by John W. Mitchell
, Senior Correspondent | October 18, 2018
Physician clinical judgment matters when it comes to the use of artificial intelligence (AI) applications, according to a study just published in the Journal of the American College of Radiology.
Combining a radiologist’s opinion and various imaging parameters in an AI algorithm resulted in greater accuracy compared to working with the imaging parameters alone.
“The way a radiologist looks at an image to provide interpretation is predominantly anatomical and to an extent, abstract [as it is a factor of] the radiologist's clinical experience and gut feeling,” Dr. Adarsh Ghosh, study author, Department of Radio-diagnosis and Imaging, AIIMS, New Delhi, India told HCB News.
Numed, a well established company in business since 1975 provides a wide range of service options including time & material service, PM only contracts, full service contracts, labor only contracts & system relocation. Call 800 96 Numed for more info.
The algorithm deals with image-derived parameters predominantly in a numerical fashion to provide interpretations. His findings, Ghosh explained, prove that the two very different AI and human approaches can be combined to yield greater accuracy and improved patient care.
In the study, he relied on breast imaging data sets from the University of California, Irvine Machine Learning Repository. Three machine learning algorithms were trained to provide cross-validation. The evaluation metrics that were used to compare the two cohorts – AI alone and radiologists/AI combined – included lesion shape, density, and patient demographics. Overall, radiologist oversight outperformed the stand-alone AI application.
“I wanted to address the notion that AI will replace radiologists,” said Ghosh. “The common hype of the radiologist being rendered redundant is unfounded. The radiologist and AI algorithms will work in tandem and, synergistically, may provide a diagnosis which will be closer to the truth.”
Ghosh noted that the study was a very simple design as he used the most basic AI algorithms available. AI algorithms, he explained, are versatile, allowing both continuous and categorical predictors. He believes the paper will set a benchmark for further AI research as radiologist-augmented AI workflow needs more evaluation - especially when it comes to deep learning algorithms.
“The key finding that will interest radiologists is that combining a radiologist's opinion and various image parameters in the AI algorithms obtained greater accuracy than the AI algorithms working on the image parameters alone,” he concluded.