Juneja Mamta, Singla Ishaan, Poddar Aditya, Pandey Nitin, Goel Aparna, Sudhir Agrima, Bhatia Pankhuri, Singh Gurzafar, Kharbanda Maanya, Kaur Amanpreet, Bhatia Ira, Gupta Vipin, Dhami Sukhdeep Singh, Reinwald Yvonne, Jindal Prashant, Breedon Philip
University Institute of Engineering and Technology, Panjab University, Chandigarh 160014, India.
Department of Neurosurgery, Government Medical College and Hospital, Sector 32, Chandigarh 160032, India.
Bioengineering (Basel). 2025 Feb 16;12(2):188. doi: 10.3390/bioengineering12020188.
Cranioplasty enables the restoration of cranial defects caused by traumatic injuries, brain tumour excisions, or decompressive craniectomies. Conventional methods rely on Computer-Aided Design (CAD) for implant design, which requires significant resources and expertise. Recent advancements in Artificial Intelligence (AI) have improved Computer-Aided Diagnostic systems for accurate and faster cranial reconstruction and implant generation procedures. However, these face inherent limitations, including the limited availability of diverse datasets covering different defect shapes spanning various locations, absence of a comprehensive pipeline integrating the preprocessing of medical images, cranial reconstruction, and implant generation, along with mechanical testing and validation. The proposed framework incorporates a robust preprocessing pipeline for easier processing of Computed Tomography (CT) images through data conversion, denoising, Connected Component Analysis (CCA), and image alignment. At its core is CRIGNet (Cranial Reconstruction and Implant Generation Network), a novel deep learning model rigorously trained on a diverse dataset of 2160 images, which was prepared by simulating cylindrical, cubical, spherical, and triangular prism-shaped defects across five skull regions, ensuring robustness in diagnosing a wide variety of defect patterns. CRIGNet achieved an exceptional reconstruction accuracy with a Dice Similarity Coefficient (DSC) of 0.99, Jaccard Similarity Coefficient (JSC) of 0.98, and Hausdorff distance (HD) of 4.63 mm. The generated implants showed superior geometric accuracy, load-bearing capacity, and gap-free fitment in the defected skull compared to CAD-generated implants. Also, this framework reduced the implant generation processing time from 40-45 min (CAD) to 25-30 s, suggesting its application for a faster turnaround time, enabling decisive clinical support systems.
颅骨成形术能够修复由外伤、脑肿瘤切除或减压性颅骨切除术引起的颅骨缺损。传统方法依靠计算机辅助设计（CAD）进行植入物设计，这需要大量资源和专业知识。人工智能（AI）的最新进展改进了计算机辅助诊断系统，用于更准确、快速的颅骨重建和植入物生成程序。然而，这些方法存在固有的局限性，包括涵盖不同位置各种缺损形状的多样化数据集有限，缺乏一个整合医学图像预处理、颅骨重建、植入物生成以及力学测试和验证的综合流程。所提出的框架包含一个强大的预处理流程，通过数据转换、去噪、连通分量分析（CCA）和图像对齐，更轻松地处理计算机断层扫描（CT）图像。其核心是CRIGNet（颅骨重建和植入物生成网络），这是一种新颖的深度学习模型，在由2160幅图像组成的多样化数据集上进行了严格训练，该数据集通过模拟五个颅骨区域的圆柱形、立方体、球形和三棱柱形缺损来制备，确保在诊断各种缺损模式时具有鲁棒性。CRIGNet实现了卓越的重建精度，骰子相似系数（DSC）为0.99，杰卡德相似系数（JSC）为0.98，豪斯多夫距离（HD）为4.63毫米。与CAD生成的植入物相比，生成的植入物在缺损颅骨中显示出更高的几何精度、承载能力和无间隙贴合度。此外，该框架将植入物生成处理时间从40 - 45分钟（CAD）减少到25 - 30秒，表明其可应用于更快的周转时间，实现决定性的临床支持系统。
