Prototype-Enhanced Multi-View Learning for Thyroid Nodule Ultrasound Classification

Yangmei Chen; Zhongyuan Zhang; Xikun Zhang; Xinyu Hao; Mingliang Hou; Renqiang Luo; Ziqi Xu

cs.CVMar 30, 2026

Prototype-Enhanced Multi-View Learning for Thyroid Nodule Ultrasound Classification

Authors: Yangmei Chen, Zhongyuan Zhang, Xikun Zhang, Xinyu Hao, Mingliang Hou, Renqiang Luo, Ziqi Xu

Organizations: Guangdong Institute of Smart Education, Jinan University, Guangzhou 510632, China

Abstract

Thyroid nodule classification using ultrasound imaging is essential for early diagnosis and clinical decision-making; however, despite promising performance on in-distribution data, existing deep learning methods often exhibit limited robustness and generalisation when deployed across different ultrasound devices or clinical environments. This limitation is mainly attributed to the pronounced heterogeneity of thyroid ultrasound images, which can lead models to capture spurious correlations rather than reliable diagnostic cues. To address this challenge, we propose PEMV-thyroid, a Prototype-Enhanced Multi-View learning framework that accounts for data heterogeneity by learning complementary representations from multiple feature perspectives and refining decision boundaries through a prototype-based correction mechanism with mixed prototype information. By integrating multi-view representations with prototype-level guidance, the proposed approach enables more stable representation learning under heterogeneous imaging conditions. Extensive experiments on multiple thyroid ultrasound datasets demonstrate that PEMV-thyroid consistently outperforms state-of-the-art methods, particularly in cross-device and cross-domain evaluation scenarios, leading to improved diagnostic accuracy and generalisation performance in real-world clinical settings. The source code is available at https://github.com/chenyangmeii/Prototype-Enhanced-Multi-View-Learning.

Figures & tables

Panel of Figure 1 — Fig. 1: Examples illustrating pronounced intra-class heterogeneity in thyroid ultrasound images, where nodules of the same pathological type exhibit diverse visual manifestations across multiple lesion attributes.

Panel of Figure 2 — Fig. 2: Overview of the proposed PEMV-thyroid framework for thyroid ultrasound classification. The MVFE module extracts multi-view mediator representations, while the PBC module refines these representations using class-conditional prototypes to mitigate spurious correlations under heterogeneous imaging conditions.

Method	ACC(%)	P(%)	R(%)	F1(%)
ResNet $18$	$79.67_{\pm 1.96}$	$80.88_{\pm 5.00}$	$62.29_{\pm 4.22}$	$70.17_{\pm 2.76}$
Fishr	$79.74_{\pm 3.02}$	$77.61_{\pm 5.87}$	$67.88_{\pm 9.78}$	$71.71_{\pm 5.45}$
MixupNet	$78.11_{\pm 2.86}$	$77.31_{\pm 3.42}$	$60.76_{\pm 6.41}$	$67.94_{\pm 5.16}$
MixStyleNet	$78.96_{\pm 1.54}$	$76.12_{\pm 2.13}$	$66.02_{\pm 4.21}$	$70.63_{\pm 2.71}$
PEMV-thyroid	$\textbf{82.08}_{\pm 1.14}$	$\textbf{79.95}_{\pm 1.11}$	$\textbf{71.27}_{\pm 3.23}$	$\textbf{75.32}_{\pm 2.04}$

TABLE I: Comparison of different methods on the TN

3

K dataset. All results are reported in percentage (%), and the best performance is highlighted in bold.

Method	ACC(%)	P(%)	R(%)	F1(%)
ResNet18	$85.68_{\pm 0.66}$	$88.78_{\pm 1.42}$	$92.09_{\pm 1.37}$	$90.39_{\pm 0.39}$
Fishr	$85.82_{\pm 0.46}$	$88.28_{\pm 1.04}$	$92.97_{\pm 1.01}$	$90.55_{\pm 2.06}$
MixupNet	$85.66_{\pm 0.60}$	$88.75_{\pm 0.48}$	$92.07_{\pm 1.52}$	$90.37_{\pm 0.50}$
MixStyleNet	$84.68_{\pm 0.79}$	$87.52_{\pm 1.19}$	$92.23_{\pm 1.41}$	$89.80_{\pm 0.52}$
PEMV-thyroid	$\textbf{86.50}_{\pm 0.55}$	$\textbf{88.88}_{\pm 0.87}$	$\textbf{93.21}_{\pm 0.97}$	$\textbf{90.99}_{\pm 0.36}$

TABLE II: Comparison of different methods on the TN

5000

dataset. All results are reported in percentage (%), and the best performance is highlighted in bold.

