Cranial Implant Design Using V-Net Based Region of Interest Reconstruction

Shashwat Pathak, Chitimireddy Sindhura, Rama Krishna Sai Subrahmanyam Gorthi, Degala Venkata Kiran, Subrahmanyam Gorthi

Abstract: Cranial implant design is a sophisticated time-intensive process performed by specialists uniquely for each patient using a set of standardized cranioplasty procedures. Automating the design of cranial implants for the ‘in-Operating-Room’ (in-OR) manufacturing pipeline is required to perform cranioplasty immediately after the primary surgery, thereby reducing the overall surgery time. In this manuscript, we propose an efficient cranial implant design workflow through a two-step approach in which we use two V-Net architectures, one to extract the region of cranial defect from the low-resolution skull and the other to reconstruct the cranial defect in the high-resolution skull. The extracted defective cranium is subtracted from the reconstructed cranium and is post-processed to obtain the fine implant. We further performed experiments to manufacture the cranial implant predicted by our proposed method through 3D printing, using titanium-aluminium (Ti6-Al4-V) alloy, a standard material used for medical prosthetics and implants. The proposed method is trained and evaluated on the data provided by the MICCAI 2021 AutoImplant Challenge. Our method performed well, giving a Dice Similarity Coefficient (DSC) of 0.90, border DSC of 0.95, and a 95-percentile of the Hausdorff Distance (HD95) of 2.02 mm over the test dataset (000.nrrd–109.nrrd).