CNN-Based Classification of Craniosynostosis Using 2D Distance Maps.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Artificial Intelligence in Medicine

Background:

• Craniosynostosis, premature fusion of infant skull sutures, often requires complex diagnostics.
• Current methods like CT scans and cephalometric measurements can be limited.
• Automated diagnosis using 3D photogrammetry remains an area for advancement.

Purpose of the Study:

• To develop an automated diagnostic method for craniosynostosis using 3D photogrammetry.
• To represent infant head shapes from 3D scans as 2D distance maps for analysis.
• To evaluate a convolutional neural network (CNN) classifier utilizing these distance maps.

Main Methods:

• Generation of 2D distance maps from 3D photogrammetric surface scans using ray casting.
• Implementation of a CNN-based classifier to analyze the 2D distance maps.
• Validation of the CNN classifier on a public synthetic dataset and a clinical dataset.

Main Results:

• Qualitative visualization of head shape differences in the generated distance maps.
• Achieved 100% accuracy on the synthetic dataset.
• Attained 98.86% accuracy on the clinical dataset, demonstrating high diagnostic performance.

Conclusions:

• 3D photogrammetry, when processed into 2D distance maps, offers significant diagnostic value for craniosynostosis.
• The proposed CNN-based approach enables accurate and automated diagnosis.
• Future work will focus on refining the mapping technique and expanding CNN application to other pathologies.