Empirical Framework for Efficient Voxel-to-Point Transformation in Medical Imaging
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Three-dimensional medical imaging is central to diagnostic decision-making and treatment planning, yet its dense voxel-based representation imposes significant computational and memory demands, restricting its use in portable or resource-limited clinical settings. To address this challenge, we introduce a voxel-to-point-cloud transformation framework that restructures volumetric CBCT data by retaining only anatomically and radiometrically meaningful structures, reducing the burden of processing large voxel grids.The proposed method converts voxel radiometric densities into a compact point-cloud representation, enriches each point with spatial–radiometric gradient features that correspond to true anatomical transitions, and applies spatial-radiometric guided downsampling to selectively preserve clinically relevant details. The framework is evaluated on mandibular canal segmentation, an anatomically critical task during implant planning and surgical risk assessment, yet challenging due to low contrast, narrow geometry, and pronounced class imbalance. Results on CBCT data show up to 87.5% memory reduction and an average processing time of 9.5 s per volume on a standard CPU, while maintaining structural interpretability and anatomical fidelity.These findings demonstrate that voxel-to-point-cloud conversion offers a computationally efficient and clinically meaningful alternative to conventional 3D processing, enabling fast, reliable analysis suitable for edge-class medical devices and consumergrade hardware.
Fields of Study & Subjects
Dental Radiography and Imaging
Medical Image Segmentation Techniques
3D Shape Modeling and Analysis
Publication details
- Journal / source
- International Conference on Consumer Electronics
- Pages
- 1–6
- Publication date
- 3 February 2026
- Language
- EN