Artificial intelligence-driven cancer prognostication

The importance of prognostication

Imagine being one of the approximately 1,8 million people diagnosed with colon cancer worldwide every year. It is likely that one of the first things you would like to know is your prognosis. But the knowing the prognosis of your specific tumor is also crucial for your doctor when deciding which treatment would be best for you.


Most colon cancers are at stage II or III when discovered. In this case your overall chance to survive the next 5 years is about 80% with only surgical removal of the tumor alone. However, there is a 20% chance that your cancer is more aggressive and requiring chemotherapy in order to stop the cancer from coming back. Until now we have been lacking precise methods to determine which patients fall into the 20% group. Because of this, most patients at these stages are today recommended so called adjuvant therapy – different forms of chemotherapy – to reduce the chance for cancer recurrence.

However, since 80% of these 1600 patients don´t need this in order to stay healthy (Quasar Collaborative Group, 2007), and many of the cancers in the high-risk group will recur despite this adjuvant therapy, the use of adjuvant therapy has only increased overall survival in the group with 2-7% across most clinical studies.



In other words – most of the patients we put through tough adjuvant therapy will have no benefit, and the treatment could be avoided if we could identify who these patients are. The chemotherapy treatment comes at a high financial and human cost, and may have serious side effects, sometimes making it impossible for patients to go back to work and or perform normal daily activities. For about 0,5-1% of patients it may even cause death, which underlines the need for a more precise marker to aid clinicians and patients deciding on adjuvant treatment.


Now imagine being one of the approximately 80% of the 1,8 million people diagnosed with colon cancer in stadium II or III every year receiving the good news that there will be no need for the most devastating form of chemotherapy.


Read more - The DoMore! project

To best treat cancer, we must first understand how the disease will develop. For five years, researchers at Oslo University Hospital, University of Oslo, University of Oxford and University College London have used deep learning and vast reserves of patient data to develop algorithms that provide the patients with a more precise prognosis to allow for better choices in treatment. The prognosis is calculated from a digitally transferred and stored scanned tumor sample – we call it “in silico pathology”. The excellent results have been published in international journals like Nature and Lancet.

You can read more about the project at Domore.no