Mamta Juneja

Maxillofacial defects impair facial aesthetics and oral function, arising from trauma, tumor resection, or congenital anomalies; however, reconstruction using Computer-Aided Design (CAD) and autologous grafts remains complex and time-intensive, and is associated with donor-site morbidity. Although deep learning (DL) has advanced automated reconstruction, existing models often address isolated tasks, lack integrated multi-scale feature learning, and rely on small datasets. This study proposes the Maxillofacial Implant-generation Network (MaxI-Net), a fast, resource-efficient three-dimensional DL framework for end-to-end maxillofacial defect reconstruction and patient-specific implant generation, with a completion step of cavity filling within the assembly. The model employs a 3D encoder–bottleneck-decoder architecture integrating hybrid dilated convolutions, residual connections, squeeze-and-excitation (SE) blocks, and 3D Convolutional Block Attention Modules (CBAM) with multi-scale feature fusion. It was trained on 921 Cone Beam-Computed Tomography (CBCT) scans, augmented to 11,973 maxillary defect pairs, using Dice loss and Adam optimisation with Automatic Mixed Precision, and benchmarked against UNet, UNETR, SegResNet, and SwinUNETR. MaxI-Net achieved the following: superior Dice Similarity Coefficient (DSC) = 0.778; 95th percentile Hausdorff Distance (HD95) = 3.453 mm; DSC Standard Deviation (SD) = 0.094; 95% confidence interval (CI) for mean DSC: 0.775–0.782). It was statistically validated against all competing architectures via pairwise Wilcoxon signed-rank tests, with significant DSC improvements confirmed across all comparators (p < 0.001) and rank-biserial effect sizes ranging from r = 0.250 against the closest competitor SegResNet* with high efficiency (0.06 s/volume; 9.6 min/epoch). Internal cavity filling of the generated implants was performed as a brief manual post-processing step in Autodesk Fusion 360 prior to biomechanical validation. Biomechanical validation using a finite element analysis (FEA) of polyether–ether–ketone (PEEK) implants (~26.53 g) showed 41% stress reduction under physiological loads (100–400 N), predicting a ~9.2-year lifespan.

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.

Cranial reconstruction using implants is critical for protecting intracranial structures and restoring cerebral hemodynamics in cases of cranial defects caused by accidents, diseases or cancer. Patient-specific implants (PSIs) made from materials such as polyether-ether-ketone (PEEK), are required to be lightweight, high in strength and capable of mimicking the natural bone structure. Effective fastening mechanisms using the required number of fixture plates are essential for seamless integration between the PSI and the cavity of a defected skull for successful cranial reconstruction. This study explores the optimal number and shape of fixture plates required to join a Skull-PSI assembly, such that the overall weight of the PSI remains minimal, and to ensure that these assemblies do not fail when subjected to heavy external loads of 950 N. PEEK material was used for PSI, natural bone for the defected skull and Titanium Alloy (Ti-6Al-4 V) for the fixture plates. Conventional straight shaped fixture plates often require manual bending for correct fitment on the Skull-PSI curved surface, which increases a surgeon's time and effort. Curved shaped fixture plates were designed, to save on this time and effort and enhance the contact surface area with the Skull-PSI surface. Four, three and two numbered, straight and curved shaped fixture plates were investigated using Finite Element Analysis (FEA) techniques. Three numbered, curved shaped fixture plates were found to be optimal, to generate a 7-gram lightweight PSI that could successfully sustain external loads up-to 950 N without failure. Ultimately, these design improvements would benefit both patient and surgeon in aspects of surgery time and patient comfort.