AI-Enhanced Device Quality Assurance

Abstract

Medical device quality assurance operates at the boundary between pharmaceutical manufacturing rigour and the fabrication variability inherent to devices with physical, electronic, and software components. Traditional quality systems built on ISO 13485 and 21 CFR 820 specify structured procedures for process validation, design control, corrective and preventive action, and complaint handling, but the implementation of those procedures has relied heavily on manual inspection, statistical process control applied to periodic sampling, and deviation investigation initiated by observed out-of-specification events. AI-enhanced quality assurance introduces continuous process monitoring, computer vision inspection at inspection rates exceeding human capability, predictive failure detection before deviations occur, and automated complaint analysis that identifies emerging quality signals faster than traditional trending methods. This study presents the AI-Enhanced Quality Assurance Assessment Framework (AIQAAF), evaluating five AI-enhanced QA approaches -- computer vision automated inspection (CV-AI), process analytical technology with ML process modelling (PAT-ML), AI-driven CAPA and complaint analysis (AI-CCA), predictive equipment and process failure monitoring (AI-PEM), and AI-assisted design control and risk management (AI-DCR) -- across four device manufacturing contexts: high-volume disposable device manufacturing, active implantable device production, diagnostic imaging device assembly, and combination product manufacturing. Performance was scored using the QA Enhancement Effectiveness Score (QEES), a weighted composite of defect detection sensitivity (0.30), false alert management (0.20), regulatory compliance support (0.20), implementation complexity-inverse (0.15), and return on investment (0.15). Computer vision automated inspection achieved the highest QEES (0.884) through the best defect detection sensitivity (0.940) and strong regulatory compliance support (0.900). Predictive equipment monitoring ranked second (0.877) with the best return on investment (0.940). Process analytical technology with ML (0.871) achieved the best false alert management (0.920). AI-assisted CAPA/complaint analysis and AI design control demonstrated meaningful value with moderate implementation complexity.