A deep-learning classifier identifies patients with clinical heart failure using whole-slide images of H&E tissue

Area of Science:

• Cardiology
• Pathology
• Artificial Intelligence

Background:

• Heart failure affects over 26 million people globally each year.
• Endomyocardial biopsy (EMB) is the gold-standard for diagnosing heart failure when the cause is unknown.
• Manual interpretation of EMB slides suffers from significant inter-rater variability.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) classifier for detecting clinical heart failure from H&E stained whole-slide images.
• To assess the CNN's performance against conventional methods and expert pathologists.

Main Methods:

• A CNN classifier was trained on H&E stained whole-slide images from 104 patients with heart failure.
• The CNN model was independently tested on images from 105 patients.
• Performance was evaluated using sensitivity, specificity, and comparison to expert pathologist diagnoses.

Main Results:

• The CNN achieved 99% sensitivity and 94% specificity in identifying heart failure or severe pathology on the independent test set.
• The CNN classifier outperformed conventional feature-engineering approaches.
• The deep learning model demonstrated superior performance compared to two expert pathologists, exceeding their accuracy by nearly 20%.

Conclusions:

• Deep learning analysis of endomyocardial biopsy images offers a highly accurate and reproducible method for detecting heart failure.
• This AI-driven approach has the potential to improve diagnostic consistency and predict cardiac outcomes.
• The study highlights the transformative potential of artificial intelligence in cardiovascular pathology and diagnostics.