Explainable AI for personalised therapy in metastatic colorectal cancer
Colorectal cancer (CRC) is the second leading cause of cancer-related deaths in Europe, with over 150 000 new cases diagnosed annually. A significant proportion of these patients develop metastatic colorectal cancer (mCRC), which is significantly more difficult to treat. Despite advances in chemotherapy and molecular targeted therapies, treatment response varies greatly between individuals, highlighting an urgent need for more personalised and effective approaches.
AI to aid decision-making in the clinic
The EU-funded REVERT(opens in new window) project has taken a major step in that direction by developing an AI-driven platform that supports personalised treatment planning for patients with mCRC. “Our goal was to create an explainable AI system that clinicians can understand and trust,” emphasises project coordinator Fiorella Guadagni of IRCCS San Raffaele in Rome. The consortium incorporated multiple independent algorithms, thereby providing a consensus-driven recommendation that supports rather than replaces the doctors’ expertise. The REVERT’s decision support system (DSS) was tested in a prospective early-phase clinical trial(opens in new window) to assess its impact on survival and quality of life. The results were very encouraging with AI-guided therapy suggestions significantly improving progression-free survival in patients with mCRC. This has the potential to reduce unnecessary toxicity, improve quality of life, and make better use of healthcare resources.
Biomarker discovery
REVERT also focused on the identification of biomarkers for personalised therapy. The consortium analysed both conventional and novel biomarkers through retrospective studies and AI-driven stratification models. One notable outcome was the development of an algorithm(opens in new window) which uses mutation profiles of 7 genes including KRAS, BRAF, ERBB2, and TP53 to predict whether a patient will respond to first-line chemotherapy. The algorithm demonstrated statistically significant accuracy in stratifying therapy responders and non-responders. The consortium also established patient-derived organoids (PDOs) from patient tumour samples to assess how different tumours respond to standard and novel therapies. Transcriptomic and miRNA analyses revealed distinct molecular signatures that could inform future treatments, particularly those targeting the KRAS-MAPK signalling pathway. A multi-omics analysis further indicated that genes involved in cell extracellular matrix interactions and proteoglycan-related pathways may play a role in treatment resistance. These insights are helping lay the groundwork for more accurate, biology-driven AI models.
Ethical design
A defining achievement of REVERT is that its AI system was designed in strict alignment with EU regulatory guidelines and the ethics guidelines of the European high-level expert group(opens in new window) for trustworthy AI. This includes transparency, fairness, safety, and most importantly, human oversight. “As doctors, we are bound by the Hippocratic principle of conscience. The REVERT system empowers doctors to remain in control, making conscious decisions supported by transparent, explainable models. This trust is key to the successful clinical implementation,” highlights Guadagni.
Future directions
Now that REVERT has demonstrated the potential of AI in real-world clinical trials, the next step is to translate these results into broader use. The project's reusable AI workflow can be adapted to other cancers or diseases, helping to build a new era of data-driven, patient-centred healthcare. Future studies will build on the REVERT dataset, AI models, and organoid systems to refine treatment strategies and further validate outcomes. With its impact on both clinical practice and digital health policy, REVERT stands as a model for how AI and ethical innovation can converge to improve personalised cancer care across Europe.
Keywords
REVERT, colorectal cancer (CRC), metastatic colorectal cancer (mCRC), algorithm, biomarker, explainable AI, mutation
