Kann KI das individuelle Rückfallrisiko von Krebs mithilfe der genetischen Sequenzierung von Tumoren vorhersagen ?

Ein Krebsrückfall hängt von einem komplexen Zusammenspiel von genetischen Mutationen, dem Tumor-Mikromilieu und der Therapieantwort ab. Die personalisierte Medizin zielt darauf ab, das Rückfallrisiko vorherzusagen, indem sie die Tumor-Genomik analysiert, doch die Integration riesiger Datensätze bleibt für menschliche Kliniker eine Herausforderung. KI könnte diesen Prozess beschleunigen, indem sie Muster identifiziert, die mit dem Wiederauftreten in hochdimensionalen Daten verbunden sind.

Background

Cancer relapse is shaped by interactions among somatic mutations, the tumor microenvironment, systemic immunity, and therapeutic selection pressures. Personalized oncology seeks to quantify recurrence risk from tumor genomics, but integrating high-dimensional genomic, epigenomic, transcriptomic, and clinical data within a single workflow remains non-trivial for human interpreters.

AI-driven pipelines now fuse whole-exome or whole-transcriptome tumor sequencing with clinical covariates to generate individualized recurrence-risk estimates. Commercial gene-expression assays such as Oncotype DX AR-V7 (prostate cancer) and FoundationOne Hemo (hematologic malignancies) and the breast-cancer panel Oncotype DX Breast Recurrence Score have received regulatory clearance and provide prognostic signatures correlated with distant recurrence and survival endpoints. Deep-learning models trained on TCGA cohorts report AUCs of ≈0.75–0.85 for predicting relapse across several tumor types, outperforming traditional histopathology-based staging in validation splits. Regulatory-cleared tools are currently labeled for prognosis (i.e., outcome prediction) rather than therapy selection (predictive use), and their performance in non-academic, multi-institution cohorts is still being evaluated. Reference: Nature Medicine, enriched May 12 2026.