine learning techniques, including ensemble methods and ADA Boost, to analyze the ADNIMERGE dataset, which includes neuropsychological assessment scores. These algorithms will address incomplete data and balance the dataset by generating synthetic observations, aiming to enhance model performance in detecting Alzheimer’s disease stages. The ensemble approach combines multiple models to improve predictive accuracy, while ADA Boost trains weak learners iteratively to enhance overall model precision. Our goal is to develop a cost-effective tool for Alzheimer’s disease detection using only neuropsychological assessments.

Human-Robot Interaction (HRI) using biological signals has garnered research interest due to its potential to assist people with motor disabilities, such as in controlling wheelchairs, speech aids, or interacting with a computer. These systems involve acquiring physiological signals generated by the user’s intentions, which are then processed by a computational algorithm. In this study, we developed a method to interact with a mobile manipulator robot in a controlled virtual environment, allowing it to pick up and move objects using eye movements acquired through Electrooculography (EOG) and classified by an eight-class Convolutional Neural Network (CNN) algorithm. A Graphical User Interface (GUI) was designed for interacting with the robot, enabling the user to control various commands through eye movements, with visual feedback provided for the robot’s workspace. The algorithm achieved 92% accuracy in classifying eight eye movements, and during the experiment, the user successfully interacted with the robot and located an object of interest within its workspace in 19 s.

The objective of this work is to develop and present ThermalDiabetesTools, an innovative Python library tailored for managing thermal and RGB images of diabetic feet. It encompasses image loading, post-processing, segmentation, and registration, addressing the unique challenges in diabetic foot imaging. These include small temperature differentials and the need for precise temperature gradient analysis and RGB image processing. The tool is intended not only to compile existing methodologies but also to calibrate images specifically for diabetic foot applications. This specialized approach ensures more accurate and efficient processing, essential for both research and clinical purposes. In the clinical domain, ThermalDiabetesTools facilitates the early detection of complications by providing detailed thermal analysis, crucial for preventive healthcare in diabetes management. From a research perspective, the library accelerates developments in diabetic foot care by simplifying initial yet critical image analysis steps. Its automation and standardization capabilities can significantly improve consistency in results, making it a valuable asset in scientific studies. Furthermore, ThermalDiabetesTools is engineered for seamless integration with various systems and workflows, ensuring its accessibility to a wide range of users. Its function interface bridges the gap between advanced image processing techniques and practical clinical applications. This accessibility is key to its wide adoption, thereby enhancing the overall quality of diabetic foot care. By offering a set of tools for specific applications, this platform contributes to both research and clinical practices in diabetes management.

Recent advancements in sports video analysis and computer vision techniques have significantly improved sports analysis, allowing for more in-depth insights into sports dynamics. Human motion analysis, the exact estimation of joint extension and acceleration, is crucial for evaluating professional athlete outcomes and improving overall athletic performance. This study aims to quantitatively assess the correlation between kinematic parameters and athletic performance in the standing long jump, by analyzing the motions of both professional and amateur athletes. We utilize Dynamic Time Warping (DTW) analysis to assess the similarities between a professional athlete and non-athletes during multiple jumping trials. Our results show that kinematic parameters significantly influence jump distance, with arm-swinging angles playing a substantial role in achieving longer jumps. However, knee angles exhibit less influence on attaining greater jumping distances. Despite technical challenges, such as markerless motion capture and low sampling frequency, our findings highlight the potential of DTW in measuring performance in the standing long jump, offering valuable insights for sports training and teaching methodologies.

This study explores the use of foot ultrasound technique for assessing the recognition of regions of interest (ROI’s) in the foot, so it can help streamline processes for detecting abnormal patterns, specifically in diabetic foot using RSWE elastography. It focuses on the heel’s microchamber and calcaneus zones as key ROI’s. Traditionally, ROI detection relied on conventional algorithms; however, this paper introduces a novel approach using deep learning (DL) for automatic ROI detection in heel segmentation employing YOLOv8 algorithms, refined with data from prior research. A detailed comparative analysis was conducted between single-categorical and multicategorical models to classify the heel’s critical areas. Results showed that single-categorical models achieved an average precision of 98%, recall of 88%, mAP50 of 93%, and mAP50–95 of 65%. In contrast, multi-categorical models showed a balanced performance with 90% precision and recall, 92% mAP50, and 58% mAP50–95%. These findings underscore the potential of deploying DL techniques for diabetic foot assessment, marking a significant advancement in automated and precise medical diagnostics for early diabetes detection through foot condition analysis.

