Artificial intelligence is as reliable as a pathologist - a tool automates the identification of a prognostic factor in colorectal cancer samples
Colorectal cancer is the second most common fatal cancer worldwide, after lung cancer. Once diagnosed, the 5-year survival rate for a Finnish patient is around 68%. It has long been hoped that AI will help both cancer research and practical cancer diagnostics. It has become a popular development target in medical imaging, including pathology.
The neural network model developed in this project identified tumors and connective tissue with 96% accuracy. After identification, the model calculated the amount of connective tissue in the tumor, which is one of the prognostic factors predicting lifespan. In six out of ten cases, the AI’s results agreed with those of a pathologist, which is in the same range as when several pathologists estimate the percentage of connective tissue.
“Pathologists do a very important job in analyzing different tumor tissue characteristics in microscopic samples of tumors. The work is done visually, so it can be difficult to replicate, and the mutual assessments of pathologists can differ greatly," describes Liisa Petäinen, who is responsible for development work in the AI Hub Central Finland project.
Petäinen believes that the spread of AI methods will speed up routine tasks in the future and free up pathologists' time to analyze more demanding cases.
“Access to treatment could also be accelerated by reducing the time it takes to process samples.”
Digital pathology makes practical work easier
The potential of AI models in pathology is significant. Digitized microscopic images provide a wealth of information, even down to the intracellular structures.
“By combining the information in the images with other clinical patient data, it is possible to find things that a human would miss and thus improve the effectiveness of treatment and also speed up access to care," Petäinen comments on the potential of digital methods.
The automation of routine tasks can bring considerable relief to the pathologist’s work, although the final decision-making power remains with the pathologist.
An annoyingly cunning cancer
The large amount of connective tissue in a tumor suggests a poor prognosis for patients, particularly in breast and colorectal cancer. In other words, colorectal cancer is regrettably cunning and ruthlessly exploits the resources of its host.
In the case of connective tissue, the tumor is likely to harness the connective tissue to reinforce its own growth. In fact, several studies have shown that a large amount of connective tissue in a tumor indicates a poor prognosis for the patient.
For these reasons, among others, Teijo Kuopio, a pathologist and a Research Director at the Central Finland Hospital District, considers research and development in cancer and pathology to be particularly important.
“AI-based technologies help to automate tasks, increasing the reproducibility of diagnosis and the speed of sample analysis.”
The method has been developed in the AI Hub Central Finland project, funded by the European Regional Development Fund and carried out by the Ģֱ's Faculty of Information Technology, in cooperation with the Central Finland Hospital District and Central Finland Biobank.
More information
Researcher Liisa Petäinen
lihesalo@jyu.fi
+358 44 553 9575
Research Director Teijo Kuopio
Central Finland Hospital District
teijo.kuopio@ksshp.fi
+358 40 546 7633