Photoplethysmography Bio-Signal Extraction for Classifying Diabetes Mellitus Diseases Using Pretrained Deep Learning Networks

• Published in: SN computer science Vol. 6; no. 2; p. 104
• Main Authors: Tarmizi, Syaidatus Syahira Ahmad, Suriani, Nor Surayahani, Mohd, Mohd Norzali Bin Hj, Bagchi, Susama, Debnath, Sanjoy Kumar, Shah, Syed Mehr Ali
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 01.02.2025; Springer Nature B.V
• Subjects: Accuracy; Aging; Algorithms; Biomarkers; Blood pressure; Channel morphology; Classification; Computer Imaging; Computer Science; Computer Systems Organization and Communication Networks; Data Structures and Information Theory; Deep learning; Diabetes; Diabetes mellitus; Disease; Glucose; Heart rate; Hypertension; Information Systems and Communication Service; Metabolic disorders; Morphology; Networks; Original Research; Oxygen saturation; Pattern Recognition and Graphics; Physiology; Respiration; Sensors; Signal processing; Skin; Software Engineering/Programming and Operating Systems; Trends; Unleashing the Advances in ICT for Digital Transformation through Data Engineering and Data Analytics; Vision; Blood glucose; Photoplethysmography; Blood pressure; Diabetes mellitus
• ISSN: 2661-8907; 2662-995X; 2661-8907
• DOI: 10.1007/s42979-024-03589-4

This study investigates the association between the presence of blood glucose and blood pressure variations to determine an individual’s health. It classified diabetes mellitus by using a publicly available PPG-BP dataset of 72 subjects. Increasing diabetes trends due to undiagnosed disease were overcome through continuous finger cuff-based blood pressure measurements via photoplethysmography to extract diabetes related features including the baseline anthropometrical characteristics and signal morphology. The variations in blood pressure obtained from single channel PPG are processed to extract signal information and the results which are the STFT transformed version of the signal as the input data to investigate the meaningful patterns of the original raw, filtered and their derivatives between healthy and diabetic individuals. This can be followed by the state of the art of pretrained deep learning networks such as DenseNet121, DenseNet169, EfficientNetB0, EfficientNetB1, InceptionResNetV2, ResNet50, ResNet101 and Xception which were discussed in the comparative analysis that can be effectively used for classification on the respective sets of signals. The result indicates that InceptionResNetV2 and Xception networks outperform the rest of the pretrained deep learning used in the study with the highest recorded accuracy, precision and recall values which making them the most effective networks for classification problems.