Deep learning can now help detect abnormal chest x-rays with an accuracy equal to that of professional radiologists, according to a new article published by a team of Google artificial intelligence researchers in the journal Nature. The deep learning system can help radiologists prioritize chest x-rays and can also serve as a first response tool in emergency settings where experienced radiologists are not available.
The paper also shows how far the AI research community has come to build processes that can reduce the risks of deep learning models and create work that can be further built on in the future.
Google’s AI researchers write in their paper, “The wide range of possible CXR (chest x-rays) abnormalities makes it impractical to detect every possible condition by building multiple separate systems, each of which detects one or more pre-specified conditions.”
“A reliable AI system for distinguishing normal CXRs from abnormal ones can contribute to prompt patient workup and management,” the researchers further added.
Google’s solution was to develop a deep learning system that could detect whether a breast scan was normal or had clinically useful results. Defining the problem domain for deep learning systems is an act of balancing specificity and generalizability.
B7, the model for X-ray anomaly detection in research, is the largest of the EfficientNet family and consists of 813 layers and 66 million parameters. The deep learning model was trained on more than 250,000 x-rays from five hospitals in India. Interestingly, the researchers used 10 Tesla V100 GPUs instead of Google’s TPU processors to train the model. Recent studies show that while deep learning is nowhere near a substitute for radiologists, it can increase their productivity at a time when the world is facing a serious problem of medical experts.
