How AI changes the prognostication of cancer

Kari Andresen
4 min readJun 7, 2019

Cancer specialists don’t know enough about how to divide between a dangerous and less dangerous tumour. The lack of objective and precise tools for the prognosis of cancer is the main reason behind the prevalence of over- and undertreatment of cancer patients worldwide.

– Calculations indicate that the pathologist’s assessment of the severity of the cancerous tumour is correct only in about 60 % of the cases. It`s the elephant in the room, says Håvard Danielsen, professor and director at Institute for cancer genetics and informatics (ICGI), Oslo University Hospital (UIO). The institute, celebrating its 15th anniversary this year, is performing research within biomedicine and informatics to develop and establish new methods for diagnosis and prognostication.

– Today, pathologists base the prognostication of cancer on subjective assessments. If the conclusion is incorrect, it may cause additional costs, undesirable side effects and, at worst, the death of the patient, says professor Danielsen.

– Nevertheless, to err on the safe side, many cancer patients are given more treatment than they might need. But overtreated cancer patients are susceptible to severe side effects and reduced quality of life.

Solving substantial societal

– We are making use of new technology to improve prognostication of cancers, says Danielsen.

In 2016 was the institute's project DoMore! selected as one of The Research Council of Norway's three Lighthouse projects, designed for solving substantial societal challenges using cutting-edge technology. The team is now working to offer a complete transferal from today’s complex human decision-making to an AI-based system ready for clinical use, by 2021.

In February this year–halfway into the 5-year timeline–the Norwegian government chose to launch the news of a national AI strategy during a visit to the research institute. Since then, both Prime Minister Erna Solberg and Minister of Digitalization Nicolai Astrup have emphasized DoMore! and the positive consequences AI-based prognosis will have for patients and society through faster and more accurate cancer prognostication.

The tumour heterogeneity

– Understanding how a tumour will develop is essential for the proper treatment of cancer patients, however, most cancerous tumours are complex –heterogeneous–and therefore, difficult to predict says the project leader, professor Håvard E. Danielsen.

The concepts in DoMore! are based on image analysis, more specifically on deep learning, texture analysis, and quantification of DNA. The team is studying the effect of sampling in the prostate- and colorectal cancers, and on histological grading, DNA measurements, and gene expression analysis.

The researchers examine multiple samples from each patient and tumour to model the heterogeneity in prostate-, colorectal- and lung cancer and to analyse the effect of sampling on the strength of the prognostic markers.

– The heterogeneity of a tumour is one of the most significant challenges in today’s cancer treatment, says Håvard Danielsen.

A cancerous tumour can consist of areas with different abnormalities and contains much more information than the human brain can handle. Some deviations may spread and be deadly. Others will remain calm.

Halfway through the project, the team is now making machines capable of mapping the heterogeneity of a tumour by the use of big data and deep learning. The goal is to have an AI-based system ready for clinical use by 2021.

– Up until today, we have only sampled a tiny portion of the tumour. The risk with this practice is that we are ignoring the cells that can kill the patient. The analysis work is also very time-consuming and resource-intensive. On top there is a worldwide shortage of specialists who can do this work, says Danielsen.

The DoMore!-team are now teaching the machine to perform the analysis of cancer samples by “feeding” neural networks with thousands of samples from previous cancer patients. The process is by far faster and safer than the experts can perform.

Initially, three of the most common cancer types in the world; colon, prostate and lung cancer are included in the work.

Optimising today’s treatment methods

Professor David Kerr, at Queens College at the University of Oxford, is one the world’s most experienced scientists and clinicians on colon cancer in the world and a collaborator in DoMore! The professor summarises the most important needs in the treatment of colon cancer this way:

  • We need new drugs more effective than those we currently have available.
  • We need better for solutions for how we can improve the safety of chemotherapy through the development of genetic tests that can identify those patients at highest risk of life-threatening side effects.
  • We need to find the patients who can benefit from treatment with today’s available means.

It is the last point that connects Kerr’s contribution to the DoMore!-project.

– One of the biggest challenges in today’s colon cancer research is developing methods that can find those patients who can benefit from treatment, says Kerr.

Proven possible

The 5-year long projects are now in the second half period. The results so far show that it is possible to teach a computer through Deep Learning and Big Data, not only to do the same but, to establish more robust grading systems in cancer types where pathology is less successful, while at the same time eliminating the subjective component. The researchers in DoMore! have published promising results for assessing the severity of several cancers in The Lancet Oncology. Now, more game-changing results are on the way. New and groundbreaking methods based on AI will give the specialist’s a much better basis for choosing the right treatment options for the patient already at the time of diagnosis.



Kari Andresen

Head of Unit, Visualisation and dissemination, Institute for cancer genetics and informatics, Oslo University Hospital. Writing about AI in cancer research.