Title: AutoFFS: Adversarial Deformations for Facial Feminization Surgery Planning

Abstract:

Facial feminization surgery (FFS) plays a pivotal role in gender affirmation for transgender and gender-diverse individuals, focusing on reshaping craniofacial structures to align with female morphology. However, traditional surgical planning depends heavily on subjective clinical evaluations, often lacking the quantitative and reproducible anatomical guidance necessary for precision. To address this gap, we introduce AutoFFS, an innovative data-driven framework that creates counterfactual skull morphologies using adversarial free-form deformations.

Our approach executes a deformation-based targeted adversarial attack on an ensemble of pre-trained binary sex classifiers, which have been trained to recognize sexual dimorphism. This process effectively transforms individual skull shapes toward the desired target sex. By generating these counterfactual morphologies, AutoFFS establishes a quantitative basis for preoperative FFS planning, offering significant advancements for a patient demographic that has historically been underrepresented in technical research.

We validate our methodology through several rigorous tests. First, we employ classifier-based evaluation to assess performance. Second, we introduce two new metrics, Morphological Fréchet Distance (MFD) and Morphological Kernel Distance (MKD), designed to evaluate the distributional alignment between generated and real populations. Finally, a human perceptual study confirms that the generated morphologies successfully exhibit characteristics associated with the target sex.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC