Development and validation of a pixel wise deep learning model to detect cataract on swept-source optical coherence tomography images