Sex estimation from the three-dimensional nasal septum on computed tomography: an allometric, size-based signal in 1,373 adults

Abstract Sex estimation from computed tomography (CT) of the head has relied on manual measurements of the skull, paranasal sinuses, and nasal cavity, with modest accuracy and small samples. The nasal septum, the midline osteocartilaginous partition of the nasal cavity, is present on every head CT but has never been evaluated as a sex estimator, and whether any sex signal in the septum reflects size or shape determines how such an estimator would transfer between populations. We tested two hypotheses in 1,373 adults (394 men, 979 women) with radiologically normal sinus CT: that septal morphometry classifies sex at least as accurately as established cranial methods, and that the discriminating information is allometric size rather than shape. The septum was segmented automatically with a validated convolutional network (held-out Dice 0.983) and 17 morphometric variables were derived from each segmentation. Septal volume was 34% larger in men (16,852 versus 12,564 mm³; Cohen's d 2.33, 95% CI 2.18 to 2.47). Under five-fold cross-validation, sex was classified with an area under the curve (AUC) of 0.95 by random forest and 0.96 by logistic regression on 17 variables; volume alone gave 0.94. After allometric adjustment no shape variable differed by more than d = 0.43, and shape variables alone gave an AUC of 0.58 to 0.63. Both hypotheses were supported: the septum is a strong, fully automatable sex estimator for CT, and its signal is size. Like other size-based skeletal estimators it will require population-specific calibration, which the automated pipeline makes practical.