Journal of Information Technology Management

Journal of Information Technology Management

A Two-Stage Progressive Transfer-Learning Framework for Binary Thyroid Nodule Classification in Ultrasound Imaging

Document Type : Research Paper

Authors
1 Department of Computer Science, College of Computer Science & Information Technology, University of Kerbala, Karbala, Iraq.
2 College of Pharmacy, University of Kerbala, Karbala, Iraq.
3 College of Business Informatics, University of Information Technology and Communications (UoITC), Baghdad, Iraq.
10.22059/jitm.2026.109979
Abstract
Thyroid nodules are highly common, mostly benign, and evaluated using ultrasound as the standard of care; however, their interpretation is often operator-dependent. This paper proposes an experimental approach based on deep transfer learning for binary thyroid nodule classification using a subset of 3,115 resized grayscale ultrasound images from Kaggle. We propose a two-stage progressive fine-tuning protocol with intermediate output suppression to prevent catastrophic forgetting, along with a preprocessing pipeline that combines bilateral filtering, contrast-limited adaptive histogram equalization (CLAHE), and single-image unsharp masking. Here, we perform a systematic architecture search, comparing six different architectures: our custom CNN trained from scratch; EfficientNetB0, DenseNet121, and ResNet152 (all with ImageNet-pretrained weights); and a CNN-handcrafted-feature-fusion model that is fine-tuned over all the compared backbone architectures and jointly evaluated against average classifier accuracies per species at 6-, 12-, and 20-fold imbalanced ratios. This yields a best-performing architecture (an ImageNet-pretrained ResNet50 with two-stage fine-tuning), achieving 77.61% test accuracy, with test AUC evaluated on a withheld test set of 469 images. Class 0 precision, recall, and F1-scores were 0.83, 0.80, and 0.81, respectively, compared with 0.70, 0.74, and 0.72, respectively, for class 1. Gradient-weighted class activation mapping (Grad-CAM) confirmed the qualitative clinical plausibility of the model, with network attention directed at hyperechoic nodule margins rather than background artifacts. We further contextualize these findings in relation to the wider literature on thyroid ultrasound image classification using CNNs and discuss how our single-backbone scheme relates to ensemble approaches reported in concurrent literature. We also discuss lessons regarding generalization, gradient shock, and the extent to which preprocessing and feature engineering are necessary for small medical imaging datasets.
Keywords

Buda, M., Wildman-Tobriner, B., Hoang, J. K., Thayer, D., Tessler, F. N., Middleton, W. D., & Mazurowski, M. A. (2019). Management of thyroid nodules seen on US images: Deep learning may match performance of radiologists. Radiology, 292(3), 695–701.
Chambara, N., & Ying, M. (2019). The diagnostic efficiency of ultrasound computer-aided diagnosis in differentiating thyroid nodules: A systematic review and narrative synthesis. Cancers, 11(11), 1759.
Dar, S. U. H., Ozbey, M., Catli, A. B., & Cukur, T. (2020). A transfer-learning approach for accelerated MRI using deep neural networks. Magnetic Resonance in Medicine, 84(2), 663–685.
Durante, C., Grani, G., Lamartina, L., Filetti, S., Mandel, S. J., & Cooper, D. S. (2018). The diagnosis and management of thyroid nodules: A review. JAMA, 319(9), 914–924.
Floridi, C., Cellina, M., Buccimazza, G., et al. (2019). Ultrasound imaging classifications of thyroid nodules for malignancy risk stratification and clinical management: State of the art. Gland Surgery, 8(Suppl. 3), S233.
Gupta, P., Malhotra, P., Narwariya, J., Vig, L., & Shroff, G. (2020). Transfer learning for clinical time series analysis using deep neural networks. Journal of Healthcare Informatics Research, 4(2), 112–137.
He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 770–778).
Hosny, A., Parmar, C., Quackenbush, J., Schwartz, L. H., & Aerts, H. (2018). Artificial intelligence in radiology. Nature Reviews Cancer, 18(8), 500–510.
Huang, G., Liu, Z., van der Maaten, L., & Weinberger, K. Q. (2017). Densely connected convolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 4700–4708).
Jia, D., Wei, D., Socher, R., Li, L.-J., Kai, L., & Li, F.-F. (2009). ImageNet: A large-scale hierarchical image database. In Proceedings of the 2009 IEEE Conference on Computer Vision and Pattern Recognition (pp. 248–255).
Lee, J. H., Kim, Y. G., Ahn, Y., et al. (2023). Investigation of optimal convolutional neural network conditions for thyroid ultrasound image analysis. Scientific Reports, 13, 1360.
Lim, H., Devesa, S. S., Sosa, J. A., Check, D., & Kitahara, C. M. (2017). Trends in thyroid cancer incidence and mortality in the United States, 1974–2013. JAMA, 317(13), 1338–1348.
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017). Grad-CAM: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE International Conference on Computer Vision (ICCV) (pp. 618–626).
Shao, J., Zheng, J., & Zhang, B. (2020). Deep convolutional neural networks for thyroid tumor grading using ultrasound B-mode images. The Journal of the Acoustical Society of America, 148(3), 1529.
Tan, M., & Le, Q. V. (2019). EfficientNet: Rethinking model scaling for convolutional neural networks. In Proceedings of the International Conference on Machine Learning (ICML) (pp. 6105–6114).
Tong, W.-J., Wu, S.-H., Cheng, M.-Q., et al. (2023). Integration of artificial intelligence decision aids to reduce workload and enhance efficiency in thyroid nodule management. JAMA Network Open, 6(5), e2313674.
Wu, M. H., Chen, K. Y., Shih, S. R., et al. (2020). Multi-reader multi-case study for performance evaluation of high-risk thyroid ultrasound with computer-aided detection. Cancers, 12(2), 373.
Xu, W., Jia, X., Mei, Z., et al. (2023). Generalizability and diagnostic performance of AI models for thyroid US. Radiology, 307(5), e221157.
Xu, Y., Xu, M., Geng, Z., Liu, J., & Meng, B. (2025). Thyroid nodule classification in ultrasound imaging using deep transfer learning. BMC Cancer, 25, 544.