MIT EECS | Angle Undergraduate Research and Innovation Scholar
Designing a Ranking System for Pulmonary Edema Severity
- Natural Language and Speech Processing
The severity of pulmonary edema is diagnosed on a discrete scale of 0 to 3 using information from chest radiographs. Unfortunately, accurate grading of disease severity is a challenging task because the boundaries of the bins are not well defined and there is often overlap between different radiologist interpretations. Our research aims to use natural language processing (NLP) techniques on radiology reports to improve the accuracy of diagnosis. We plan to use NLP to extract coarse-grained labels for pulmonary edema severity and intra-patient pairwise comparisons from chest radiograph reports. If time permits, we would also like to investigate the feasibility of developing a continuous-value scale for grading the severity of pulmonary edema using the pairwise comparisons.
“Through SuperUROP, I hope to gain more hands-on experience with applied machine learning (ML). I have taken multiple ML courses and I am interested in healthcare and medicine, so I believe this is an exciting opportunity to learn how to leverage NLP to make sense of unstructured clinical data and use it to improve patient outcomes. I look forward to improving my research and communication skills and working on a successful project.”