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

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Early identification of cardiac structural abnormalities is crucial for managing heart failure.
• Chest X-rays (CXRs) are common diagnostic tools offering potential for scalable screening.
• Deep learning (DL) can analyze CXRs for signs of Stage B heart failure or worse.

Purpose of the Study:

• Develop a DL model to identify severe left ventricular hypertrophy (SLVH) and dilated left ventricle (DLV) using CXRs.
• Evaluate the model's performance in detecting these cardiac structural abnormalities.

Main Methods:

• Utilized 71,589 CXRs from 24,689 patients, linked to echocardiogram data.
• Extracted labels for SLVH, DLV, and a composite outcome from echocardiograms.
• Trained and validated a DL model, assessing performance via AUROC and external validation.

Main Results:

• The DL model achieved an AUROC of 0.79 for SLVH, 0.80 for DLV, and 0.80 for the composite outcome.
• Performance was consistent on an external dataset.
• The model surpassed individual radiologists and demonstrated higher sensitivity than the radiologist consensus.

Conclusions:

• DL analysis of CXRs is effective for detecting LV hypertrophy and dilation.
• This approach shows promise for early identification of patients with these cardiac conditions.
• A public dataset of 71,589 CXRs with echocardiographic labels is now available to foster research.