Multi-Kernel Gated Decoder Adapters for Robust Multi-Task Thyroid Ultrasound under Cross-Center Shift

Maziar Sabouri; Nourhan Bayasi; Arman Rahmim

cs.CVMar 9, 2026

Multi-Kernel Gated Decoder Adapters for Robust Multi-Task Thyroid Ultrasound under Cross-Center Shift

Authors: Maziar Sabouri, Nourhan Bayasi, Arman Rahmim

Organizations: Department of Astronomy and Physics, University of British Columbia, Vancouver, Canada · Department of Basic and Translational Research, BC Cancer Research Institute, Vancouver, Canada · Department of Radiology, University of British Columbia, Vancouver, Canada · School of Biomedical Engineering, University of British Columbia, Vancouver, Canada

Abstract

Thyroid ultrasound (US) automation couples two competing requirements: global, geometry-driven reasoning for nodule delineation and local, texture-driven reasoning for malignancy risk assessment. Under cross-center domain shift, these cues degrade asymmetrically, yet most multi-task pipelines rely on a single shared backbone, often inducing negative transfer. In this paper, we characterize this interference across CNN (ResNet34) and medical ViT (MedSAM) backbones, and observe a consistent trend: ViTs transfer geometric priors that benefit segmentation, whereas CNNs more reliably preserve texture cues for malignancy discrimination under strong shift and artifacts. Motivated by this failure mode, we propose a lightweight family of decoder-side adapters, the Multi-Kernel Gated Adapter (MKGA) and a residual variant (ResMKGA), which refine multi-scale skip features using complementary receptive fields and apply semantic, context-conditioned gating to suppress artifact-prone content before fusion. Across two US benchmarks, the proposed adapters improve cross-center robustness: they strengthen out-of-domain segmentation and, in the CNN setting, yield clear gains in clinical TI-RADS diagnostic accuracy compared to standard multi-task baselines. Code and models will be released.

Figures & tables

Figure 1: Overview of the proposed MKGA and ResMKGA decoders architecture.

	ThyroidXL (In-domain Test Set)								DDTI (External Test Set)
Model Architecture	Segmentation		TI-RADS			Position			Segmentation		TI-RADS
	Dice	IoU	Acc	F1	AUC	Acc	F1	AUC	Dice	IoU	Acc	F1	AUC
Convolutional Neural Networks (ResNet34)
ResNet34 (Frozen)	0.857	0.764	0.855	0.857	0.920	0.855	0.852	0.947	0.631	0.512	0.357	0.396	0.556
ResNet34 (Unfrozen)	0.861	0.772	0.872	0.873	0.931	0.864	0.863	0.953	0.590	0.466	0.406	0.466	0.577
ResNet34 + MKGA (Frozen)	0.869	0.784	0.873	0.873	0.915	0.870	0.868	0.952	0.671	0.551	0.430	0.489	0.600

Table 1: Segmentation and diagnostic (TI-RADS) performance evaluated on ThyroidXL and DDTI, and positioning on ThyroidXL only. Best in Bold .

	ThyroidXL (In-domain Test Set)								DDTI (External Test Set)
Model Architecture	Segmentation		TI-RADS			Position			Segmentation		TI-RADS
	Dice	IoU	Acc	F1	AUC	Acc	F1	AUC	Dice	IoU	Acc	F1	AUC
Convolutional Neural Networks (ResNet34)
ResNet34 (Unfrozen)	0.861	0.772	0.872	0.873	0.931	0.864	0.863	0.953	0.590	0.466	0.406	0.466	0.577
ResNet34 + PCGrad (Unfrozen)	0.859	0.768	0.872	0.874	0.934	0.873	0.872	0.959	0.642	0.517	0.594	0.653	0.633
ResNet34 + MKGA (Unfrozen)	0.862	0.775	0.872	0.872	0.922	0.875	0.870	0.956	0.659	0.533	0.632	0.683	0.642

Table 2: Impact of PCGrad on multi-task optimization. Best in Bold .

