Research Article
Mobile Application for Remote Monitoring of Peripheral Edema
@INPROCEEDINGS{10.1007/978-3-031-59717-6_18, author={Aaron John Bernante and Khristine Joie Recto and Jhoanna Rhodette Pedrasa}, title={Mobile Application for Remote Monitoring of Peripheral Edema}, proceedings={Pervasive Computing Technologies for Healthcare. 17th EAI International Conference, PervasiveHealth 2023, Malm\o{}, Sweden, November 27-29, 2023, Proceedings}, proceedings_a={PERVASIVEHEALTH}, year={2024}, month={6}, keywords={Peripheral edema MobileNetV3 mHealth Convolutional neural networks Deep learning}, doi={10.1007/978-3-031-59717-6_18} }- Aaron John Bernante
Khristine Joie Recto
Jhoanna Rhodette Pedrasa
Year: 2024
Mobile Application for Remote Monitoring of Peripheral Edema
PERVASIVEHEALTH
Springer
DOI: 10.1007/978-3-031-59717-6_18
Abstract
The prevalence of heart disease continues to become a relevant issue globally. One of the most common symptoms in heart failure (HF) patients is peripheral edema. Peripheral edema can be caused by various underlying conditions. Thus, early detection and consistent monitoring are vital for its appropriate treatment. Several studies have explored the use of telehealth for a more accessible remote monitoring of HF patients. With the current gap in monitoring patients remotely, the proposed solution is a mobile application that detects the presence and severity of peripheral edema in HF patients. It allows patients to take a video of their extremities and a deep learning model in the application will evaluate the presence and severity of peripheral edema. The dataset collected consists of 150 photos for each edema stage. Transfer learning was utilized on a MobileNetV3 model with pre-trained weights from ImageNet. The model yielded an accuracy of 95.24% and recall of 0.96 on the test dataset, and an accuracy of 86.67% and recall of 0.95 during the field testing of the application. The high accuracy indicates that the model performs well in classifying different peripheral edema severity. Moreover, the high recall value shows that the model is able to accurately detect the presence of edema by minimizing false negatives.
