Google did deep learning research using a large data set of retinal images to diagnose diabetic retinopathy which leads to blindness. AI could prove transformative in healthcare. Google has developed an algorithm which is trained by a data set of 1,28,000 images. Each image has been reviewed by a 3-7 expert ophthalmologists. These are fundus images. The symptoms could be micro-aneurysms, haemorrhages, hard exudates etc. AI detects referable diabetic retinopathy (DR). It is possible to do this fast and in greater numbers. Early diagnosis is preventive. The software was tested using 12000 images. The performance rating was 0.95, the highest being 1.The ophthalmologists rating was 0.91. Google now is focusing on 3D retinal images. It may point out cases which doctors would have missed. Instead of a doctor alone or algo-alone, it is better to have an algo complementing a doctor.

Such algos can be used to detect cardio-vascular disease.