Chest X-Ray Medical Diagnosis and Brain Tumor Auto-Segmentation for MRI
April 2020 - May 2020
AI for Medicine Specialization Course, deeplearning.ai
In the first part of this project, I trained the top-layers of pre-trained DenseNet121 model to diagnose pathologies (Pneumonia, Edema, Cardiomegaly) in Chest X-rays of ChestX-ray8 dataset. To address class imbalance, which is usually very high in medical diagnosis data, a weighted cross-entropy loss is used to train the model. To increase the interpretability of model, a GradCAM’s technique is used to produce a heatmap highlighting the important regions in the image for predicting the pathological condition.
In the second part of the project, a 3D U-Net model is trained for Volumetric Segmentation of MRI Images (DICOM format) with Multi-Class Soft Dice Loss as the loss function using the data from the Decathlon 10 Challenge.
