Self-Supervised Learning Featuring Small-Scale Image Dataset for Treatable Retinal Diseases Classification

Automated medical diagnosis through image-based neural networks has increased in popularity and matured over years. Nevertheless, it is confined by the scarcity of medical images and the expensive labor annotation costs. Self-Supervised Learning (SSL) is an good alternative to Transfer Learning (TL) and is suitable for imbalanced image datasets. In this study, we assess four pretrained SSL models and two TL models in treatable retinal diseases classification using small-scale Optical Coherence Tomography (OCT) images ranging from 125 to 4000 with balanced or imbalanced distribution for training. The proposed SSL model achieves the state-of-art accuracy of 98.84% using only 4,000 training images. Our results suggest the SSL models provide superior performance under both the balanced and imbalanced training scenarios. The SSL model with MoCo-v2 scheme has consistent good performance under the imbalanced scenario and, especially, surpasses the other models when the training set is less than 500 images.

A Greedy-optimized Framework for Heart Rate Variability Monitoring during Daily Activities using Wearable Photoplethysmography

Continuous monitoring of inter-beat-interval (IBI) and heart rate variability (HRV) provides insights in cardiovascular, neurological, and mental health. Photoplethysmography (PPG) from wearables assures convenient measurement of IBI. However, PPG is susceptible to motion artifacts, considerably deteriorating the accuracy of IBIs estimation. Although a multi-channel model in previous study improves accuracy, prevailing compact commercial wearables would favor single-channel sensors, causing benefits of multi-channel applications to have restrictions. In this paper, a greedy-optimized framework is proposed for measurement of IBI and HRV featuring single-channel and multi-channel PPG signals collected during daily activities. Utilizing the fact of continuity in heartbeats, the IBI estimation problem is converted into the shortest path problem in a directed acyclic graph, where candidate heartbeats from the noisy PPG are regarded as vertices. The framework exploits a convex penalty function to optimize weight assignment in the shortest path calculation and a greedy-optimized fusion method to mitigate overly fluctuating patterns in estimated IBIs. The results achieve correlation of 0.96 with percentage error of 3.2% for IBI estimation using single-channel PPG signals from the 2015 IEEE Signal Processing Cup dataset, where percentage error is reduced by 58.4% and correlation is improved by 11.6% in comparison to those without greedy-optimized fusion. In the multi-channel model, it achieves correlation of 0.98 with percentage error of 2.2%. Estimated and true HRV parameters are also highly correlated with low percentage errors. This paper further validates these techniques on the PPG-DaLiA dataset, indicating the robustness of the proposed framework.