Prototype of CareRare Rare Disease Screening Machine
Website of online CareRare Rare Disease Screening System
What it does
CareRare is an AI-based rare disease diagnosis auxiliary. One can quickly get the possible diagnose solutions by providing the abnormal face and body pictures. We can diagnose up to 47 rare diseases with high accuracy rate of 94%.
Rare genetic syndromes (rare diseases) are not rare. There are about 20million rare disease patients in China, and about 300 thousand new patientsare added every year. However, each rare disease patient needs toexperience 7.6 years and be misdiagnosed 2.3 times before being diagnosed rightly. We have come up with the idea of whether we can use AI to intervene in the early screening process of rare diseases. With lower misdiagnosis rate, the patients can carry out treatment intervention as soon as possible to improve prognosis, reducing disease injury and economic loss of the patient's family
How it works
CareRare is a characteristic analysis and diagnosis system for genetic syndromes. W utilize the information includes pictures, videos, words of body parts of the patient (suspected patients), body morphological characteristics, and human phenotype information. We construct a morphological model of the patient's face, palm and joints based on images of various parts of the patient, and construct a posture model based on the patient's action video. We take morphological model, postural model and human phenotype ontology (HPO) as input. Based on the self-developed AI-based algorithms, we independently designed a clinical screening machine for genetic syndromes that is easy to deploy and loaded with functions. The output results include the name of suspected diseases, the probability of each disease, follow-up examination suggestions, medication recommendations and other auxiliary diagnosis and treatment suggestions for doctors.
In 2019, we visited 4 tertiary hospitals in Shanghai and conducted research on hundreds of rare disease patients and practitioners. We came up with the idea of whether we can use AI to intervene in the early screening process of rare diseases. In 2020, we completed the first edition of the case collection of 47 rare diseases, and designed the first set of auxiliary diagnosis algorithms for genetic diseases in China. We solve the problems of insufficient training data, same feature on different diseases, and privacy protection step by step. Finally, the accuracy rate of the CareRare was 100%, and the diagnostic accuracy rate was 94%. At the same time, we develop a diagnosis and treatment service system that can run on the online and offline model. In 2021, we constructed the first rare disease dataset of patient cases on the order of tens of thousands. In September of the same year, the team approached the Shanghai Children's Medical Center as the first testing unit to conduct clinical tests and screened 350 high-risk patients within half a year. The feasibility of the model is verified. In 2022, we have submitted an application for National Medical Products Administration (NMPA III) as the medical device qualification in China.
How it is different
CareRare is the first solution in China to use artificial intelligence technology to assist in the diagnosis of rare diseases. We build the largest rare disease case dataset in China. We design a rare disease screening algorithm based on self-supervised few-shot learning to cope with the data scarcity problem. We design a saliency-based facial feature anomaly detection algorithm to solve the problem of the same features on different diseases. We also design a user privacy protection algorithm based on federated meta-learning to deal with privacy data protection. This solution can significantly improve the work efficiency of doctors, shorten the diagnosis time. Importantly, the program can reduce the cost of patients,from 15,000 yuan to 800 yuan a single time, and the time required to be reduced from a few months to two weeks.
We have negotiated with several hospitals in Shanghai and plan to cooperate with several hospitals to carry out pilot projects. We have collected 8 intent orders so far. Our NMPA application is in the classification and definition stage. It is expected to obtain NWPA certification by the end of 2023 and open the Series A financing. In the future, we will continue to iteratively develop professional large-scale diagnosis and treatment equipment, and use multiple diagnostic indicators to realize multi-modality physiological diagnosis technology to prepare.
National Postgraduate Artificial Intelligence Innovation Competition, First Prize China Innovation Challenge—AI Medicine Track, Top Five in China National Smart Healthcare Challenge, Second Prize Yantai Innovation and Entrepreneurship Competition, Second Prize