Electrosurgery equipment calibration is essential for maintaining safety and effectiveness in medical procedures, ensuring that devices function correctly and predictably. Accurate calibration minimizes risks and enhances patient outcomes by guaranteeing that electrosurgical tools operate within specified parameters. This study investigates the comparative performance of manual versus automated calibration techniques, focusing on two critical metrics: uncertainty and time efficiency. Through a comprehensive analysis, the results reveal that automated calibration techniques significantly outperform manual methods. Specifically, automated systems achieve a higher degree of precision and reliability, resulting in a narrower uncertainty interval. Additionally, the calibration process is completed approximately three times faster with automation. These advancements highlight the advantages of integrating automated calibration processes into clinical practice. By adopting automated methods, clinical engineering environments can experience improved accuracy, increased productivity, and enhanced safety, ultimately contributing to better patient care and more efficient operation of electrosurgical equipment.

Ancestral Andean medicine conceives the energy of the human body as a vital force that balances both the animate and inanimate elements of the cosmos represented by the 3 worlds: Hanan pacha, Kai pacha and Uku pacha. The measurement of this energy, known as “samai”, is performed by the wise men of the community, Yachaks, Healers, Midwives, among others. From biomedicine there are western methods for the measurement of energy, for example, electrodermal activity (EDA). The latter is an adaptable alternative to the approach of ancestral medicine because it can record the biological electrical potential from the 3 regions of the body (head, trunk, abdominal-pelvic region) corresponding to the worlds of the Andean cosmos, i.e., it integrates western biomedical technology from the Andean Cosmogony. This proposal proposes the design of an inclusive and culturally sensitive device for the holistic understanding of health, not only due to the high cost of support devices for measuring and recording energy, but also capable of protecting and revitalizing the cultural values and identity of Andean communities by incorporating this knowledge in the design, as it promotes the recognition and empowerment of these ancestral traditions.

This study addresses the challenges faced by visually impaired individuals who struggle with autonomous navigation due to limited obstacle detection at the head and torso levels. We propose a prototype that seamlessly integrates with eyeglasses to complement the white cane and provide enhanced safety and mobility. Our approach involved a structured methodology including requirements analysis, conceptual design, prototype development, and testing. Results show that our prototype effectively detects obstacles at varying distances and provides real-time alerts via vibration. This innovation promises to improve the independence and safety of the visually impaired by overcoming the limitations of existing devices and providing a cost-effective solution.

Through various studies, it has been confirmed that sitting in one position for prolonged periods of time can cause a variety of health problems, from lower back discomfort to cardiovascular disease. In addition, people with spinal cord injuries are significantly prone to develop pressure ulcers, which can have fatal consequences. This article proposes the implementation of sensing-mat technology to prevent these conditions by redistributing pressure points while sitting. It also includes an analysis of pressure points in different postures, revealing significantly high levels of pressure, with up to three critical points detected in the buttocks and adjacent areas.

Epilepsy is a brain disorder characterized by recurrent episodes of abnormal brain activity, known as epileptic seizures, and this condition can have a significant impact on patients’ quality of life, as well as their emotional, social, and physical well-being. This study investigates the development and evaluation of a signal processing-based algorithm for detecting tonic-clonic epileptic seizures using a smartphone accelerometer. The algorithm was implemented using Python and evaluated using a dataset available on IEEE DataPort. Data were analyzed in the frequency domain, calculating energies for different types of normal and epileptic activities. The results indicate that the developed algorithm is capable of distinguishing between normal and epileptic states, with a precision of 42.8%, an accuracy of 88.2%, and a specificity of 87.1%. This approach provides an effective tool for detecting epileptic episodes, thereby improving clinical management and patients’ quality of life.

Effective medical equipment management is crucial for delivering high-quality healthcare services, especially in resource-limited settings. In Peruvian hospitals, existing inventory systems are inadequately equipped and lack the necessary tools to guarantee accurate and reliable management of medical equipment inventories. This paper introduces and evaluates the usability of an RFID-based system specifically designed to manage medical equipment and its accessories. The proposed system involves the development of an RFID reader integrated with a user-friendly web interface. This setup facilitates the registration and tracking of medical equipment and ensures the availability and management of associated accessories. This system advances medical equipment management practices by focusing on accessory management. Initial usability studies suggest that the system significantly enhances the accuracy and efficiency of inventory management. However, comprehensive evaluations are needed to assess its effectiveness across diverse healthcare environments in Latin America. This helps generalize the system’s benefits and optimize its components for broader application.

