Purpose: Thyroid nodules are one of the manifestations of thyroid lesions, and their rapid and accurate diagnosis is crucial to help develop an appropriate treatment for patients. The lack of representative features between benign (especially TI-RADS 3) and malignant nodules limits the accuracy of diagnosis and leads to inconsistent interpretation, overdiagnosis and unnecessary biopsies. Thus, deep learning methods are needed to improve accuracy of diagnosis and specificity of biopsy recommendations.
Methods: In this paper, we propose a thyroid nodule classification model based on ViT with contrast learning called TC-ViT for TI-RADS 3 and malignant nodules. The model inputs the original thyroid and nodule image pairs, and uses contrast learning to promote the representation learning of robust features. The contrastive learning can minimize the representation distance between nodules of the same category, strengthen the representation consistency between global and local nodule features, and finally achieve accurate diagnosis of TI-RADS 3 or malignant nodules. ViT can well explore the global features of thyroid nodules. And nodule images are used as ROI to enhanced the local features of the ViT.
Results: We collect 689 ultrasound images of thyroid nodules to train and test our classification network. The test results achieve an accuracy of 85.5%. The evaluation metrics demonstrate the superior classification performance of the network to other classical deep learning-based networks.
Conclusion: The proposed contrastive learning-based classification ViT can accurately realize automatic classification for TI-RADS 3 and malignant nodules on ultrasound images. It can also be used as an key step in computeraided diagnosis for comprehensive analysis and accurate diagnosis. It shows superior performance to other classical methods and demonstrates the potential for accurate classification of nodules in clinical applications.
Funding Support, Disclosures, and Conflict of Interest: This work is supported by Changzhou Sci&Tech Program (No. CJ20210128 and CJ20200099), General Program of Jiangsu Provincial Health Commission (No. M2020006), Changzhou Key Laboratory of Medical Physics (No. CM20193005).
Ultrasonics, Classifier Design, Diagnostic Radiology
Not Applicable / None Entered.