Prostate biopsy with cancer probability (blue is low, red is high). This case was originally diagnosed as benign but changed to cancer in a further review. The AI accurately detected the cancer in this complicated case. Credit: Ibex Medical Analytics
A study published today (July 27, 2020) in The lancet Digital health by UPMC and the University of Pittsburgh researchers demonstrate the highest accuracy to date to recognize and characterize prostate cancer using an artificial intelligence (AI) program.
“Humans are good at recognizing abnormalities, but they have their own biases or past experience,” said lead author Rajiv Dhir, MD, MBA, chief pathology and vice president of pathology at UPMC Shadyside and professor of biomedical informatics at Pitt. “The machines are separate from the whole story. There is definitely an element of standardization of care. “
To train the AI to recognize prostate cancer, Dhir and colleagues provided images of more than a million parts of stained tissue slides taken from patient biopsies. Each image was tagged by expert pathologists to teach AI how to discriminate between healthy and abnormal tissue. The algorithm was tested on a separate set of 1,600 slides taken from 100 consecutive patients seen at UPMC for suspected prostate cancer.
During the tests, the AI demonstrated a sensitivity of 98% and a specificity of 97% in the detection of prostate cancer, significantly higher than previously reported for algorithms that work from tissue slides.
Additionally, this is the first algorithm that extends beyond cancer detection, reporting high performance for classification, size, and invasion of surrounding nerves. These are all clinically important features required as part of the pathology report.
AI also marked six slides that were not noticed by expert pathologists.
But Dhir explained that this does not necessarily mean that the machine is superior to humans. For example, in the course of evaluating these cases, the pathologist might simply have seen enough evidence of malignancy elsewhere in that patient’s samples to recommend treatment. However, for less experienced pathologists, the algorithm could act as a failure test to detect cases that might otherwise be missed.
“Algorithms like this are especially helpful in injuries that are atypical,” said Dhir. “A non-specialized person may not be able to make the correct evaluation. That’s a great advantage of this type of system. “
While these results are promising, Dhir cautions that new algorithms must be trained to detect different types of cancer. Pathology markers are not universal in all tissue types. But he didn’t see why that couldn’t be done to adapt this technology to work with breast cancer, for example.
###
Reference: July 27, 2020, The Lancet Digital Health.
Other study authors include Liron Pantanowitz, MBBCh., From the University of Michigan; Gabriela Quiroga-Garza, MD, of UPMC; Lilach Bien, Ronen Heled, Daphna Laifenfeld, Ph.D., Chaim Linhart, Judith Sandbank, MD, Manuela Vecsler, of Ibex Medical Analytics; Anat Albrecht-Shach, MD, from Shamir Medical Center; Varda Shalev, MD, MPA, of Maccabbi Healthcare Services; and Pamela Michelow, MS, and Scott Hazelhurst, Ph.D., from the University of the Witwatersrand.
Funding for this study was provided by Ibex, which also created this commercially available algorithm. Pantanowitz, Shalev and Albrecht-Shach report the fees paid by Ibex, and Pantanowitz and Shalev are part of the medical advisory board. Bien and Linhart are authors of pending patents US 62 / 743,559 and US 62 / 981,925. Ibex had no influence on study design or interpretation of results.
