Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Computer-Based Biomarker May Aid Prognosis Prediction In Colorectal Cancer

A digital biomarker, developed using deep-learning techniques, may help select patients with colorectal cancer who might benefit from adjuvant chemotherapy
05 Feb 2020
Colon and Rectal Cancer;  Translational Research

Author: By Laura Cowen, medwireNews Reporter

medwireNews: A computer-based biomarker, which scans conventional haematoxylin- and eosin-stained sections of colorectal cancer (CRC), can accurately distinguish between patients with good and bad prognoses, study findings indicate. 

“The marker could potentially be used to improve selection of adjuvant treatment after resection of colorectal cancer by identifying patients at very low risk who could have been cured by surgery alone, as well as patients at high risk who are much more likely to benefit from more intensive regimes”, write Håvard Danielsen, from Oslo University Hospital in Norway, and co-authors of the study published in The Lancet

The researchers developed the DoMore-v1-CRC classifier using a deep-learning method that initially involved scanning more than 12,000,000 image tiles from 828 patients with a distinctly good or poor disease outcome. 

The resulting 10 convolutional neural networks were integrated into a prognostic biomarker using images from 1645 patients with a non-distinct outcome. The biomarker was then tested on a separate cohort of 920 patients and independently validated among 1122 patients with stage II and III CRC who were treated with single-agent capecitabine during the QUASAR 2 randomized controlled clinical trial. 

After adjustment for established prognostic markers, such as pN and pT stage, lymphatic invasion and venous vascular invasion, the investigators found that the biomarker was significantly associated with cancer-specific survival. Specifically, the hazard ratio (HR) for CRC-related death with a poor versus good prognosis according to DoMore-v1-CRC was 3.04 overall, 2.71 for stage II cancers and 2.95 for stage III cancers. 

By comparison, the HRs for CRC-related death according to pN stage, pT stage, and lymphatic invasion were 5.94 (pN2 vs pN0), 1.86 (pT2 vs pT1) and 1.75 (pT4 vs pT3), and 1.66 (lymphatic invasion vs none), respectively. Venous vascular invasion was not significantly associated with cancer-specific survival in this cohort. 

The biomarker accurately classified 76% of patients as having a good prognosis and significantly correlated with variables such as age, pN and pT stage, histological grade, location, tumour sidedness, BRAF mutation status and microsatellite instability. 

Håvard Danielsen et al note that the DoMore-v1-CRC classifier “is technically simple to apply and can be delivered at pathology laboratories in a variety of settings”, with full processing complete within a median of 2.8 minutes. 

The investigators say they now “plan to undertake prospective adjuvant trials stratifying patients into different prognostic groups using the DoMore-v1-CRC biomarker and randomly assigning patients into observation, low intensity, and high intensity regimes, depending on relative risk score.” 

In an accompanying comment, Adrian Specogna (Prolatent Health Technologies, Calgary, Alberta, Canada) and Frank Sinicrope (Mayo Clinic, Rochester, Minnesota, USA) caution that although the study “adds value to the application of deep learning methods in cancer research”, automated bias can be a concern with new technology. 

They say that “clinicians should be mindful that although a system might be automated, it is entirely data driven and not trained to understand why, using a causal perspective, an outcome has occurred.” 

Noting that “little attention is given to how learned biases carry forward and relate to errors in clinical decisions with the potential to harm patients”, the commentators conclude: “Recognising these limitations, and given the consequences of medical errors, automated prognostication procedures are unlikely to eliminate the need for expert human intervention, although they could allow for greater efficiency.” 


Skrede OJ, De Raedt S, Kleppe A, et al. Deep learning for prediction of colorectal cancer outcome: a discovery and validation study. Lancet 2020; 395:350–360. doi:10.1016/S0140-6736(19)32998-8

Specogna AV, Sinicrope FA. Defining colon cancer biomarkers by using deep learning. Lancet 2020; 395:314–316. doi:10.1016/S0140-6736(20)30034-9

medwireNews (www.medwireNews.com ) is an independent medical news service provided by Springer Healthcare. © 2020 Springer Healthcare part of the Springer Nature group

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.