The Lancet Digital Health: AI algorithm snares prostate cancer with 98% accuracy

On July 27, 2020, researchers at the University of Pittsburgh Medical Center (UPMC) and researchers at the University of Pittsburgh published a new study published in The Lancet Digital Health that showed that the use of artificial intelligence (AI) programs to identify and characterize prostate cancer was by far the most accurate. Rajiv Dhir, chief pathologist and vice president of pathology at upmcIN,
, and a professor of biomedical informatics at the University of Pittsburgh and senior author of the study, says humans are good at identifying anomalies, but they have their own biases or past experience.
machines are not like this.
machines have an element of standardized care.
to train artificial intelligence to identify prostate cancer, Dhir and his colleagues provided images of more than a million slices of dyed tissue extracted from a patient biopsy.
each image is marked by an expert pathologist to teach artificial intelligence how to distinguish between healthy and abnormal tissues.
, the algorithm was then tested on a separate set of 1,600 slides taken from 100 suspected prostate cancer patients at UPMC.
during the testing process, Artificial Intelligence showed 98 percent sensitivity and 97 percent specificity in detecting prostate cancer, significantly higher than previously reported algorithms using tissue slices.
, this is the first algorithm to go beyond cancer detection, reporting high performance in violation of tumor grading, size, and peripheral nerves.
these are important clinical characteristics that need to be part of the pathology report.
AI also marked six slides that pathologists did not notice.
an overview of the algorithms and clinical deployment of Galen Prostate's second reading system, but Dhir explains that this does not necessarily mean that machines are superior to humans.
, for example, in assessing these cases, pathologists can simply see enough malignant evidence from other samples of patients to recommend treatment.
However, for inexperienced pathologists, the algorithm can act as a failure insurance to catch cases that may be missed.
in the paper, researchers report on the development of an algorithm for evaluating digital prostate CNB slices, medical-grade artificial intelligence, and the successful use of the ai-intelligence tool in conventional clinical practice.
based on a large, blind, externally validated data set, the researchers demonstrated the algorithm's high precision to identify and quantify prostate cancer, distinguish between low- and high-level tumors, and detect nerve peripheral immersion.
This is the first report based on an artificial intelligence algorithm that transcends cancer detection and grading of prostate cancer in histopathological images and is one of the first clinical examples of using artificial intelligence algorithms in conventional pathology practice.
in summary, researchers at the University of Pittsburgh Medical Center reported on the development of artificial intelligence-based algorithms, external clinical validation, and deployment in routine practice, which detect, grade, and evaluate other clinically relevant tumor characteristics of digital prostate CNBs.
the data show that the artificial intelligence-based algorithm can be used as a tool for automatic screening of primary diagnoses of prostate CNBs, evaluating registered cases for quality control purposes, and standardizing reporting to improve patient management.
studies on clinical efficacy in other laboratories are also under way.
.