In a research paper published in the international journal Nature Communications, Ping An AskBob Doctor's smart imaging model for diagnosing pelvic and hip injuries has been shown to overcome the limitation of other artificial intelligence (AI) systems, which can only detect individual fractures.
The model, a deep learning algorithm, can help physicians make faster, more accurate diagnoses and save lives. Hip fractures mainly occur in elderly people and patients with major trauma. Although they are not directly fatal, complications can lead to a high mortality rate. Reducing the rate of missed diagnoses, improving the comprehensiveness of the detection and providing accurate diagnoses are critical.
Ping An Health Technology Research Institute, Ping An Smart City, Ping An Good Doctor and the Department of Trauma and Emergency Surgery of Chang Gung Memorial Hospital in Taiwan jointly released the research paper "A scalable physician-level deep learning algorithm detects universal trauma on pelvic radiographs".
The research team gathered the pelvic X-ray data of 1,888 emergency room patients to evaluate the model's performance in obtain the fracture results and locations. It achieved an overall accuracy of 92.4%. Compared with general clinical diagnoses, the model substantially improves detection accuracy, speeds up the treatment progress, improves the treatment effectiveness on high-risk patients, and reduces the economic cost to patients through earlier detection and treatment.
As of March, 2021, the AskBob Doctor AI pelvic trauma detection technology has been used in the real-life clinical environment in Chang Gung Memorial Hospital in Taiwan for nearly eight months.
Since adopting the AI system, the misdiagnosis rate has dropped from 9.7% to 0.7% among emergency physicians, 11.3% to 1.58% among resident physicians and 6% to 0.5% among specialist physicians. The model has performed comparably to radiologists and certain orthopaedic specialists in terms of quantitative indicators such as sensitivity and specificity.