Detection of Inner Ear Malformations Based on Simple Anatomical Measurements: A Model Approach
Riccardo Di Micco, Angelika Illg, Omar Abu-Fares, Levent Sennaroglu, Thomas Lenarz, Daniel SchurzigObjectives:
Accurate identification of the specific inner ear malformations can assist the otologist in anticipating surgical challenges and potentially optimizing postoperative hearing outcomes. In this study, we explored the feasibility of applying similar cochlear morphology principles, that is, ones based on easily quantifiable measurements of the basal turn, to differentiate normal lateral wall geometry from malformed variants.
Design:
We retrospectively collected 60 patients who underwent cochlear implantation in our center between 2005 and 2023 in the presence of a preoperatively recognized inner ear malformation. Of the 120 analyzed cochleae, 111 were eligible for segmentation, which included 8 cochlear hypoplasia type II, 15 cochlear hypoplasia type III, 2 cochlear hypoplasia type V, 38 Incomplete partition type I, 44 Incomplete partition type II, and 4 incomplete partition type III. A control cohort of 141 normal cochleae was selected. Using manual segmentation of the cochlea on preoperative cone beam computed tomography scans, three-dimensional lateral wall spirals were obtained. The spirals were then used to compute simple anatomical measures of the basal turn, namely the cochlear diameter A and width B, the basal turn length computed based on A and B using the elliptic circular approximation approach, the ratio B/A and the B ratio, and cochlear height (H). Initially, two-sided Mann–Whitney–Wilcoxon tests were conducted to derive statistical differences in the aforementioned geometrical parameters between normal and malformed anatomies. Second, logistic regression analyses were performed to define whether the derived geometrical parameters may be used to predict if a specific cochlear morphology is normal or malformed, and to subsequently investigate if such a model may even be capable of distinguishing between different malformation types.
Results:
Four geometrical basal‑turn parameters easily assessable in clinical imaging, namely the basal turn length along the lateral wall, cochlear height (H), B/A ratio, and B‑ratio, provide enough information to reliably distinguish normal cochleae from malformed variants. The binary logistic model achieved 94% overall accuracy with precision and recall ≥0.92 across classes, with cochlear height H emerging as the dominant predictor. In our models, H and B b /B were the strongest discriminants for normal versus malformed anatomies and for incomplete partitions versus cochlear hypoplasia, respectively. Moving beyond binary discrimination, the three‑class model (normal versus IP versus CH) retained 90% accuracy. In summary, the models reliably recognize normal versus malformed, incomplete partition versus cochlear hypoplasia, and CHIII among individual subtypes due to reduced height. They are less reliable for IPI versus IPII differentiation, where basal‑turn measures alone appear insufficient, and IPIII, CHII, CHIV, where very small sample sizes depress recall.
Conclusions:
Simple, basal‑turn measurements of the lateral wall were demonstrated to provide a fast, interpretable, and accurate means to detect cochlear malformations on clinical imaging and to differentiate IP from CH. The approach reliably flags abnormal cases, offers actionable preoperative information for patient‑tailored implantation, and identifies domains where additional features or larger cohorts are needed (IPI versus IPII; rare subtypes). Such a model could lay the groundwork for future automated recognition of cochlear malformations during preoperative planning.