ARTIFICIAL INTELLIGENCE-ENHANCED LIQUID BIOPSY IN CANCER MONITORING: FROM TREATMENT RESPONSE TO EARLY DETECTION OF DISEASE RECURRENCE

Abstract

Liquid biopsy has emerged as an important minimally invasive technology in contemporary oncology, enabling repeated assessment of tumor-derived molecular information through blood and other biological fluids. Circulating tumor DNA (ctDNA), circulating tumor cells, cell-free DNA methylation patterns, extracellular vesicles, circulating RNA and fragmentomic signatures may provide dynamic information about tumor burden, treatment response, molecular residual disease and cancer recurrence. However, liquid biopsy generates complex, multidimensional datasets in which tumor-derived signals may be extremely small relative to biological and technical background. Artificial intelligence (AI), including machine learning and deep learning, offers a potential computational framework for integrating these heterogeneous signals and generating clinically meaningful predictions. This review examines the emerging convergence of AI and liquid biopsy, with particular emphasis on longitudinal cancer monitoring rather than cancer detection alone. The available PubMed-indexed literature was reviewed to evaluate the potential role of AI-enhanced liquid biopsy in treatment-response assessment, minimal residual disease detection, prediction of treatment resistance and early identification of disease recurrence. Evidence from colorectal, lung, breast, hepatocellular and other solid tumors is discussed. The review also addresses major translational barriers, including low tumor shedding, clonal hematopoiesis, analytical heterogeneity, external validation, algorithmic bias, interpretability, data privacy, cost and uncertainty regarding clinical actionability. Current evidence indicates that AI can extract complementary biological information from complex circulating biomarker datasets, but prospective validation is still required before widespread implementation. AI-enhanced liquid biopsy may ultimately support a transition from episodic, imaging-centered surveillance toward personalized, dynamic and molecularly informed cancer monitoring.