ARTIFICIAL INTELLIGENCE
OBSERVATIONS ON THE CURRENT STATUS OF ARTIFICIAL INTELLIGENCE IN DIAGNOSTIC IMAGING Elizabeth A . Amin , MD
Artificial intelligence ( AI ) can be defined as a technique that enables machines to mimic human behavior . The first iteration of AI in diagnostic imaging was the Computer Aided Detection ( CAD ) model . This became commercially available in the late 1990s , initially applied to mammography . It was marketed as a second pair of eyes , the equivalent of a second reader , a tool that would help busy radiologists detect early breast cancer on screening mammograms . At the time of the original launch of this particular CAD program , mammograms were still being performed as filmscreen ( analog ) studies . Individual mammograms were scanned into the CAD device . The resultant images were projected on a screen with calcifications and masses marked for the radiologist ’ s interpretation . This early CAD model was not successful . The system was static . It was not designed to improve with use . Calcifications were over called as suspicious ; masses were under called and subtle asymmetries in the breast were completely missed . 1 Radiologists , afraid of ignoring the CAD for potential medico-legal reasons , incurred the wrath of patients ( too many mammographic recalls ) and general surgeons ( an excess of benign biopsies ). This was not an auspicious start in convincing radiologists of the day-to-day use of AI in clinical practice .
In the last two decades , two major events have led to a fresh look at AI in diagnostic imaging .
1 ) All imaging studies are now digital ( pixel based ). 2
2 ) The increased sophistication of imaging equipment , in particular CT and MRI scanners , has led to a marked increase in the number of images in any given patient series .
This increase , for example CTs of abdomen / pelvis and MRIs of brain , coupled with the need in many instances to compare the exams with prior studies , means that a radiologist can be faced with dozens if not hundreds of individual images - for a single patient exam . Somewhere along this trajectory there ought to be a place for AI to effectively help the radiologist ; help being defined as :
A ) Enable greater efficiency in handling the workload . ( A simple example here would be an algorithm to sort for appropriate prior studies )
B ) Assist in accurate diagnosis .
Beyond the initial definition of AI , new terms are becoming familiar to radiologists . 3 ( continued on page 10 ) APRIL 2022 9