Last January 2022, I was accepted as an intern at the Nara Institute of Science and Technology (NAIST) under the Computational Systems Biology Laboratory. During this internship, I was introduced to a project of the lab which aims to detect scoliosis on chest X-ray images via deep learning. The model created by the lab was based on the EfficientNet architecture.
I was able to create a model that outperformed the lab's model in terms of accuracy, precision, F1-score, and ROC-AUC score, with the lab's model edging out mine in recall. See the presentation deck for more info on this performance comparison. I trained my model via transfer learning on the CheXNet model with weights obtained from this Github repository.
This repository contains all the files that I used during this internship period. Unfortunately, due to privacy issues, I will not be sharing the dataset and the files for the models themselves. Only the notebooks and final presentation deck will be included in this repository.