Stanford researchers use AI to monitor rare cancer - San Jose Spotlight

""The major finding was that this is a very feasible way that we can apply AI in this area of health care," Chang said. "Because a lot of what we do is very reliant on images and interpretation of images, there's a lot of opportunity for computer vision to help augment clinicians' ability to diagnose, to treat and to surveil different disease processes.""

"In the study, a computer vision model was trained on about 1,500 endoscopic images of the nasal pharynx from 192 patients. While the model performed strongly at detecting osteoradionecrosis versus healthy tissue, the system showed moderate accuracy in identifying recurrences of nasopharyngeal carcinoma. At times, the model confused recurrent NPC with radiation-related damage or normal-appearing tissue. Still, Dr. Chang said with a larger dataset, the models could perform even better."

A computer-vision model was trained on about 1,500 endoscopic images of the nasopharynx from 192 patients. The model detected skull base osteoradionecrosis versus healthy tissue with about 85% accuracy, comparable to experienced clinicians. The model showed moderate accuracy identifying recurrent nasopharyngeal carcinoma and sometimes confused recurrence with radiation-related damage or normal-appearing tissue. Nasopharyngeal carcinoma develops deep behind the nose, can be subtle, is often diagnosed late, and disproportionately affects Asian American communities. Larger datasets are likely to improve model performance. Funding was provided by the Center for Asian Health Research and Education.

Read at San Jose Spotlight