I was an ML researcher at UNC's Biomedical Image Analysis Group under Dr. Marc Niethammer. I extended a unified multi-modal BERT-transformer model in PyTorch for osteoarthritis prognosis using tabular, X-ray, and MRI data—evaluating outcomes across diverse socioeconomic groups in medicine. I also trained computer vision models for patient pain prediction that reduced disparity gaps by 30% versus objective X-ray based pain ratings.
I spent a year doing research under Dr. Martin Styner at UNC focused on unsupervised anomaly detection in 3D brain MRIs to identify autism via differences in brain structure. Worked on building/optimizing a U-Net style network that learns the gradient of distribution from healthy scans and determines anomalies by maintaining spatial awareness.
I performed exploratory research under Dr. Jorge Silva focused on a variety of U-Net and Swin-UNET segmentation methods for multimodal brain imaging. For this, I built PyTorch implementations of the models along with Encoder-Decoder architectures. Trained on BRaTS 2021. Additional work focused on building various model architectures for Heart Disease Prediction.
I was 1 of 82 students worldwide at the Research Science Institute, hosted at MIT. My work here under Dr. Anna Kosovicheva was focused on building a methodology quantifying how image features influence human object localization. This work was featured in the Vision Sciences Society in 2020.
I was a multi-time ISEF finalist and award winner for my projects AzureWare and AzureVoice; both of which focused on building algorithms and ML methods for Parkinson's disease diagnosis and monitoring. I won $10,000 at the Intel International Science and Engineering Fair over the course of work.