Cataract disease is the world's leading cause of blindness with a majority of the world's population affected by it; however, it is easily treatable with surgery.
In Mexico, of the 16 million people who suffer from some form of vision loss, 63% of the population (~10 million people) are blind due to cataract disease.
There are four obstacles to receiving cataract surgery:
-
Lack of awareness
-
Bad service
-
Cost
-
Distance for treatment
63% of the Mexican population has a smartphone, so a smartphone application would be the best mode to reach our audience.
Transfer Learning & ResNet-50
Due to the relatively small size of our dataset (~2400 images), transfer learning utilizes pre-trained deep learning models to train on a new dataset. This ML approach is commonly used in medical image applications (Zhou et al., 2020). Transfer learning works best when there is scarce medical data and achieves better performance with significantly reduced training time and computational costs.
Of various pre-trained models, we found ResNet-50 to provide the most optimal results. ResNet-50 is a 50 layer convolutional neural network from the ImageNet Database.
