Machine Learning Engineer
Date:
Developed a Brain Tumor Segmentation Software using state-of-the-art deep learning models to detect different regions of tumor in 3D MRI scan of human brain.
- Developed SaaS model for brain tumor segmentation in 3D MRI scan of human brain using Machine Learning for radiologists
- Implemented various state of the art deep learning models including Mask R-CNN, Yolo and U-Net using keras and tensorflow; achieved dice score of 84%for full tumor segmentation
- Trained model on BraTS dataset containing 3D MRI scan images of human brain in multiple contrast sequences
- Increased dice score up to 76% on enhancing tumor and 74% on non-enhancing tumor region by designing and implementing a separate U-Net model and a custom cropping algorithm
- Added functionality of custom dilation and erosion of contour around the full tumor