An AI-based biomarker developed by Norwegian company DoMore Diagnostics could help oncologists determine which colorectal cancer patients are likely to benefit from chemotherapy after surgery. The technology analyses digital images of tumour tissue and predicts the risk of cancer recurrence. Clinical validation studies suggest that it can distinguish between patients with different risk profiles, potentially supporting more personalised treatment decisions and reducing unnecessary chemotherapy.
Colorectal cancer is among the most frequently diagnosed cancers in Europe and remains a major cause of cancer-related mortality. Approximately half of patients are diagnosed with stage II or III disease, for which surgical removal of the tumour is generally the primary treatment.
Many patients, particularly those with stage III cancer, subsequently receive adjuvant chemotherapy to reduce the risk of recurrence. However, not everyone benefits from this additional treatment, while chemotherapy can cause significant side effects. Identifying which patients are most likely to benefit remains a clinical challenge. Conventional risk assessment methods, including tumour staging and molecular testing, do not always provide sufficient information to guide individual treatment decisions.
AI analyses existing tumour samples
The European Innovation Council-funded Histotype Px project, coordinated by DoMore Diagnostics, focuses on commercialising an AI-based biomarker that extracts prognostic information from digitised tumour tissue slides. The Histotype Px Colorectal system uses deep learning to analyse tissue samples collected during surgery. According to the developers, it can generate results within minutes using existing digital pathology infrastructure, without requiring additional tissue samples or specialised laboratory equipment.
The technology is designed to identify patients at relatively low risk of recurrence and distinguish them from those who may benefit from additional chemotherapy. Project coordinator Andreas Berg Storsve says the system can provide prognostic information that conventional tests may fail to identify. The company estimates that improved treatment selection could potentially spare up to 250,000 patients worldwide each year from unnecessary chemotherapy, with substantial associated healthcare savings. These figures are projections rather than demonstrated outcomes of widespread clinical implementation.
Clinical studies assess predictive value
According to the project team, clinical validation studies have been completed in hospital settings across Europe, Japan and the United States. One Japanese study examined more than 900 patients who had tested negative for circulating tumour DNA (ctDNA), a biomarker increasingly investigated for detecting residual cancer after treatment. Although ctDNA testing can help identify patients at risk of recurrence, negative results do not necessarily exclude the presence of residual disease. In the Japanese cohort, Histotype Px classified patients into low-, intermediate- and high-risk groups. According to the developers, the AI system identified approximately two-thirds of the recurrences that had not been detected by ctDNA testing.
The findings also suggested that combining tissue-based AI analysis with ctDNA testing could improve risk assessment and help identify patients less likely to benefit from additional treatment. Separate studies evaluating Histotype Px independently reported differences in outcomes between the identified risk groups. A US study also indicated that patients classified as high risk experienced a survival benefit from adjuvant chemotherapy, whereas no comparable benefit was observed among low-risk patients. These findings support further investigation into treatment selection based on AI-derived biomarkers. However, their clinical significance depends on the underlying study designs, patient populations and additional validation.
AI supports personalised cancer care
The Histotype Px development reflects a broader shift towards using AI to extract clinically relevant information from existing diagnostic data. A comparable approach was previously described by in research on AI-assisted prognosis for throat cancer. Researchers at Mass General Brigham and Dana-Farber Cancer Institute developed an AI model that analyses routine CT scans to predict extranodal extension, a prognostic factor in oropharyngeal cancer.
The model was trained and tested using imaging data from 1,733 patients. By estimating the extent of cancer spread beyond lymph nodes, the researchers demonstrated improvements in risk stratification when AI predictions were combined with existing clinical indicators. While the two technologies analyse different types of medical information, both aim to support treatment decisions by identifying patients who may require more intensive therapy and those who could potentially avoid unnecessary treatment. Neither approach replaces clinical judgement. Instead, AI-generated risk assessments provide additional information for oncologists when considering treatment options.
Regulatory progress and next steps
According to DoMore Diagnostics, Histotype Px Colorectal has achieved CE marking and ISO 13485 certification. The project has also generated seven patents. Further regulatory submissions in the Netherlands, United States and United Kingdom are planned for 2027. The developers report consistent performance across different countries, hospitals and laboratories, suggesting that the technology may be applicable in various clinical settings. The potential value lies in combining AI-based tissue analysis with established diagnostic methods, including ctDNA testing, rather than treating them as competing approaches.
Further clinical evidence will be important to determine how reliably the biomarker can guide treatment decisions and whether its use improves patient outcomes. If these findings translate into routine clinical practice, AI-assisted pathology could help oncologists select adjuvant chemotherapy more precisely, balancing the benefits of preventing recurrence against the risks of unnecessary treatment.
References
Cordis (result reports)
Add ICT&health on Google
Show more content from ICT&health in Google Search.
