Internship Project @ Hanlun AI

This project implements a U-Net for binary segmentation of Covid-19 symptoms in chest axial CT scans. I trained the model using the COVID-19 CT segmentation dataset, which contains 100 radiologist-annotated CT images labeled with ground-glass opacities, consolidations, and pleural effusions.

Due to the limited dataset size, I used data augmentations to improve generalizability (random rotations (15°), width/height shifts (10%), and zooms (10%)). The model was trained in Keras with the RMSprop optimizer and achieved a Dice coefficient of 0.79.