Did An AI Medical Imaging System Outperform A Radiologist?

The use of artificial intelligence as a tool to help with the detection of potential medical conditions through medical imaging has become a particularly animated field of study, but one particular study has generated significant interest.

Published in The Lancet Oncology, the study claimed that a dedicated image reporting and data system focused on prostate cancer was better on average at detecting clinically significant cases compared to radiologists.

As with many studies with conclusions of this significance, there are caveats to bear in mind, the biggest of which revolves around the difference between study readings and multidisciplinary routine practice.

There were two parts of the study, one which involved study readings of a set of radiologists compared to the PI-RADS system whilst the other compared the AI readings to historical ones made in real-world conditions with a multidisciplinary system.

The first test, according to the description of the clinical trial, was a head-to-head comparison between a radiologist and an AI, with the former limited to using the same variables that the AI has access to.

In this first part of the test, PI-RADS was noticeably better, which is in line with a lot of machine learning detection algorithms being at a comparable level to radiologists but able to detect and report the information quicker than a human could.

The second part of the test is somewhat less conclusive and highlights the fundamental difference between comparing AI against an individual experienced radiologist and a multidisciplinary team able to explore the information from different perspectives in real-world conditions.

As well as this, the test was on 10,000 MRI tests, which is a relatively small proportion of tests to draw conclusions about its clinical applicability.

However, it is also a promising study and highlights the level of progress machine learning has made with detecting prostate cancer, and the potential for an AI support tool to highlight cases that merit further examination could be potentially life-saving.