Figure 2: Qualitative segmentation comparison on ThyroidXL and DDTI datasets.

	ThyroidXL (In-domain Test Set)								DDTI (External Test Set)
Model Architecture	Segmentation		TI-RADS			Position			Segmentation		TI-RADS
	Dice	IoU	Acc	F1	AUC	Acc	F1	AUC	Dice	IoU	Acc	F1	AUC
Core MKGA Ablations (ResNet34 Backbone)
ResNet34 + MKGA (NoGate)	0.862	0.775	0.869	0.869	0.930	0.861	0.859	0.953	0.673	0.552	0.499	0.569	0.589
ResNet34 + MKGA (NoMulti)	0.855	0.768	0.862	0.863	0.922	0.874	0.871	0.955	0.629	0.509	0.359	0.402	0.565
ResNet34 + MKGA (K1_3)	0.868	0.780	0.874	0.874	0.924	0.872	0.870	0.953	0.652	0.527	0.479	0.547	0.591

Table 3: Ablation studies isolating the architectural components of the proposed MKGA and ResMKGA modules. Best in Bold .

Explore similar work

Representation-Level Adversarial Regularization for Clinically Aligned Multitask Thyroid Ultrasound Assessment
Mar 22, 2026Dina Salama, Mohamed Mahmoud, Nourhan Bayasi +2Multi-Task Learning Clinical Risk Prediction
University of British Columbia, Vancouver, Canada · University of Calgary, Calgary, Canada
ThyCLIPNet: A BiomedCLIP-Guided Lightweight Attention-Enhanced DeepLabV3+ Framework for Robust Thyroid Nodule Segmentation
Oct 3, 2026Tasnim Jahan, Md Easin Arafat, Swakkhar ShatabdaUltrasound Image Segmentation Medical Image Segmentation
Department of Computer Science and Engineering, United International University, United City, Madani Ave, Dhaka 1212, Bangladesh · Faculty of Informatics, Institute of Industry-Academia Innovation, Department of Data Science and Engineering, Eötvös Lor´and University, P´azm´any P´eter S´et´any 1/C, Budapest 1117, Hungary · Department of Computer Science and Engineering, BRAC University, Pragati Sarani, Merul Badda, Dhaka 1212, Bangladesh
AURORA: Adaptive Unified Representation for Robust Ultrasound Analysis
Mar 19, 2026Ufaq Khan, L. D. M. S. Sai Teja, Ayuba Shakiru +4Medical Image Analysis Ultrasound Imaging
MBZUAI, Abu Dhabi, UAE · NIT Silchar, India · Birmingham City University, UK

Mar 22, 2026J/K move · Enter open · S save

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

Dina Salama, Mohamed Mahmoud, Nourhan Bayasi, David Liu+1

University of British Columbia, Vancouver, Canada · University of Calgary, Calgary, Canada

Multi-Task Learning Clinical Risk Prediction Ultrasound Imaging Adversarial Representation Learning Ultrasound Image Segmentation

Open paper AskPDF arXiv

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 compete within the shared representation. To make this competition explicit and controllable, we introduce RLAR, a representation-level adversarial gradient regularizer. Rather than performing parameter-level gradient surgery, RLAR uses each task's normalized adversarial direction in latent space as a geometric probe of task sensitivity and penalizes excessive angular alignment between task-specific adversarial directions. On a public TI-RADS dataset, our clinically guided multitask model with RLAR consistently improves risk stratification while maintaining segmentation quality compared to single-task training and conventional multitask baselines. Code and pretrained models will be released.

Abstract

Explore similar work

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

ThyCLIPNet: A BiomedCLIP-Guided Lightweight Attention-Enhanced DeepLabV3+ Framework for Robust Thyroid Nodule Segmentation

AURORA: Adaptive Unified Representation for Robust Ultrasound Analysis