Livestock production faces urgent challenges, including efficiency, sustainability, and animal welfare. In this context, remote animal monitoring is a crucial tool for sustainable livestock management. This paper presents the design and development of an Internet of Things (IoT) based biotelemetric device for animal welfare monitoring in ruminants. The device measures physiological variables such as body temperature, heart rate, and motor activity, in addition to the animal’s location. The collected data is transmitted wirelessly to a database and displayed on a mobile or web application, allowing users to monitor animal status in real time. The device meets the established minimum requirements and demonstrates its feasibility for collecting data relevant to animal welfare. Successful tests have been conducted with accelerometer, body temperature, and heart rate sensors, validating their functionality. The feed plate and device housing design was optimized to improve its performance and adaptation to the ruminant’s anatomy.

This work is about an innovation project that offers a solution for monitoring patients’ health with COVID-19 in home isolation. The project has been developed under the Agile approach of the Project Management Institute (PMI), also using the tool for continuous improvement DMAIC (Define, Measure, Analyze, Improve, and Control). In addition, a control panel will also be made to monitor the global quality of service (QoS) metrics; a financial analysis based on the theory of constraints (TOC) was implemented. The presented project seeks to solve the growing demand for health services triggered by the COVID-19 pandemic. We propose a monitoring system based on wireless body area networks (WBAN).

The summer school is an international high-level academic training program that addresses the need to educate new generations in the design, use, and adaptation of emerging technologies for the social inclusion of the current elderly population in our countries, promoting their autonomy and healthy aging. It was established in 2017 at Pontificia Universidad Javeriana in Bogota, Colombia, in collaboration with IFMBE, CORAL, ABIOIN, and IEEE/EMBS, in Biomedical Engineering education with a high impact on undergraduate and postgraduate students across 16 countries, including Argentina, Bolivia, Brazil, Canada, Chile, Colombia, Costa Rica, Cuba, Ecuador, El Salvador, Honduras, Guatemala, Mexico, Nicaragua, Peru, and Panama. Over 300 students have participated in the program, focusing on leveraging emerging technologies to enhance elderly well-being. The participants have the opportunity to interact with the elderly in combination with a strategic pedagogical such as lectures, workshops, case analyses, visiting living labs, and networking, and the main activity is to carry out a Hackathon Challenge based on the elderly needs in a real study case from a multidisciplinary perspective, in a multicultural environment to develop technical and soft skills that allow the students to explore different challenges to develop innovative technical and ergonomic solutions, by integrating emerging technologies and addressing social needs. A multidisciplinary distinguished professor mentors this learning model on the innovation phases of conception, design, prototyping, validation with the user, and business model, considering the standards, ethical, and commercialization process to propose affordable solutions. Finally, the students present a pitch to an expert panel for evaluation and awards.

The application of telemedicine represents a promising paradigm for enhancing healthcare services in remote and inadequately served areas, with the potential to bring significant benefits. In Ecuador, rural communities situated along the Cayapas River face significant challenges pertaining to the provision of healthcare services. These challenges are attributed to two key factors: geographical isolation and the lack of medical specialists in the region. This study evaluates the efficacy of a telemedicine platform implemented on Amazon Web Services (AWS) in improving healthcare accessibility in these rural areas. The study was conducted over a period of two years, between January 2020 and September 2022. The use of a telemedicine platform on AWS EC2 with a free software stack enabled the delivery of teleconsultations via a responsive Joomla template. The vulnerability analysis was based on data security standards, specifically the National Institute of Standards and Technology (NIST) 800-115 standard. The telemedicine platform demonstrated potential for enhancing healthcare delivery, connecting rural doctors with specialists, and improving diagnostic accuracy. However, the platform’s adoption was limited, with a decreasing number of teleconsultations in the previous year. The main challenges were identified as intermittent Internet connectivity, technological barriers, and the lack of willingness to adopt new technologies. Despite the promise of telemedicine, it is imperative to address the technological and training challenges that must be overcome to facilitate its widespread adoption and sustainability in rural areas of Ecuador.