Clinician‐Guided Deep Learning Segmentation of Skull Base Pneumatization on Computed Tomography: Radiomic Analysis of Temporal Bone and Sphenoid Sinus Volumes

Skull base pneumatization is anatomically variable and clinically relevant for temporal bone and transsphenoidal surgical corridors, but manual volumetric segmentation is time-consuming. This retrospective pilot study evaluated a clinician-guided deep learning workflow for computed tomography segmentation of temporal bone/mastoid and sphenoid sinus pneumatization. Bone-window computed tomography (CT) datasets were curated, converted to NIfTI format, and segmented using 3D Slicer, MONAI Label, and a three-dimensional SegResNet-based model. The mastoid workflow used side-specific temporal bone crops, while the sphenoid workflow used full-head CT volumes. The final mastoid development dataset included 122 side-cases, with an independent 28 side-case expert-validation cohort. The sphenoid arm included 117 expert-submitted development labels and a separate 17-case expert-validation cohort. The mastoid model achieved strong expert agreement, with mean Dice 0.9691 and median Dice 0.9794 on the independent validation set. The sphenoid workflow achieved a best internal validation Dice of 0.8750, while expert comparison of saved automatic masks showed mean Dice 0.9426 and median Dice 0.9408. Exploratory mask analysis enabled volumetric, densitometric, and extension-pattern assessment of pneumatized skull base compartments. This research supports clinician-guided deep learning as a feasible approach for reproducible CT-based skull base pneumatization segmentation and radiomic analysis.