Summary
During the Summer of 2023, I had the privilege of collaborating with Dr. Neal Bangerter and Dr. Pete Lally on a fascinating Undergraduate Research Opportunity (UROP) focused on Deep Learning applications for Magnetic Resonance Images (MRI). My task was to develop a Deep Learning model for automatic knee MRI image segmentation, while keeping computational resources at a manageable level. This was my first exposure into the field of Deep Learning and I spent a few weeks learning about the fundamentals before starting the research project. Over the next three weeks, I refined my model for knee MRI images, exploring recent research on advanced Deep Learning techniques. Despite challenges due to limited resources, I optimized my model’s hyperparameters, achieving an 80% average DICE coefficient, compared to 93% for more sophisticated techniques. The model struggled to distinguish between the medial and lateral planes, which I believe could be improved by transitioning to a 3D UNET model. This experience highlighted the differences between Deep Learning and biological neural networks, sparking my interest in computational neuroscience. I gained a deep appreciation for the complexities of Deep Learning, recognizing its ‘black box’ nature and its significant data requirements, especially in medical applications.