STUDY SPACE
Deep learning to detect left ventricular structural abnormalities in chest X-rays

Deep learning to detect left ventricular structural abnormalities in chest X-rays

2024 | Shreyas Bhave¹, Victor Rodriguez¹, Timothy Poterucha², Simukayi Mutasa³, Dwight Aberle³, Kathleen M. Capaccione³, Yibo Chen⁴, Belinda Dsouza³, Shifali Dumeer³, Jonathan Goldstein³, Aaron Hodes⁵, Jay Leb³, Matthew Lungren⁶, Mitchell Miller⁵, David Monoky⁵, Benjamin Navot³, Kapil Wattamwar⁷, Anoop Wattamwar⁵, Kevin Clerkin², David Ouyang⁸, Euan Ashley⁹, Veli K. Topkara², Mathew Maurer², Andrew J. Einstein²,³, Nir Uriel², Shunichi Homma², Allan Schwartz², Diego Jaramillo³, Adler J. Perotte¹,²*, and Pierre Elias¹,²*
This study investigates the use of deep learning to detect left ventricular (LV) structural abnormalities, specifically severe LV hypertrophy (SLVH) and dilated left ventricle (DLV), from chest X-rays (CXRs). The researchers developed a deep learning model that inputs a pre-processed CXR, age, and sex to output probabilities for SLVH, DLV, and a composite label indicating the presence of either structural abnormality. The model outperformed 15 board-certified radiologists in detecting these abnormalities, achieving higher sensitivity at the same specificity. The model's performance was consistent across different datasets from Columbia University Irving Medical Center and Stanford University Medical Center. Saliency maps revealed that the model is sensitive to the cardiac silhouette and areas around the left ventricle. The study demonstrates the potential of deep learning as a tool for early detection of LV structural abnormalities, which could lead to earlier diagnosis and improved outcomes for patients with heart failure.This study investigates the use of deep learning to detect left ventricular (LV) structural abnormalities, specifically severe LV hypertrophy (SLVH) and dilated left ventricle (DLV), from chest X-rays (CXRs). The researchers developed a deep learning model that inputs a pre-processed CXR, age, and sex to output probabilities for SLVH, DLV, and a composite label indicating the presence of either structural abnormality. The model outperformed 15 board-certified radiologists in detecting these abnormalities, achieving higher sensitivity at the same specificity. The model's performance was consistent across different datasets from Columbia University Irving Medical Center and Stanford University Medical Center. Saliency maps revealed that the model is sensitive to the cardiac silhouette and areas around the left ventricle. The study demonstrates the potential of deep learning as a tool for early detection of LV structural abnormalities, which could lead to earlier diagnosis and improved outcomes for patients with heart failure.
Terms of Service
Privacy Policy
Reach us at info@study.space