Efficient and Private ECG Classification on the Edge Using a Modified Split Learning Mechanism

Cardiovascular diseases are the number one cause of death, comparable only to cancer in high-income countries. Electrocardiography (ECG) is an important non-invasive diagnostic technique for evaluating a patient's cardiac clinical status. Because of the large number of ECGs that are routinely taken, advanced decision support systems based on automatic ECG interpretation algorithms promise significant assistance to medical staff. This paper proposes a modified distributed machine learning system that builds on top of a technique called split learning. The modified system aims at reducing the communications and computation overhead of split learning achieving a private and secure ECG classification that is possible to implement on edge hardware. The modified SL system is tested on the PTB-XL data set using hardware representing edge devices. The results showed a significant reduction in the computation and communication overhead with minimal loss of performance.