FedCarepod: Federated Quantum Transfer Learning Approach in Screening Process for Early Detection of Liver Abnormalities

Abstract

The liver is the most frequently injured intra-abdominal organ that leads to complications. Visual interpretation depends on the expertise, and this justifies the need for automatic accelerated assistive prediction system. Classical and Quantum machine learning approaches are combined to leverage their benefits, and the feasibility of designing a novel system is explored. This approach successfully ends up with health care system for early detection and screening of liver diffused diseases to assist doctors and radiologists. The local model in each location uses quantum transfer learning and the learned neural parameters are transferred to the global model for update, and the aggregated parameters are used to update neural parameters by federated learning and communicated to local models.  This learning approach is implemented using PennyLane. The detection performance is 99% using public liver dataset, hospital dataset and real time data from patients with updated model parameters. Due to the low cost and noninvasive nature, B mode ultrasound scanner is used. Amplitude encoding and optimizer reduces training time compared to classical approach and the result shows that this distributed Quantum approach is a promising research area if suitable Encoding and Ansatz circuit design is identified with cutting edge technologies. The proposed design outperforms the existing state-of-the-art techniques.