Title: Deep learning-based malignancy prediction in thyroid nodules

Authors: L. Mohana Sundari; M.S. Maharajan; T. Senthil Kumar; Leo John Baptist Andrews

Addresses: Department of Electronics and Communication Engineering, Saveetha Engineering College (Autonomous), Chennai, India ' Department of Computer Science and Engineering, Sri Sai Ram Institute of Technology, Chennai, India ' Biomedical Engineering, GRT Institute of Engineering and Technology, Tiruttani, India ' Department of Information Technology, Faculty of Engineering and Technology, Botho University, Botswana

Abstract: Although the vast majority of thyroid nodules are non-cancerous, determining whether or not a nodule is cancerous may be a difficult and time-consuming process that often involves unnecessary surgical events. In addition, we discussed the process of developing a model that might anticipate the presence of cancer in thyroid nodules by including a number of the core demographic and ultrasound parameters. A combined sensitivity and specificity score was used to assess the diagnostic performance, and their accuracy was compared to that of radiologists. The comparison between model prediction and expert evaluation reveals the benefit of our approach over human judgement in predicting thyroid nodule malignancy. The results of the experiments show that the suggested algorithm performs better. Nodules of TI-RADS category 4 were used. The area under the receiver operating characteristic curve in the validation dataset was 0.92 (with accuracy of 0.70, sensitivity of 0.81 and specificity of 0.58).

Keywords: thyroid; AI; deep learning; medical imaging; deep learning; CNNs; preoperative diagnosis.

DOI: 10.1504/IJMEI.2025.145041

International Journal of Medical Engineering and Informatics, 2025 Vol.17 No.2, pp.103 - 115

Received: 01 Jun 2022
Accepted: 22 Jul 2022
Published online: 18 Mar 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article