AI Screening for DR and Chronic Diseases 2018-03-29 Emma Xu, Dr. - PowerPoint PPT Presentation

AI Screening for DR and Chronic Diseases 2018-03-29 Emma Xu, Dr. You Li

Impact of Blindness in Elderly Survival Curve for All-cause Mortality in Singapore Malay Impact of Visual Impairment and Eye diseases on Mortality: the Singapore Malay Eye Study (SiMES) Scientific Reports volume 5 , Article number: 16304 (2015)

Outline • Why are there high rate for blindness in China • How AI can be used for the early detection of eye diseases • How AI DR screening is used in hospital

Patients of Blinding Diseases in China • Blindness or Low Vision: 1 / 20 • No. of Patients Age-related • Cataract: 60 million Age-related Macula 30 million Degeneration: Glaucoma: 10 million • Chronic-diseases-related 12 million Diabetic Retinopathy: Retinal Vessel 7 million Occlusion:

Example: Diabetic Retinopathy (DR) Taking fundus image Healthy Fundus Fundus with DR Lesions

Deficiency of Ophthalmologists in the Rural China 1/5 County Hospitals have Ophthalmology Department 40% - 200% Higher prevalence of blindness than cities

AI Fundus Screening Algorithm 8.37M Fundus Photos 100+ professors/3yrs Cross labelling 30+ Diseases Recognized 2 million fundus photos Uploaded in 2017

Architecture Web Server UI Data Flow Annotation Storage Training System Storage Nvidia DR Grading GPU Server Lesion Detection General Classification Macula detection Cataract Classification Client Web Server Glaucoma Service …… Front-end GPU Server Computer or smart device AI services: Recognition/detection/inference.

Community Screening for Elderly in 4 rd -tier Cities Disease found in 60+ yrs 2/3 Positive 15% have whole-body chronic diseases 80% did not know the disease before 90% will go to hospital 28% Cataract: 14% Retinal Arteriosclerosis: 10% Age-Related Macular Degeneration: 2.1% Diabetic Retinopathy:

Screening for Working Population 21% Positive 10% Need to see a doctor 85% Did not know the disease before Age-related Macula 5% Degeneration: Probable 4% : Glaucoma: Pathological Myopia: 2.6% Retinal 2.3% Arteriosclerosis:

Prof. Li: Airdoc AI in DR screening • Li You M.D. • Dr. of Ophthalmology • Professor of Medicine • Chief Physician • Department of Ophthalmology, Changzheng Hospital Shanghai, China

Prof. Li: Airdoc AI the screening of DR(Diabetic Retinopathy)

Probably the Most Famous DR Patient

What exactly is DR? No apparent symptom in early stage Irreversible lesion Lower the risk of blindness by controlling on time Early Intervention to lower risk of glaucoma

DR Prevalence Patients of Diabetes 110 million DR Prevalence in Patients of Diabetes Prevalence of DR in Patients of Diabetes 25%－40% NDR: 62.5%-75.3% No. of Ophthalmologists 35,000 NPDR: 21.4%-30.1% No. of Doctors/No. of DR Patients PDR: 3.3%-7.4% 1/3250 DME: 5.2% CDME: 3.5%

Romote Regions of China Need Medicine and Doctors

Cataract Operation in Village

Later Period of Glaucoma TAO

Deep Learning in DR Automatic Detection on lesions

Features of Airdoc DR • Accuracy in DR grading • Lesion Detection • Seamless integration with fundus cameras

DR Patient found by Airdoc AI :Mr. Sheng

The Future of Fundus Screening More chronic diseases are found to be related to lesions in fundus • Senile Dementia • Parkinson • AIDS • Cerebral Stroke • Coronary Heart Disease • Nephropathy • Cerebral Small Vessel Disease(CSVD)

Thank You! emma@airdoc.com