A Multi-Modal, Multi-Task Thyroid Ultrasound Imaging Dataset for Computer-Aided Thyroid Nodule Diagnosis
Zelan Li, Jianning Chi, Geng Lin, Huixuan Wu, Jiahui Chen, Ying Huang · 2025
The incidence of thyroid cancer has significantly increased over the past decade, and accurately identifying malignant thyroid nodules is crucial for early diagnosis and treatment. With the rapid development of deep learning algorithms, the diagnosis of benign and malignant thyroid nodules, as well as the detection and segmentation of lesions, has become more efficient. However, data-driven medical image deep learning algorithms still face challenges such as annotation difficulties, limited datasets, and the use of single-modal data. To address these issues, this paper presents a multi-modal, multi-task computer-aided diagnosis dataset named SHUS-CEUS specifically designed for thyroid disease research. The dataset integrates ultrasound sequence images with corresponding contrast-enhanced ultrasound videos, combining multi-modal information. It also includes benign/malignant classification labels and nodule segmentation annotations made by experienced clinicians, supporting both nodule classification and segmentation tasks. We conducted studies on thyroid nodule segmentation tasks, benign/malignant diagnostic tasks, and pre-training transfer learning based on this dataset. Extensive experimental results further demonstrate the contribution of the proposed multi-modal dataset in developing AI-based transfer learning, diagnostic, and segmentation models for thyroid nodules.