1. Research Questions & Objectives
Abdominal tumors (e.g., hepatocellular carcinoma) cause high mortality worldwide. Non-invasive stratification of tumor biology, body composition, and immune status using imaging, multi-omics, and clinical data remains challenging and benefits from large cohorts like UK!Biobank.
1.1 Primary Question
Can an AI-driven radiomics signature from abdominal CT/MRI-integrated with genetics, proteomics, metabolomics, body composition, lifestyle, and clinical data-detect and predict prognosis of abdominal tumors more effectively than standard clinical measures?
1.3 Objectives
1.Build and validate an AI radiomics pipeline to extract CT/MRI features.
2.Link imaging features with germline genetics (e.g., polygenic risk).
3.Combine imaging with proteomic, metabolomic, and clinical markers for diagnostic/prognostic evaluation.
4.Analyze body composition (VAT/SAT, muscle index) in relation to tumor phenotype and outcomes.
5.Assess lifestyle and medical history effects on radiogenomic signatures.
6.Develop AI models for tumor detection, treatment response, and survival prediction.
7.Compare multimodal models against clinical-only models to measure added value.
2. Scientific Rationale
Radiomics quantifies imaging features from CT/MRI to detect tumor heterogeneity beyond visual assessment, improving HCC grading, staging, and prognosis. Radiogenomics demonstrates imaging signatures can non-invasively predict genetic, transcriptomic, proteomic, metabolomic, and immune characteristics. Body composition metrics-like visceral fat and muscle mass-reflect metabolic health and inflammation, influencing tumor progression. The UK!Biobank provides extensive imaging, multi-omics, lifestyle, and outcome data, ideal for AI-driven integrative research. Our approach promises to improve non-invasive tumor profiling, support personalized treatment, reduce invasive biopsies, and advance precision oncology.
