Dr. Sunaina Kapoor MD, MPH is currently a post-doctoral fellow in the Eudowood Division of Pediatric Respiratory Sciences at the Johns Hopkins University School of Medicine and one of the Global Program in Respiratory Sciences fellows. She has background in public and global health, implementation science, and medical education. Her areas of research includes development, implementation and, utilization of digital stethoscopes and machine learning in resource-limited settings.
We developed an expert-annotated dataset utilizing simultaneous conventional stethoscope expert auscultation as the reference for AI algorithm training. Higher than anticipated misclassification of normal reference lung sounds as abnormal by the panel suggests digital stethoscope artifact and environmental noise are contributing to classification disagreement. To address this limitation, we are applying a noise-reduction filter to recordings before panel interpretation. This expert-annotated dataset will then train the AI algorithm before testing with non-expert recordings.
Source: Conference 2024Auscultation Lab utilizes spectrograms to provide a three-dimensional visual and auditory aid for detailed lung sound annotation and classification. This platform is an innovative tool for training lung sound listening panel physicians and thus advancing AI algorithm performance, and for teaching lung auscultation in medical education.
Source: Conference 2024