Meehan A, Zhang Z, Williams B, Jiang R. An Investigation on ECG-based Cardiological Diagnosis via Deep Learning Models. InRecent Advances in AI-enabled Automated Medical Diagnosis 2022 Oct 20 (pp. 304-316). CRC Press.
Abstract
This paper compares three of the most popular and suitable types of neural networks for identifying abnormalities in ECG data. The ECG data set used was carefully and meaningfully pre-processed and aligned with the time dimension between samples. A 2-layer Convolutional Neural Network performed best across 25 train-test trials, achieving an accuracy of 99.2%. A 3-layer MLP also performed well, achieving an accuracy of 98.8%, but the LSTM network couldn’t match either’s performance with an accuracy of 97.9%. Further studies could consider more realistic and complex ECG data, as a continuous-time series, with live simultaneous inputs from several different sensors. Similarly, more sophisticated multi-class classification could be applied for different types of arrhythmias. These datasets are likely to require deeper and more sophisticated neural networks.
