A Clinical Decision Support System Using Ultrasound Textures and Radiologic Features to Distinguish Metastasis From Tumor‐Free Cervical Lymph Nodes in Patients With Papillary Thyroid Carcinoma. (30th March 2018)

• Record Type: Journal Article
• Authors: Abbasian Ardakani, Ali; Reiazi, Reza; Mohammadi, Afshin
• Is Part Of: Journal of ultrasound in medicine. Volume 37:Number 11(2018)
• Page Start: 2527
• Page End: 2535
• Publication Date: 2018-03-30
• Subjects: computer‐assisted -- diagnosis -- head and neck -- informatics/image processing -- lymph nodes -- pattern recognition -- thyroid carcinoma -- thyroid/parathyroid -- ultrasound; Ultrasonics in medicine -- Periodicals; Ultrasonics; Ultrasonography; Ultrasonics in medicine; Electronic journals; Periodicals; Periodicals; 616.07543
• DOI: 10.1002/jum.14610
• Languages: English
• ISSNs: 0278-4297

Abstract

Abstract : Objectives: This study investigated the potential of a clinical decision support approach for the classification of metastatic and tumor‐free cervical lymph nodes (LNs) in papillary thyroid carcinoma on the basis of radiologic and textural analysis through ultrasound (US) imaging. Methods: In this research, 170 metastatic and 170 tumor‐free LNs were examined by the proposed clinical decision support method. To discover the difference between the groups, US imaging was used for the extraction of radiologic and textural features. The radiologic features in the B‐mode scans included the echogenicity, margin, shape, and presence of microcalcification. To extract the textural features, a wavelet transform was applied. A support vector machine classifier was used to classify the LNs. Results: In the training set data, a combination of radiologic and textural features represented the best performance with sensitivity, specificity, accuracy, and area under the curve (AUC) values of 97.14%, 98.57%, 97.86%, and 0.994, respectively, whereas the classification based on radiologic and textural features alone yielded lower performance, with AUCs of 0.964 and 0.922. On testing the data set, the proposed model could classify the tumor‐free and metastatic LNs with an AUC of 0.952, which corresponded to sensitivity, specificity, and accuracy of 93.33%, 96.66%, and 95.00%. Conclusions: The clinical decision support method based on textural and radiologic features has the potential to characterize LNs via 2‐dimensional US. Therefore, it can be used as a supplementary technique in daily clinical practice to improve radiologists' understanding of conventional US imaging for characterizing LNs.