MIT AI learns to spot breast cancer risk factor like an expert
A newly developed deep-learning system is capable of identifying a breast cancer risk factor in mammograms with the same level of accuracy as expert radiologists, according to MIT. The university teamed with Massachusetts General Hospital to develop what they call an automated model that is able to analyze mammograms and make assessments of the tissue that was captured.
According to MIT, the deep-learning model is the first of its kind used on real medical patients. The technology may one day serve as a screening option, working to automatically review mammograms and analyze them for signs of breast cancer. The system was trained using real mammogram scans.
The AI doesn't look for breast cancer in these scans, but rather a specific risk factor: dense tissue, which is common among women in the US and increases their odds of developing cancer. Radiologists make these determinations, which researchers say are subjective — one human radiologist may be less inclined to designate any given scan as dense verses a different radiologist.
This is something of an issue, in that the risk factor is significant enough that many states require women to be notified if they have it. By training this system to identify the tissue, hospitals and medical clinics may one day have an automated screening tool that helps make determinations about tissue density.
This technology was put to use at Massachusetts General Hospital, where it first received mammogram scans before they were passed onto the human radiologist. The system's assessment of the scan was passed along to the human expert with the scan, where the radiologist could then view the AI's results and make their own determination. Overall, the AI and radiologists were in agreement about scan results 94-percent of the time.
SOURCE: MIT