[2603.21095] Representation-Level Adversarial Regularization for Clinically Aligned Multitask Thyroid Ultrasound Assessment

Computer Science > Computer Vision and Pattern Recognition arXiv:2603.21095 (cs) [Submitted on 22 Mar 2026] Title:Representation-Level Adversarial Regularization for Clinically Aligned Multitask Thyroid Ultrasound Assessment Authors:Dina Salama, Mohamed Mahmoud, Nourhan Bayasi, David Liu, Ilker Hacihaliloglu View a PDF of the paper titled Representation-Level Adversarial Regularization for Clinically Aligned Multitask Thyroid Ultrasound Assessment, by Dina Salama and 4 other authors View PDF HTML (experimental) Abstract:Thyroid ultrasound is the first-line exam for assessing thyroid nodules and determining whether biopsy is warranted. In routine reporting, radiologists produce two coupled outputs: a nodule contour for measurement and a TI-RADS risk category based on sonographic criteria. Yet both contouring style and risk grading vary across readers, creating inconsistent supervision that can degrade standard learning pipelines. In this paper, we address this workflow with a clinically guided multitask framework that jointly predicts the nodule mask and TI-RADS category within a single model. To ground risk prediction in clinically meaningful evidence, we guide the classification embedding using a compact TI-RADS aligned radiomics target during training, while preserving complementary deep features for discriminative performance. However, under annotator variability, naive multitask optimization often fails not because the tasks are unrelated, but because their gradients c...