Cochlear Diameter Matches Complex Shape Models for Routine CT Sex Estimation

A semi-automated analysis of routine clinical computed tomography scans can extract sex-related anatomical information from the human cochlea, but adding complex three-dimensional curvature shape descriptors offers no detectable advantage over measuring simple cochlear size alone. In an internal validation study testing multiple predictive configurations, a classifier relying entirely on scalar cochlear diameter matched the predictive performance of models that combined dimensional metrics with intricate geometric curvature and torsion descriptors. The investigation demonstrated that while cochlear geometry reflects biological sex differences, advanced shape parameters fail to provide incremental diagnostic value beyond basic linear measurements.

Sex estimation remains a foundational step in forensic anthropology and bioarchaeology, where investigators routinely work to reconstruct biological profiles from recovered human remains. Standard identification and profiling methodologies typically depend on pelvic, cranial, or postcranial skeletal elements, but these bones are frequently absent, fragmented, or degraded in forensic and archaeological recovery contexts. By comparison, the bony labyrinth of the inner ear is securely encased within the dense petrous portion of the temporal bone, granting it exceptional taphonomic resistance against environmental degradation, physical trauma, and postmortem decay. Furthermore, because the inner ear completes its morphological development and maturation early in ontogeny, its anatomical structures have attracted increasing interest as potential indicators of biological sex.

Although previous anthropological literature indicated that cochlear morphology carries a sex-specific anatomical signal, earlier investigative frameworks faced substantial technical hurdles that constrained their practical adoption. Most prior analytical approaches relied upon labor-intensive manual segmentation, high-resolution micro-computed tomography scanning, or complex mathematical shape-analysis pipelines that are inaccessible in routine forensic casework or standard clinical settings. To overcome these practical barriers, the researchers designed an investigation to evaluate whether semi-automated segmentation using routine clinical computed tomography could perform sex estimation, while formally quantifying whether three-dimensional shape descriptors yield predictive value beyond standard size metrics.

The research team evaluated a dataset comprising 139 cochleae collected from 87 adult individuals examined with routine clinical computed tomography scans. To ensure adequate spatial resolution for morphological modeling, the protocol utilized diagnostic scans with a slice thickness of 0.6 millimeters or less. Using OTOPLAN clinical software, the researchers performed semi-automated extraction of the three-dimensional lateral wall path of each cochlear duct. Following path extraction, the coordinate trajectories underwent normalization and spatial resampling, enabling the calculation of scalar cochlear measurements alongside curvature-based and torsion-based geometric descriptors to train a supervised machine learning classifier.

To rigorously evaluate model performance while preventing data leakage and optimistic bias, the investigators implemented a nested cross-validation framework where feature selection, algorithmic preprocessing choices, and hyperparameter tuning were conducted entirely separately from final performance testing. Hyperparameters and preprocessing steps were systematically optimized using Bayesian optimization algorithms within the internal cross-validation folds. The experimental workflow systematically compared six distinct feature configurations across five repeated random data partitions, with all partitioning executed strictly at the patient level so that cochleae from the same individual never appeared simultaneously in training and testing splits.

Across the nested cross-validation evaluations, the top-performing classifier configurations reached an area under the receiver operating characteristic curve of approximately 0.71. Strikingly, a supervised model relying on cochlear diameter as its sole predictive variable performed indistinguishably from models integrating both size and multidimensional shape descriptors, achieving an area under the curve of 0.71 with a 95 percent confidence interval spanning from 0.62 to 0.80. This single-variable diameter classifier also delivered the highest overall balanced accuracy at 0.70, accompanied by an empirical sensitivity of 0.71 and a specificity of 0.70 across the evaluated patient cohort.

In contrast, classifier configurations restricted solely to curvature-based and torsion-based shape descriptors demonstrated markedly inferior discriminatory power, yielding a median area under the receiver operating characteristic curve of only 0.59. The lower confidence bounds for these purely shape-based configurations ranged between 0.50 and 0.51, placing their performance at or just marginally above chance level. Furthermore, when the researchers added mathematical curvature metrics directly to the cochlear diameter baseline, the median area under the curve shifted by an increment of only +0.011, with a wide 95 percent confidence interval extending from -0.064 to +0.107, confirming the absence of a statistically detectable predictive gain.

The study authors concluded that although sex-related anatomical information can be successfully recovered from cochlear geometry using semi-automated clinical computed tomography workflows, classification accuracy remains moderate. Because the semi-automated OTOPLAN pipeline is non-destructive, scalable, and fully compatible with existing clinical imaging protocols, the methodology offers a potential adjunct for bioarchaeology and forensic anthropology when conventional diagnostic skeletal elements are missing or compromised. However, the authors explicitly cautioned that this approach is not an identification method and has not been validated as a stand-alone forensic tool, emphasizing that its diagnostic performance was established solely through internal cross-validation on in vivo clinical computed tomography scans.

Subject of Research: Semi-automated sex estimation from cochlear CT measurements

Article Title: Semi-automated sex estimation from the cochlea on routine clinical CT: limited incremental value of curvature-based shape descriptors

Article References: Vishnumurthy, P., Daval, M., Helpard, L., Ouzzine, I., Haddad, R., Ivanovic, N., Jourdes, V., Michel, J., & Gargula, S. (2026). Semi-automated sex estimation from the cochlea on routine clinical CT: limited incremental value of curvature-based shape descriptors. International Journal of Legal Medicine. https://doi.org/10.1007/s00414-026-04063-x

Image Credits: AI Generated

DOI: 10.1007/s00414-026-04063-x

Keywords: cochlea, sex estimation, clinical CT, OTOPLAN, bony labyrinth, petrous bone, forensic anthropology, bioarchaeology, taphonomic resistance, cochlear diameter, shape descriptors, curvature, torsion, supervised classifier, nested cross-validation, Bayesian optimization, AUC-ROC, lateral wall path