What problems are you trying to solve?
Over 400 million people worldwide have diabetes, and it is estimated that 1/3 of all patients with diabetes will develop DR over the course of their lives. DR is a silently progressing disease with minor or no symptoms in the beginning, but could eventually lead to blindness if not properly managed. Blindness is largely preventable with regular screening, typically on a yearly basis. But with the growing number of people with diabetes worldwide, the traditional practice of manual screening by diabetes and eye care specialists simply cannot scale. Furthermore, diabetes specialists are not trained to interpret retinal images and therefore may not have the confidence to do so.
While annual DR screening is recommended for all diabetic patients, on average only less than 50% of patients get screened, even in the developed world. This is mainly due to the following problems:
(1) Shortage of eye care specialists (ophthalmologist to diabetic patient ratio = 1:30002 (Thailand); 1:1600 (USA))3
1 JAMA ophthalmology 135.7 (2017): 706-714.
2 “A human-centered evaluation of a deep learning system deployed in clinics for the detection of diabetic retinopathy.” Proceedings of the 2020 CHI conference on human factors in computing systems. 2020.
3 of ophthalmologists in the US = 19,216 (https://reurl.cc/bzL0aM)
of diabetes patients in the US = 30,987,900 (https://reurl.cc/YW2ypx)
(2) While DR screening may be conducted at the diabetes clinics, diabetes specialists are not trained to interpret retinal images. Therefore, diabetes specialists may not have the expertise or confidence to interpret retinal images.
Why is the problem critically chosen?
The longer a person has diabetes, the more likely they will develop DR. Among them, 20 to 30% will experience visual loss, and DR is the main cause of legal blindness for people ages 20 to 65.
What technologies are applied to solve the problem?
VeriSee DR was developed using the state of the art deep learning (neural network) and feature enhancement technologies. The performance of VeriSee DR was 95% sensitivity, 90% specificity, and an overall 93% accuracy for referable DR. Such performance is comparable to ophthalmologists.
Any particular areas of improvement?
VeriSee DR can effectively assist diabetes specialists in interpreting retinal images for diabetic retinopathy and output the results within seconds. With the implementation of VeriSee DR, the entire DR screening patient journey is expected to be greatly optimized, allowing health care providers to screen more patients, effectively direct referable patients to ophthalmologists in time, and achieve better patient outcomes.
What have you learned and how would the technology advancement potentially or practically help?
We have learned that understanding the end users’ pain points and unmet needs is essential to the development of medical AI products. Unlike the tech industry, in the field of medical AI, we must also take the current regulatory and clinical limitations into account during the planning and development stage. Once the industry becomes more mature, medical AI has the potential to really transform the future of healthcare. For example, if an AI product can see what doctors cannot see, or predict diseases in advance, we can then take preventive measures to delay disease progression or intervene with treatments earlier. In the long run, AI has the potential to decrease overall medical expenditure and bring better patient outcomes.
Technology at Work
This Technology is from
