‘The AI is simply outstanding’: Algorithm spots prostate cancer similarly to expert pathologists

A new AI system performed just as well as expert pathologists at identifying prostate cancer, and holds promise for improving some of the specialty’s shortcomings, experts wrote Jan. 8 in the Lancet Oncology.

"Our results show that it is possible to train an AI system to detect and grade prostate cancer on the same level as leading experts," study author Martin Eklund, with Karolinska Institutet in Sweden, said in a statement. “This has the potential to significantly reduce the workload of uro-pathologists and allow them to focus on the most difficult cases."

Pathology, as it’s practiced today, comes with a level of inherent subjectivity, Eklund wrote. Pathologists analyzing biopsies may come to different conclusions when looking at the same samples. And doctors are then forced to determine a patient’s best course of treatment based off multiple different assessments.

Eklund, along with colleagues at Tampere University in Finland, developed an approach to change this.

They created digital images of more than 8,000 prostate cancer biopsies taken from nearly 1,200 men in Sweden. Approximately 6,600 of those samples were used to train the AI to spot differences between biopsies with or without cancer.

Results showed the system was almost perfect at picking out cancer and estimating how far along the disease was in biopsy samples. The AI was also just as accurate at assessing the diseases’ severity as international experts who used Gleason scoring.

“When it comes to grading the severity of the prostate cancer, the AI is in the same range as international experts, which is very impressive, and when it comes to diagnostics, to determine whether or not it is cancer, the AI is simply outstanding," co-author Lars Egevad, professor in pathology at Karolinska Institutet said in the statement.

Although promising, the findings require further validation before moving toward clinical use, the researchers noted. That’s why a multicenter study taking place across nine European countries has already started. It’s goal is to train the system to identify cancer in samples taken from various labs, with different scanners and growth patterns. Egevad and colleagues say the research should be completed by the end of this year.

"AI-based evaluation of prostate cancer biopsies could revolutionize future health care," said fellow researcher Henrik Grönberg, with Karolinska Institutet. "It has the potential to improve the diagnostic quality, and thereby secure a more equitable care at a lower cost."