Deep learning for disease diagnosis

Challenge Diabetic Retinopathy is currently the leading cause of blindness in working age adults. 23% of all 110 million diabetics in China suffer from Diabetic Retinopathy (DR), if untreated DR will lead to blindness. Poor distribution of eye care providers compounds the problem with more than 90% of the 22,000 eye physicians in the nation are in large- or  average-sized cities, while only 3,000 eye physicians are in China’s 1,237 rural counties, or about one to two physicians per county.

Opportunity The Global Ophthalmic Devices Market is expected to reach USD 55.5 billion by 2024. Rising prevalence of diabetics, increasing management financial support, change in lifestyle and global aging population are the factors driving the market growth.

Solution Retinal images can be used for diagnosis as changes or anomalies in microvasculature can be easily viewed with a fundus camera.

Retinacheck’s Convolutional Neural Network algorithm analyses these images and can diagnose Diabetic Retinopathy with an accuracy of over 94%, better than an opthamologist.

Advantages
✓ Uses the latest in convoluted neural network technology which mimic mathematical processes in the visual system.
✓ Learning can be done on an image level (disease/healthy) and the result is given on a lesion level.
✓ Low cost solution that is easily available in rural areas

Status Algorithm is trained on a dataset of almost 90,000 fundus photographs from the 2015 Kaggle Diabetic Retinopathy competition and a private dataset of almost 110,000 photographs. RetinaCheck has a patent pending for deep learning techniques that improve the algorithm’s
learning capabilities.