In collaboration with a small group of data science students, developed and trained a convolutional neural network to classify pathologies in a dataset of 112,120 chest X-rays. We created a multi-task weighted loss function to both allow for multiple predictions (pathologies) per image and to account for class imbalances. Given the potential for bias in biomedical AI techniques, we evaluated computational bias in our model across different demographic groups
Co-authors of project and code: Kejung Ying, Mahnum Shazad, and Sebastian Pereira