AI-POWERED MEDICAL IMAGING FOR PRIVACY-PRESERVING EARLY CANCER DIAGNOSIS AND SECURE INTEGRATION INTO U.S. HEALTHCARE SYSTEMS

Abstract

This study evaluated the effectiveness of AI-powered medical imaging for privacy-preserving early cancer diagnosis and secure integration into U.S. healthcare systems using a federated, multi-institutional dataset of 68,425 imaging examinations. The analysis demonstrated that AI augmentation substantially improved diagnostic outcomes, with mixed-effects logistic regression revealing an 84% increase in odds of early cancer detection compared to standard interpretation alone. AI confidence scores exhibited strong alignment with confirmed early-stage malignancies (correlation r = 0.72), and radiologist agreement improved significantly when supported by AI-generated probability maps and lesion annotations. Workflow evaluation indicated that AI integration reduced median time-to-report by 6.42 minutes, despite a modest increase of 1.17 clicks per case associated with AI-panel interactions. Privacy and security assessments showed that encryption coverage averaged 89.6%, and higher encryption levels significantly lowered privacy leakage scores (β = –0.031). Differential privacy mechanisms further reduced membership inference risk (β = –0.22) without producing measurable declines in diagnostic accuracy. Interoperability testing confirmed that AI outputs achieved high integration compliance across heterogeneous electronic health record and PACS environments, with consistent data throughput and minimal system latency. Reliability and validity analyses demonstrated strong internal consistency for diagnostic indices (Cronbach’s α = 0.91) and high temporal stability across federated rounds. While institutional variability in imaging quality and security maturity introduced some limitations, the overall findings indicate that AI-powered, privacy-preserving imaging systems can enhance early cancer detection, strengthen security protections, and improve workflow efficiency in diverse U.S. healthcare settings. These results support the scalable deployment of AI-driven diagnostic infrastructure capable of maintaining both clinical accuracy and stringent privacy safeguards.

Ähnliche Arbeiten