A diagnostic strategy for pulmonary fat embolism based on routine H&E staining using computational pathology

Area of Science:

• Forensic Pathology
• Computational Pathology
• Digital Pathology
• Artificial Intelligence

Background:

• Pulmonary fat embolism (PFE) is a significant cause of death in trauma cases.
• Traditional PFE diagnosis relies on subjective methods and specialized stains (e.g., oil red O).
• There is a need for objective, quantitative, and accessible diagnostic methods for PFE.

Purpose of the Study:

• To develop and validate a computational pathology approach for precise quantification of fat emboli in lung tissue.
• To utilize deep learning algorithms on conventional hematoxylin-eosin (H&E) stained slides for PFE diagnosis.
• To establish an objective diagnostic threshold for fatal PFE.

Main Methods:

• Digital pathology was employed to analyze whole slide images.
• Deep learning algorithms were developed to identify and quantify fat droplet morphology in lung microvessels.
• AI-quantified results were compared with traditional methods (Falzi scoring, oil red O staining).

Main Results:

• Deep learning achieved high accuracy (AUC of 0.98) in identifying fat emboli morphology.
• AI-quantified fat globules correlated well with manual scoring using oil red O.
• A diagnostic threshold of 8.275% relative fat emboli quantity was established, yielding an AUC of 0.984 for fatal PFE diagnosis.

Conclusions:

• Computational pathology, using deep learning on H&E slides, provides a precise and objective method for diagnosing fatal PFE.
• This AI-driven approach surpasses traditional H&E staining and rivals special stains in accuracy.
• The method offers a potentially affordable, rapid, and highly accurate diagnostic solution for forensic pathology.