CraniaMap: Joint Solution for Auto-Segmentation and Landmarking from CBCT Scans

Ther-Ai Llc, Kissimmee FL

Investigators

Abstract

SUMMARY / ABSTRACT: Cone beam computed tomography (CBCT) is revolutionizing medical diagnostics, offering critical 3D insights for orthodontics, precise craniomaxillofacial (CMF) procedures, and comprehensive reconstructive surgery plannings. Its pivotal role in implant planning and broad applications in neurosurgery and oncology highlight its versatility too. However, the challenge lies in managing the vast 3D data for accurate segmentation and landmark digitization (i.e., landmarking), which are crucial for treatment accuracy and personalized patient care. Our team at Ther-AI LLC is pioneering the development of the Crania software suite to tackle the complex challenges of CBCT scan analysis. Our goal in this SBIR Phase-I proposal is to develop, CraniaMap, the foundational module of the Crania software suite, that is engineered to dramatically enhance auto-segmentation and landmarking in CBCT scans. This initiative is driven by the need for precision and efficiency in orthodontic, CMF, and reconstructive surgery planning, where current artificial intelligence (AI) tools fall short. CraniaMap is designed to excel where others falter, providing a robust, comprehensive solution that promises to enhance both the speed and accuracy of these critical diagnostic processes. In Aim 1, we will use our FuseNet segmentation tool to outperform existing models, particularly in handling anatomical variations and pathologies. Our self-supervised learning approach, supported by solid preliminary data, aims to achieve at least a 90% accuracy, as measured by the area under the curve (AUC), even in severe cases, applied to 250 volumetric CBCT scans and the following tissues will be segmented: Cranial Bones (frontal bone, parietal bones, temporal bones, occipital bone, sphenoid bone, ethmoid bone), Facial Bones (maxilla, mandible, nasal bones, zygomatic bones, palatine bones, lacrimal bones, interior nasal concha, vomer), Orbit, Teeth, and Airways. This represents a significant leap forward in medical imaging technology. Aim 2, automatic landmarking, will harness the power of our advanced relational reasoning network (RRN) to pinpoint 105 essential anatomical landmarks with high precision. This innovative approach ensures accuracy even in complex cases, with our goal being to maintain less than 2 mm mean square error across 250 diverse CBCT scans. CraniaMap leads as the first integrated software to simultaneously offer segmentation and landmarking, and it can even provide landmarks without the need for segmentation maps. This innovation is supported by our years of experience in medical AI algorithms, and substantial preliminary data. CraniaMap will be a precursor for several clinical applications spanning from orthodontics, CMF surgery and general dentistry. We expect that CraniaMap to provide a standardized, efficient, and precise quantification landscape for cephalometric analysis in the short term, and diagnosis, surgery/treatment planning, and outcome prediction in the long term. Millions of patients who need CMF, reconstructive, or orthodontic operations will greatly benefit from CraniaMap.