<![CDATA[Medical artificial intelligence relies heavily on imaging for lesion detection and anatomical quantification, providing essential support for cancer diagnosis and treatment planning (General Background). While foundation models are rapidly transforming radiological workflows, traditional evaluations focus predominantly on technical segmentation metrics rather than clinical utility (Specific Background). Existing literature lacks a comprehensive synthesis combining model taxonomies, multi-task assessments beyond segmentation, and standardized clinical readiness appraisals (Knowledge Gap). This systematic review evaluated forty-two studies published between 2024 and 2026 across eleven databases using PRISMA 2020 guidelines and AI-specific quality frameworks (Aims). Results demonstrated high organ segmentation accuracy with Dice coefficients exceeding 0.90, whereas performance declined for infiltrative tumours, and external validation was present in only one-third of studies (Results). We propose an eight-level clinical-readiness scale and research roadmap to transition medical AI from technical benchmarks to deployment (Novelty). Standardized validation protocols are critical for advancing safe decision support (Implications). Key Findings Highlights Current medical foundation models achieve high technical precision in organ segmentation but show reduced performance in low-contrast and post-treatment lesions. Less than one-third of evaluated studies report external validation, and no interventional prospective trials currently exist. Transitioning AI to routine oncology care requires establishing standardized clinical endpoints, fairness reporting, and multi-institutional data frameworks. Keywords: Foundation Models, Medical Imaging, Tumour Quantification, Clinical Decision Support, Systematic Review]]>
Copyrights © 2027
