CURRENT ISSUE

ABSTRACT: Cancer behaves as a dynamic evolutionary system, whereas conventional tissue biopsy and imaging provide incomplete temporal information about how a tumour is changing. This review critically examines whether longitudinal liquid-biopsy measurement, combined with artificial intelligence (AI), can move oncology from tumour monitoring toward genuine prediction of cancer evolution and treatment resistance. We synthesise mechanistic, clinical, and computational evidence on circulating tumour DNA, circulating tumour cells, extracellular vesicles, and multimodal liquid-biopsy platforms, together with AI approaches applied to longitudinal cancer data, drawing on landmark and recent studies across colorectal, lung, breast, and ovarian cancer. Current evidence distinguishes three increasingly demanding tasks: early detection, longitudinal monitoring, and genuine forecasting. Longitudinal circulating tumour DNA analysis can detect molecular progression and emerging resistance mutations months before radiographic progression, and machine-learning models can stratify survival using serial biomarker metrics; however, most published "AI prediction" studies demonstrate earlier detection or prognostic association rather than prospectively validated forecasting of a future, patient-specific event. Biological variability in tumour shedding, analytical noise, small and irregularly sampled longitudinal datasets, overfitting, data leakage, weak external validation, and limited interpretability constrain the transition from detection to actionable prediction. AI-enabled liquid biopsy has genuine potential to forecast tumour trajectories, but current evidence is considerably stronger for molecular monitoring and early detection than for prospectively validated, clinically actionable prediction. Realising this potential will require standardised assays, prospective longitudinal cohorts with predefined prediction horizons, rigorous external and temporal validation, calibrated and interpretable models, and interventional trials linking predictions to treatment decisions before AI-enabled liquid biopsy can be considered ready to guide clinical care.

ABSTRACT: Environmental carcinogens remain significant contributors to cancer development; however, conventional risk assessment has largely focused on population-level exposure without adequately accounting for variation in individual susceptibility. Individuals exposed to comparable environmental hazards can nevertheless face different cancer risks, as susceptibility is influenced by genetic and epigenetic variation, metabolic capacity, immune function, age, lifestyle, comorbidities, and cumulative exposure histories. The growing availability of environmental, clinical, molecular, and longitudinal health data offers new opportunities to characterize these differences with greater precision. In this context, artificial intelligence (AI), particularly machine learning and other multimodal analytical approaches, can identify complex patterns across heterogeneous datasets that may remain difficult to detect through conventional statistical methods. This narrative review explores the potential of AI to enable more precise assessment of individual susceptibility to environmental carcinogens and proposes a framework that brings together exposure profiles, host susceptibility, molecular responses, lifestyle and contextual factors, and longitudinal information. Within this framework, AI serves as an analytical layer for generating individualized risk profiles and informing prevention strategies, including exposure reduction, risk-informed surveillance, and targeted preventive interventions where clinically appropriate. However, several challenges may limit its translation into practice, including weaknesses in exposure assessment, data quality, algorithmic bias, limited interpretability, privacy concerns, uncertainty surrounding causal inference, insufficient external validation, and implementation barriers. Accordingly, AI should complement, not replace, established epidemiological and mechanistic approaches. With rigorous prospective validation, transparent model development, and appropriate ethical and clinical safeguards, AI-enabled susceptibility assessment may support a transition from predominantly population-based approaches to more individualized and risk-informed environmental cancer prevention.