Automatic Assessment of Pectus Excavatum Severity From CT Images Using Deep Learning

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Thoracic Surgery

Background:

• Pectus excavatum (PE) is a common chest wall deformity.
• Current severity assessment relies on manual analysis of CT images, which is time-consuming and variable.
• Objective quantification of PE severity is crucial for treatment planning.

Purpose of the Study:

• To develop and validate a fully automatic framework for quantifying pectus excavatum severity using CT images.
• To compare the accuracy and reproducibility of the automated framework against manual assessment.
• To assess the framework's performance in terms of intra-patient variability.

Main Methods:

• A novel framework employing heatmap regression networks (Unet++ architecture) for keypoint detection and measurements.
• Automatic identification of the sternum's greatest depression point and 8 relevant anatomical keypoints.
• Geometric regularization and extraction of Haller, correction, and asymmetry indices from CT scans.

Main Results:

• The automated framework demonstrated good agreement with manual measurements, with low mean relative absolute errors (4.41% for Haller, 5.22% for correction, 1.86% for asymmetry).
• Limits of agreement were comparable to inter-observer variability.
• The framework showed superior reproducibility in intra-patient analysis compared to expert manual assessment.

Conclusions:

• The developed framework provides an accurate, automatic, and reproducible method for quantifying PE severity from CT images.
• This automated approach has the potential to enhance clinical workflow and diagnostic consistency.
• The system's performance supports its feasibility for clinical application in pectus excavatum assessment.