Normscan: open-source Python software to create average models from CT scans

• Published in: International journal for computer assisted radiology and surgery Vol. 20; no. 1; pp. 157 - 165
• Main Authors: Nahass, George R., Marques, Mitchell A., Bou Zeid, Naji, Zhao, Linping, Alkureishi, Lee W. T.
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.01.2025; Springer Nature B.V
• Subjects: Algorithms; Augmented reality; Bar codes; Birth defects; Computed tomography; Computer Imaging; Computer Science; Craniosynostoses - diagnostic imaging; Craniosynostoses - surgery; Health Informatics; Humans; Imaging; Imaging, Three-Dimensional - methods; Infant; Medical imaging; Medicine; Medicine & Public Health; Models, Anatomic; Open source software; Original Article; Pattern Recognition and Graphics; Printing, Three-Dimensional; Python; Radiology; Registration; Skull; Skull - diagnostic imaging; Software; Surgery; Three dimensional models; Three dimensional printing; Tomography, X-Ray Computed - methods; Vision; Visualization; Plastic surgery; Neurosurgery; Surgical planning; Average model generation; Craniosynostosis; Python
• ISSN: 1861-6429; 1861-6410; 1861-6429
• DOI: 10.1007/s11548-024-03185-0

Purpose Age-matched average 3D models facilitate both surgical planning and intraoperative guidance of cranial birth defects such as craniosynostosis. We aimed to develop an algorithm that accepts any number of CT scans as input and generates highly accurate, average models with minimal user input that are ready for 3D printing and clinical use. Methods Using a compiled database of ‘normal’ pediatric computed tomography (CT) scans, we report Normscan, an open-source platform built in Python that allows users to generate normative models of CT scans through user-defined landmarks. We use the basion, nasion, and left and right porions as anatomical landmarks for initial correspondence and then register the models using the iterative closest points algorithm before downstream averaging. Results Normscan is fast and easy to use via our user interface and also creates highly accurate average models of any number of input models. Additionally, it is highly repeatable, with coefficients of variance for the surface area and volume of the average model being less than 3% across ten independent trials. Average models can then be 3D printed and/or visualized in augmented reality. Conclusions Normscan provides an end-to-end pipeline for the creation of average models of skulls. These models can be used for the generation of databases of specific demographic anatomical models as well as for intraoperative guidance and surgical planning. While Normscan was designed for craniosynostosis repair, due to the modular nature of the algorithm, Normscan has many applications in other areas of surgical planning and research.