Advancements in cancer research and detection have been far and wide in 2016, with the launch of the National Cancer Institute’s Genomic Data Commons, advancements in medical data visualization, and now Beth Israel Deaconess Medical Center (BIDMC) and Harvard Medical School’s cancer-detecting Artificial Intelligence (AI).
The machine, which won a competition at the International Symposium on Biomedical Imaging, uses deep learning, a method of training AI to recognize speech, images and objects, in order to read pathology images and identify hazardous cells.
After being fed hundreds of images of both cancerous, and noncancerous lymph nodes, the AI was able to reach a diagnostic accuracy of 92 percent, and decrease the human rate of error by 85 percent.
Until now, it has been up to physicians and their microscopes to identify cells and make diagnoses. Although human pathologists have a 96 percent accuracy rate of diagnoses, problems such as standardization across the board and the time it takes pathologists to manually load and examine slides are areas that the AI can drastically improve.
According to BIDMC’s Andrew Beck, co-author of the study, when human pathologists’ analyses are combined with that of the AI’s the accuracy rate shoots to a nearly perfect 99.5 percent. He says, “Our results … show that what the computer is doing is genuinely intelligent and that the combination of human and computer interpretations will result in more precise and more clinically valuable diagnoses to guide treatment decisions.”
The goal now is to further train the AI, by feeding it even more larger and more diverse datasets. Pathologists will also be trained on how to effectively engage with the AI and other similar technology emerging in the medical field.
Despite the AI’s outstanding performance and accuracy in diagnosing, Beck says there is still much room for improvements. “I hope to see the model performance continue to improve at this task as well as other tasks in cancer pathology. And in parallel I hope new systems and processes are developed to enable implementation of the these tools in the clinical workflow.”
Aptly named, Enclothed Cognition is the official Medelita blog for medical professionals interested in topics relevant to a discerning and inquisitive audience. Medelita was founded by a licensed clinician who felt strongly about the connection between focus, poise and appearance.
