Developing A Semi‑Automatic Parametrization Algorithm For Preoperative Ct-Based Shape Determination Of An Ear Canal Wall Implant For Sanitizing Ear Surgery

INTRODUCTION: Cholesteatoma is the undesired accumulation of keratinised epithelial cells and keratinous debris in the middle ear cavity. Mastoidectomy is the gold standard for treating cholesteatoma. The three-dimensional anatomy of the temporal bone is complex, with many critical structures at-risk. Also, a clear difference between preoperative CT-scans (Figure 1) and microscopic views durante operationem (Figure 2) exists. This difference makes shaping the ear canal wall implant design hard. This research proposes an algorithm that uses the air‑containing regions of the ear. The algorithm guides implant design by semi-automatically parametrizing ear canal shape and define dimensions using preoperative CT-scans in cholesteatoma patients that qualify for sanitizing ear surgery.

METHODS: 92 clinical petrosum CT-scans from MST patients with suspected cholesteatoma were obtained with institutional board approval (PaNaMa id:1320). An in-house unified deep learning algorithm that combined multiple semi-automatic temporal bone segmentation methods parsed clinical CT-petrosum scans into anatomical landmark objects, such as the external auditory canal and facial nerve. Thereafter, gradient descent optimization was performed to fit the ear canal. The ear canal was modelled as an S-shaped, cylinder-like object parameterised by twist, writhe, translation, rotation, length, and radius. These3D-objects were refined using other segmented critical structures to respect critical structures in proximity of the middle ear cavities. Patient-specific implants will be compared with manually segmented implants using DICE-score and Hausdorff distance. Additionally, ENT-surgeons will judge the ear canal wall implants on correctness by completing a semi-structured evaluative questionnaire, which were analysed using descriptive statistics and intraclass correlation coefficients (ICC).

RESULTS: 92 CT-scans were segmented with the unified deep learning-based algorithm and planned to be converted into 92 3D-shapes resembling ear canal wall implants. These 3D-shapes will be compared against 5 manual shaped counterparts. Results of the ear canal wall implants and ICC are expected in March 2026.

DISCUSSION: Previous research showed that manually shaping ear canal wall implants in 2D CT-slices or static 3D models is difficult and time-consuming. Commercial ear canal wall implants are non-existing. The use of virtual planning software models may provide insights on how to manually shape ear canal wall implants for sanitizing ear surgery in the near future.

CONCLUSION: 92 3D-shapes of ear canal wall implants were constructed by applying an in-house developed semi-automatic parametrization algorithm on preoperative CT-scans for sanitizing ear surgery.