It should come as no surprise that artificial intelligence (AI) is continuing to show promise in the world of medicine. Last week, a team at the Shanghai Institute of Medical Imaging published a study in Diagnostic and Interventional Imaging in which they looked into whether or not a convolutional neural network (CNN) could assist radiologists in the event of “the complicated imaging appearances and anatomical relationships of pancreatic diseases.”
The study incorporated data from over 500 patients who had undergone TI-weighted MRI screens that had been enhanced using contrast. Studies derived from almost 400 patients were used to train the CNN, while the remaining data was used as validation sets, both internal and external; these sets were collected from varying facilities. These training sets were then augmented via synthetic images derived from “generative adversarial networks,” and said synthetic images were looked over by two professional radiologists.
The team used a model known as InceptionV4; at the study’s completion, they found that their CNN’s patch-level average accuracy was 79.46 percent for the external validation set and 71.56 for the internal set. This was comparable to that of an experienced radiologist, which is incredibly encouraging.
“Our study demonstrated that deep learning had the potential to differentiate between various pancreatic diseases on contrast-enhanced MR images,” lead authors X. Gao and X. Wang wrote. This shows that AI has the potential to help out radiologists in challenging cases, rather than steal their livelihoods, a common fear held by many as promising AIs continue to crop up.
