Background: Accurate identification and isolation of the nasopalatine neurovascular bundle are essential for reducing iatrogenic complications during maxillofacial surgical procedures. Manual delineation of this structure on cone beam computed tomography volumes remains challenging because of small anatomical dimensions, interindividual variability, and complex morphology. Automated three dimensional segmentation methods support precise surgical planning, improve intraoperative navigation, and facilitate integration with robotic assisted surgical systems, thereby contributing to improved patient outcomes in high risk maxillofacial interventions. Methods: A retrospective analysis was performed on 297 anonymised cone beam computed tomography scans acquired from diagnostic centers and dental clinics across India. DICOM image series and corresponding segmentation masks were converted to NIfTI format to ensure model compatibility. Labeled and unlabeled datasets were curated, followed by extraction of morphometric shape features and clustering based analysis for anatomical characterization. An nnU Net based deep learning model was trained using a three dimensional full resolution configuration with isotropic voxel spacing of 0.15 mm over 750 epochs, incorporating dynamic learning rate scheduling to optimize segmentation performance. Results and Discussion: The nnU Net model was trained on 237 volumes and validated on a holdout dataset of 60 volumes. A peak pseudo Dice score of 0.7942 was achieved at epoch 747, with a mean validation Dice score of 0.6879. Segmentation accuracy improved progressively throughout training, demonstrating stable convergence and effective generalization despite the anatomical complexity of the nasopalatine neurovascular bundle. Integration of morphometric feature analysis supported consistent and shape aware segmentation across heterogeneous cone beam computed tomography datasets. Conclusions: The proposed nnU Net based framework provides an automated, objective, and scalable solution for segmentation of the nasopalatine neurovascular bundle from cone beam computed tomography volumes. Reduced dependence on manual segmentation and improved anatomical precision support applications in maxillofacial surgery, dental implantology, robotic assisted interventions, and forensic analysis. The findings highlight the potential of deep learning driven segmentation to enhance surgical navigation, reduce neurovascular injury risk, and improve overall clinical workflow efficiency.

Aims Late diagnosis of Oral Squamous Cell Carcinoma (OSCC) contributes significantly to its high global mortality rate, with over 50\% of cases detected at advanced stages and a 5-year survival rate below 50\% according to WHO statistics. This study aims to improve early detection of OSCC by developing a multimodal deep learning framework that integrates clinical, radiological, and histopathological images using a weighted ensemble of DenseNet-121 convolutional neural networks (CNNs). Material and Methods A retrospective study was conducted using publicly available datasets representing three distinct medical imaging modalities. Each modality-specific dataset was used to train a DenseNet-121 CNN via transfer learning. Augmentation and modality-specific preprocessing were applied to increase robustness. Predictions were fused using a validation-weighted ensemble strategy. Evaluation was performed using accuracy, precision, recall, F1-score. Results High validation accuracy was achieved for radiological (100\%) and histopathological (95.12\%) modalities, with clinical images performing lower (63.10\%) due to visual heterogeneity. The ensemble model demonstrated improved diagnostic robustness with an overall accuracy of 84.58\% on a multimodal validation dataset of 55 samples. Conclusion The multimodal ensemble framework bridges gaps in the current diagnostic workflow by offering a non-invasive, AI-assisted triage tool that enhances early identification of high-risk lesions. It supports clinicians in decision-making, aligning with global oncology guidelines to reduce diagnostic delays and improve patient outcomes.

Orthodontic treatment planning often involves visualizing skeletal profile modifications, with CONVEX and CONCAVE morphologies frequently requiring geometric corrections toward more balanced configurations. However, existing GAN-based latent editing methods achieve insufficient skeletal control for clinically significant Pogonion displacement. A hybrid geometric–generative framework is introduced thatdecouples deterministic skeletal modification from photorealistic synthesis to address the manipulability limits of StyleGAN-based editing. Radial basis function warping with adaptive amplification (1.8–4.5×) produces controlled mandibular advancement or setback, while StyleGAN2 refinement with 100× landmark weighting restores anatomical realism without compromising structural accuracy. Multi-task regularization combining diversity loss with landmark-consistency constraints mitigates dataset imbalance and enables balanced generation across profile classes (approximately 31% each), countering the predominance of CONVEX cases (63%). The system achieves 10–15 pixel Pogonion displacement with < 3-pixel landmark drift (0.9 mm average), providing reliable and anatomically stable correction. Interactive profile morphing animations provide explainable visualization of transformation trajectories through transparent, frame-by-frame display, enabling clinically interpretable exploration of treatment outcomes across the full CONVEX-to-CONCAVE spectrum. Integrated Vision Transformer classifiers and regressors automate the Visual Treatment Objective workflow, reaching 89% profile-classification accuracy and 95.6% landmark-regression accuracy within a 2-pixel tolerance. Ablation analysis confirms that standard latent editing produces only1.4 pixels of Pogonion movement, demonstrating the necessity of explicit geometric control for skeletal transformations where skeletal landmarks define individual patient morphology.