Pneumonia Detection Using Asynchronous Split Learning Method

Use of deep learning in medical healthcare is a rapidly developing sector. Training a deep learning model requires huge amount of data. As individual Hospital does not have substantial medical data, Split Learning Scheme is used where multiple hospitals share their sensitive medical data securely and train a single model in the server. In such scenario effective communication between the hospitals and the server is extremely necessary. The work done in this paper uses asynchronous split learning scheme to solve challenges caused by unreliable connectivity guaranteeing that the learning process continues even when network connections fail. Asynchronous split learning minimizes data transmissions across networks, optimizing bandwidth usage. This, in turn, facilitates seamless operation for consumer electronics and medical wearable devices in varied connectivity conditions. To validate the robustness and effectiveness of our proposed work, chest X-ray data set is used to identify pneumonia by comparing the result using traditional Convolution Neural Networks and Split Learning algorithms.