FOR RESEARCHERS at the Karolinska Institutet in Sweden and the Radboud University Medical Center in the Netherlands, there it is: artificial intelligence (AI) for the diagnosis and classification of prostate cancer is ready to be included in clinical practice. Not to replace the pathologists, but to support them, to ensure a good level of performance and to guide towards the choice of the best treatment for the patient. A position supported by solid data, published on Nature Medicine.
Improve accuracy
Prostate cancer is the most common male cancer and its diagnosis is based on a microscopic analysis of the tissue biopsy by the pathologist. It follows that the process is a subjective evaluation, which can be affected by different factors (for example the operator’s experience), so much so that it can happen that more professionals evaluate the same biopsy differently. The introduction of artificial intelligences for image diagnosis and classification could contain the problem.
Prostate cancer: so artificial intelligence helps in the diagnosis
by Viola Rita
The PANDA challenge: here are the best AI
That some AIs were capable of achieving human-like performance in diagnostics was already known, but broad validation was still lacking. In fact, AI often achieve good results on datasets similar to those on which they were trained, but when tested on databases from other sources, they fail.
To verify how advanced the technology was, so if machines can really improve diagnostic techniques on a global scale, the Swedish and Dutch researchers launched the PANDA challenge (Prostate cANcer graDe Assessment). The competition, which lasted three months, saw the participation of more than a thousand developers of artificial intelligence algorithms, who challenged each other on the largest diagnostic data set for prostate cancer: over 10,000 images of prostate cancer biopsies. The three “best” AIs received a cash prize.
Medicine, artificial intelligence gives you the “second opinion”
What matters most, however, is that the 15 best AIs developed were included in more in-depth analysis work, tested on other datasets, and confirmed to perform similarly to that of more experienced pathologists. However, there is still work to be done to train the systems even on rare biopsy datasets or from countries with a more diverse demographics.
Unity is strength
Even in this case, however, the transition to the clinic is neither automatic nor taken for granted. The same researchers point out that the next step will be figuring out how to introduce AI into physicians’ workflow. “The idea is not that AI replaces human experts, but rather acts as a safety net to prevent pathologists from not recognizing cancer cases and as a support for standardizing assessments,” reiterated Lars Egevad of the Karolinska Institutet. In fact, follow-up research data shows that pathologists who use AI to diagnose prostate cancer perform better and more consistently, an indication that, as Wouter Bulten of Radboud University Medical Center pointed out, “the synergy of pathologists and combined artificial intelligence is superior to both taken individually ”.
Blood cancers, so the patient is followed online
by Tina Simoniello
“Now – added Bulten – we must demonstrate that more accurate diagnostics using AI improves outcomes for patients with prostate cancer, while reducing healthcare costs. For this, we need to integrate AI into the diagnostic workflow. A problem, since only 10% of hospitals worldwide use digital images, while the other 90% continue to view biopsies under a microscope. And this is one of the biggest obstacles to the adoption of AI in clinical practice “.
.