Fig. 3: Sensitivity analysis of the number of expert networks ( num_att ) in the MVFE module on the TN $3$ K dataset, evaluated using ACC, P, R, and F $1$ (%).

TN $3$ K
Method	ACC(%)	P(%)	R(%)	F1(%)
AB $1$	$79.67_{\pm 1.96}$	$\textbf{80.88}_{\pm 5.00}$	$62.29_{\pm 4.22}$	$70.17_{\pm 2.76}$
AB $2$	$79.87_{\pm 1.81}$	$79.06_{\pm 4.75}$	$65.51_{\pm 4.43}$	$71.42_{\pm 2.36}$
AB $3$	$78.73_{\pm 1.57}$	$73.80_{\pm 4.41}$	$70.25_{\pm 5.44}$	$71.69_{\pm 2.09}$
AB $4$	$78.73_{\pm 1.44}$	$78.85_{\pm 5.54}$	$62.37_{\pm 8.88}$	$68.99_{\pm 3.71}$
AB $5$	$\textbf{82.08}_{\pm 1.14}$	$79.95_{\pm 1.11}$	$\textbf{71.27}_{\pm 3.23}$	$\textbf{75.32}_{\pm 2.04}$

TABLE III: Ablation results on the TN

3

K and TN

5000

datasets (mean

\pm

std, %). AB

1

: ResNet

18

(ERM baseline); AB

2

: AB1+MVFE; AB

3

: AB

2

+PBC; AB

4

: AB

3

+IP; AB

5

: full model. The best and second-best results are highlighted in bold and underlined, respectively.

Explore similar work

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
CMCNet: Aligning Ultrasound Image Embeddings with Textual TI-RADS Representations for Fine-Grained Thyroid Classification
Aug 14, 2026Bingxin Yu, Xueli Wang, Jerry Zhou +6Medical Image Classification Medical Image Representation Learning
China-Japan Union Hospital of Jilin University, Changchun, 130012, P.R. China. · College of Computer Science and Technology, Jilin University, Changchun, 130012, P.R. China. · Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University, Changchun, 130012, P.R. China. +3
Thyroid Nodule Segmentation in Ultrasound Images with Conformal Uncertainty Estimation
Oct 9, 2026Francesco Teodori, Pierpaolo Brutti, Umberto Ferraro PetrilloUltrasound Image Segmentation Medical Image Segmentation
Department of Statistical Sciences, Sapienza University of Rome, Rome, Italy.

Oct 3, 2026J/K move · Enter open · S save

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

Tasnim Jahan, Md Easin Arafat, Swakkhar Shatabda

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

Ultrasound Image Segmentation Medical Image Segmentation

Open paper AskPDF arXiv

Accurate thyroid ultrasound segmentation is often challenged by low contrast, speckle noise, and unclear boundaries. Although recent methods have improved segmentation accuracy, many rely on resource-intensive architectures or lack explicit integration of multiscale features with global biomedical visual guidance. In this paper, we introduce ThyCLIPNet, a lightweight semantic-guided hybrid encoder-decoder framework that integrates BiomedCLIP-derived biomedical semantic guidance into a lightweight multi-scale CNN segmentation pipeline. The encoder integrates MobileNetV2 with efficient channel attention, while atrous spatial pyramid pooling and a custom convolutional block attention module enrich bottleneck features. The decoder combines hierarchical skip connections and lightweight attention refinement with a BiomedCLIP-guided gated fusion pathway that projects vision-only global biomedical embeddings into decoder feature space and selectively integrates them through semantic-local fusion and spatial gating. To the best of our knowledge, ThyCLIPNet is among the first lightweight thyroid ultrasound segmentation frameworks to use BiomedCLIP's vision encoder alone for image-only global semantic guidance without text prompting. Experiments on TG3K, TN3K, DDTI, and PKTN achieve dice similarity coefficients of 96.22%, 87.58%, 84.73%, and 80.70%; intersection over union scores of 92.72%, 77.91%, 73.51%, and 67.64%; and 95th-percentile hausdorff distances of 3.75, 16.38, 18.23, and 10.86, respectively. ThyCLIPNet uses 8.55M parameters and 22.99G FLOPs. Overall, the results support integrating global biomedical semantic guidance with lightweight multi-scale CNN representations for robust and computationally efficient thyroid ultrasound segmentation. Source code: https://github.com/Tasnim-Jahan/ThyCLIPNet. [Abstract shortened for arXiv. See PDF for full abstract.]

Abstract

Explore similar work

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

CMCNet: Aligning Ultrasound Image Embeddings with Textual TI-RADS Representations for Fine-Grained Thyroid Classification

Thyroid Nodule Segmentation in Ultrasound Images with Conformal Uncertainty Estimation