Robert Wood Johnson Hospital - Echocardiogram Detection with AI

Under Dr. Naveena Yanamala, a former Carnegie Mellon University research professor, I served as a Data Science Intern.

My project involved developing a standardized image-processing pipeline to normalize variable ultrasound datasets.

The project identified the most effective standardization techniques, significantly improving dataset quality and enhancing the precision and recall of a CNN model for clinical detection.

RWJ Illustration 1
RWJ Illustration 2
RWJ Illustration 3

My Work

My project focused on developing a standardized image-processing pipeline to normalize variable ultrasound datasets. By leveraging Radiomics, a quantitative method of analyzing medical images via feature extraction, I was able to extract numerical information out of images and develop my own datasets. Each dataset represented a unique image-processing technique, and for each I did statistical analysis, using Python and supporting libraries, to test for similarity in feature values across different hospitals. This study helped pick the most effective image standardizing technique that significantly improved the quality of ultrasound datasets from various hospitals, in turn, enhancing the precision and recall of their CNN model for clinical detections. The new model, trained on a standardized dataset, doubled its accuracy compared to the model that was trained on a non-standardized dataset, showing promise for future use in echocardiographic imaging and disease interpretation.