Proving the Business Case for the Internet of Things

Ping An brings AI to medical examinations

Steve Rogerson
September 4, 2018

Chinese insurance giant Ping An demonstrated five smart medical technologies at last month’s China Smart City International Expo at Shenzhen Convention & Exhibition Center.
The first was an intelligent eye screening platform (pictured) to show how diseases of the ocular fundus can be detected and diagnosed early with AI. Ping An Medical Technology unveiled the OCT intelligent diagnostic system for ocular fundus diseases jointly developed with US-based Optovue. The system is said to be the industry's first AI image screening system that seamlessly combines OCT fundus examination and AI lesion screening.
From OCT examination to patient scan code to obtain an intelligent screening report, the entire process can be completed in three minutes. Based on Ping An's deep learning technology, the intelligent eye screening platform can study and analyse thousands of data points collected during the course of the examination of the fundus, to screen for diabetic retinopathy and age-related macular degeneration (AMD).
At present, the platform has an accuracy rate for recognising diabetic retinopathy of 98.5%, and its AMD model can automatically locate the drusen (yellow deposits under the retina) and conduct relevant quantitative analysis.
Second was an intelligent imaging quality control platform. Based on AI algorithms, the platform identifies lesions automatically and smartly assists in the diagnosis according to the images displayed in a standard medical image format. In addition, the system can control the quality of the imaging diagnosis using AI technology with the aim of reducing the number of missed imaging-based diagnoses and improving the usefulness of imaging for clinical diagnosis as well as productivity at hospitals.
After the hospital technician uploads the images, quality control experts use the online image quality control assessment functionality to deliver final rating results. The advantages of the platform include examining the images using AI technology and clearly displaying the conditions of lesions. The platform provides physicians with real-time visibility of data on quality control results and the comparisons between quality and quantity of the images examined in terms of the device, location of the lesions, profile of the patient and timestamp.
Third was an intelligent imaging diagnostic system. Early this year, Ping An Technology broke world records for the detection of lung nodules and reduction in the number of false positives with accuracy rates of 95.1 and 96.8%, respectively. These two indexes are part of the Luna lung nodule analysis assessment standard, an internationally accepted standard in the medical imaging field.
The company's intelligent medical imaging screening system can complete the AI-based interpretation of CT scans and intelligent reporting within one minute. Ping An has developed medical imaging models targeting over 30 types of diseases, 200 million patients and 600 million people identified as in need of disease screening.
The system plays a key role in local hospitals. Within one week of deploying the system at a county-level hospital, more than 60 patients were detected with lung nodules and one patient was detected with suspicious lung nodules and immediately transferred to a better hospital due to the severity of the finding.
Fourth was an intelligent disease risk prediction system. Built on big data, AI and machine learning technology, the intelligent disease risk prediction system mines the data that can identify disease risk factors from a large number of characteristics. The system, which covers prediction models for 30 chronic diseases including cardio-cerebrovascular diseases, diabetes mellitus and respiratory diseases as well as their complications, automatically screens for disease risk factors from more than 3.5 million physical examinations and electronic medical records and builds intelligent disease prediction models using a machine learning method.
Taking cardio-cerebrovascular diseases as an example, it only takes approximately five minutes to generate prediction results for five main cardio-cerebrovascular diseases including coronary heart disease, stroke, atrial fibrillation, heart failure and myocardial infarction.
Developed in collaboration with several medical authorities including the Chongqing Center for Disease Control & Prevention and the Health & Family Planning Commission of Shenzhen Municipality, the system has achieved success in the prediction of infectious diseases and risk factor screening of chronic diseases. It can predict the incidence of infectious diseases one week in advance of existing methods, with an over 86% accuracy rate for influenza and hand-foot-and-mouth disease and an over 90% accuracy rate during peak periods of occurrence.
Finally, the intelligent medical assistant is a chat robot based on natural language processing technology. Rooted in medical knowledge and covering pre- and post-diagnosis medical scenarios, the assistant comes equipped with four main functions: triage, guidance, patient Q&A and follow-up.
The algorithms of the triage model support the identification of more than 2000 symptoms and the diagnosis of 500 diseases including 35 common diseases, with the accuracy rate for the top most common five answers reaching 95%.
A diabetes Q&A model is available and has compiled over 800 frequently asked questions covering 90 key categories from more than 40 million questions, with a 90.1% accuracy rate, 20% higher than data-trained open AI platforms for general purposes.
The five flagship systems have all seen application in real life. The company has established partnerships with nearly 100 Chinese healthcare organisations, including Fudan University Shanghai Cancer Center and the Fourth People's Hospital of Datong, among several other hospitals, as well as the people's hospitals of Longli county, Guizhou province and Xinhuang Dong autonomous county, Hunan province, in addition to several other local medical organisations.