Automated Deep Learning Segmentation of Cardiac Inflammatory FDG PET.

Area of Science:

• Cardiology
• Nuclear Medicine
• Artificial Intelligence

Background:

• Fluorodeoxyglucose positron emission tomography (FDG PET) is crucial for diagnosing cardiac sarcoidosis (CS).
• Accurate myocardial segmentation and image reorientation are vital for CS FDG PET analysis but are challenging and time-consuming.
• Current methods for image processing in CS FDG PET are labor-intensive and can lack consistency.

Approach:

• Developed a 3D U-Net deep learning (DL) algorithm for automated myocardial segmentation in CS FDG PET.
• Trained the DL model on 316 patient FDG PET scans and left ventricular contours from perfusion datasets.
• Compared DL segmentation with standard methods using qualitative analysis and quantitative metrics on a 50-patient test set.

Key Points:

• DL segmentation significantly enhanced clinical readability in over 90% of cases compared to standard software.
• The DL tool demonstrated performance comparable to a trained technologist, outperforming standard segmentation for key metrics.
• DL segmentation improved left ventricle displacement, angulation, and SUVmax correlation, surpassing standard methods.

Conclusions:

• The DL-based automated segmentation tool offers a substantial improvement for processing CS FDG PET scans.
• This tool promises to enhance clinical workflow, accelerate practice, and improve the consistency and quality of CS diagnosis.
• Further validation with diverse datasets is recommended to expand the applicability of this automated segmentation tool.