Hybrid Deep Learning Approach for AI Medical Device Post-Market Surveillance: Combining CNN and LSTM for Image and Time-Series Data Analysis

Abstract

Artificial intelligence-based medical devices are transforming healthcare by delivering precise diagnoses, real-time monitoring, and improved patient care. But their safety, efficacy, and compliance with regulations after they are made available in the market continue to pose a challenge. Conventional post-market surveillance practices, mainly based on sporadic assessments and clinical trials, are typically time-consuming, expensive, and ineffective in offering real-time information about device performance. This work introduces a Hybrid Deep Learning Approach combining CNN for medical image processing and LSTM networks for time-series analysis and presents a scalable and efficient solution for AI medical device post-market surveillance. With these two powerful models, the proposed solution will be used to guarantee the performance of devices in terms of capturing images and monitoring patient health over a period of time. The model has been tested against two vital datasets, the Heart Disease Dataset and ISIC where it is spectacular in tracking the performance of the monitoring devices and also the health condition of the patients. The model achieved 99 percent in overall performance evaluation of the device, excellent precision, recall, and F1-scores of the Skin Cancer Dataset. The hybrid model elaborated upon is deemed superior to traditional approaches like SaMD + PMCF in terms of mitigating risks, following up in clinics, and fusing disparate data. Moreover, it highlights the need for secure processing of the data and the implementation of cloud security systems for patient confidentiality and regulatory compliance. The findings thus illustrate the potential enhancement of effectiveness, scalability, and reliability of post-market surveillance of AI medical devices through the hybrid deep learning approach.