Traditional precision management of gastric and colorectal cancers largely relies on static, single-point models, which struggle to capture spatial heterogeneity and clonal evolution.

In a study published in LabMed Discovery, the researchers propose a new ‘spatio-temporal integration, adaptive, closed-loop’ paradigm, centred on three dimensions: the spatial dimension integrates multi-omics and multi-modality imaging to construct a comprehensive tumour atlas; the temporal dimension utilises longitudinal monitoring via liquid biopsy, such as ctDNA, to dynamically track clonal evolution and achieve functional stratification of biomarkers; AI-driven fusion, meanwhile, provides decision support at three key stages: T1 prediction, T2 adaptation and T3 interception.

The article classifies biomarkers into four categories: early screening, prognostic stratification, treatment response prediction and recurrence monitoring, emphasising that their performance and clinical applicability are cancer-type and scenario-specific, and should not be simply extrapolated across gastrointestinal tumours.

The article also provides an objective evaluation of key studies such as DYNAMIC-III: ctDNA-guided downstaging of adjuvant therapy can reduce oxaliplatin exposure and toxicity; however, as the pre-specified non-inferiority criteria were not met, its clinical utility must still be interpreted with caution.

Finally, the authors propose strategies to address translational barriers such as testing standardisation, cost, clinical validation and the generalisation capability of AI, whilst outlining future directions including standardised regulation, multimodal AI and digital twins, spatial multi-omics clinical translation, and adaptive trial design.

Source: Shanghai Jiao Tong University Journal Center