Deep Learning has indeed become an important part of our life, it has gained considerable attention in the field of study and is getting explored with a good amount of enthusiasm. It's incredible power to detect complex patterns have led us to believe that it can be used for a task as difficult as Diabetic Retinopathy disease diagnosis. The idea of such a model being used by doctors for gaining certainty in their diagnosis inspired us to get started with it approximately 2.5 weeks back!
eyeTorch is a website which takes as input a zoomed in image of the patient's eye and detect the level of severity of the disease with 0 being perfectly fine and 4 being critical situation.
We decided to use AlexNet framework after referring to some amount of research papers, to classify images, after pre-processing them. This was followed by deploying the model, which we did using flask. The templates for the web pages were made purely out of HTML and CSS. The model, hence, was successfully run on our local machine after implementing the website we made.
There were quite a lot of challenges that we faced, which ranged from something as little as pre-processing a large amount of input images to linking our model with the flask server; from deciding the correct network architecture to hosting the website on quite a lot of hosting services available. I'm very glad to mention that we were able to resolve most of these challenges.
Both of us have been practicing our skills in the world of AI for quite some time now, and naturally we had never touched upon website building, but this project helped us into not only understanding how flask works but also into making a full fledged website of our own! Hence, this is an accomplishment that we are really proud of.
We gained a good amount of knowledge about Convolutional Neural Networks and while looking for the correct architecture, we also stumbled upon a few open source codes to solve this problem - such as using fastai models - which not only motivated us to understand their solutions but also look for improvements in their already existing solution. This also enhanced our debugging skills which are a must for every programmer.
Last but not the least, we would be planning to extend this solution to other eye defects as well, such as that of Cataract, Glaucoma, and Refractive Errors.
Log in or sign up for Devpost to join the conversation.