Prior work in building machine learning predictors for healthcare has achieved strong performance on individual medical imaging tasks, but these models often lack versatility across different modalities and struggle with interpretability in clinical contexts. Through my research, I aim to develop a foundation model capable of clinical reasoning across multiple modalities, inspired by how doctors constantly interface with multiple forms of data (e.g. EEGs, vitals, MRI scans) to assess patient status. By embedding explainability as a core objective for the model’s performance, I will investigate how model scale correlates with reasoning ability across clinical scenarios. This research explores scalability and interpretability for healthcare models, two challenges vital for clinical adoption.

After taking 6.7930 Machine Learning for Healthcare last year, I was inspired to think about how machine learning may resolve inefficiencies or uncover undesirable biases in the healthcare system. Participating in the SuperUROP program gives me an opportunity to pursue formal research with that goal in mind. I am excited to dive headfirst into a challenging and fast-moving field and to grow as a researcher through the program’s mentorship.