e22576 Background: Early diagnosis is vital for cancer prognosis, yet high-level screening evidence (Grade A/B) currently exists only for breast, cervical, lung, and colorectal cancers. Liquid biopsy has emerged as a promising alternative, but gaps remain in prediction accuracy and Tumor of Origin (TOO) identification. While ML/AI is established in radiology, its role in liquid biopsy remains under-recognized. Integrating ML/AI with multi-omics data offers significant potential to elevate positive predictive values and bridge the gap between technological potential and clinical application. Methods: We systematically identified over 50 commercial and developmental cancer screening kits via technical files, peer-reviewed publications, and clinical reports. Test features were extracted to evaluate diagnostic performance. We categorized underlying ML/AI algorithms based on their architectural framework, data integration capacity, and clinical validation metrics. Results: No multi-cancer early detection (MCED) kits have received full regulatory approval; most are currently accessible only as Laboratory Developed Tests (LDTs). Evaluation reveals that market-leading products still rely on preliminary frameworks, such as Logistic Regression (LR) or Random Forests (RF). While robust, these models lack the complex feature extraction capabilities of deep learning. A key finding is the prevailing technical conservatism in the diagnostic sector. Mature products prioritize interpretability and robustness over complex "black-box" models, as regulatory bodies (e.g., FDA, NMPA) favor transparent decision boundaries and predictable performance. Consequently, the primary hurdle is developing advanced AI that meets these rigorous standards for clinical transparency. Conclusions: The diagnostic sector remains a technologically conservative niche, lagging in cutting-edge AI integration. Bridging this gap requires deep synergy between biomedical researchers, kit developers, and algorithm engineers. Transitioning from traditional models to advanced, yet interpretable, AI is essential to unlock the full potential of liquid biopsy for next-generation cancer screening. List of selected cancer screening kits and their algorithms. Dx Country Target Algorithm Galleri US Methylation Penalized Multinomial Logistic Regression Shield US Methylation + Fragmentomics + DNA Mutation LR Freenome US Multi-omics LR / SVM / Ensemble Delfi US Fragmentomics LR / LASSO / PCA OverC CN Methylation LR / SVM PanSeer CN Methylation LR / LASSO Mercury CN Fragmentomics XGBoost / GLM / GBM / RF Genecast CN Methylation, Fragmentomics KNN / SVM / LR / GBDT / RF GeneSolutions SG Methylation LR / LASSO
Qi Lei (Thu,) studied this question.