A Computer-Aided Diagnosis System Using Artificial Intelligence for the Diagnosis and Characterization of Thyroid Nodules on Ultrasound: Initial Clinical Assessment

Abstract

The sensitivity of the CAD system using AI for malignant thyroid nodules was as good as that of the experienced radiologist, while specificity and accuracy were lower than those of the experienced radiologist. The CAD system showed an acceptable agreement with the experienced radiologist for characterization of thyroid nodules.

Computer‐aided system for diagnosing thyroid nodules on ultrasound: A comparison with radiologist‐based clinical assessments

“…The AUC of the CAD system was 0.73. Our results are partially comparable to those from a recent study that indicated a pooled sensitivity of 0.91 for the ultrasound reporting system in the differential diagnosis of thyroid nodules . However, our results showed a relatively low specificity.…”

“…The TIRADS category 2 consists of benign lesions (including simple cysts, spongiform nodules, isolated macrocalcifications, and typical subacute thyroiditis). According to the newly published TI-RADS patterns proposed by the ACR, 13 points are given for all the ultrasound features of a nodule, and more suspicious features are awarded additional points. The point total determines the nodule's ACR TIRADS level, which ranges from TR1 (benign) to TR5 (high suspicion of malignancy).…”

“…Computer‐aided diagnosis (CAD) was recently developed to diagnosis thyroid nodules according to sonographic features. Previous study results showed that the CAD system had an acceptable diagnostic accuracy for thyroid nodules . We compared the diagnostic performance of the thyroid ultrasound CAD system using artificial intelligence with that of an experienced radiologist using the K‐TIRADS pattern, the ACR TIRADS pattern, and the ATA guidelines.…”

Deep convolutional neural network for the diagnosis of thyroid nodules on ultrasound

“…However, when the diagnostic performances of quantitative analysis and visual assessment by radiologists were compared in predicting thyroid malignancy, quantitative analysis showed relatively inferior diagnostic performance compared to traditional grayscale US . A number of machine learning algorithms using US images have been investigated for the differential diagnosis of thyroid nodules and have shown conflicting results when diagnostic performances were compared between these algorithms and visual assessment by radiologists . The convolutional neural network (CNN) is a typical type of deep learning technique with fully trainable models and is accepted as a state‐of‐the‐art method .…”

“…13,14 A number of machine learning algorithms using US images have been investigated for the differential diagnosis of thyroid nodules and have shown conflicting results when diagnostic performances were compared between these algorithms and visual assessment by radiologists. [15][16][17][18][19][20] The convolutional neural network (CNN) is a typical type of deep learning technique with fully trainable models and is accepted as a state-of-the-art method. 21,22 Ma et al 23 first attempted a CNN-based method to classify thyroid nodules and the method was shown to be eligible for thyroid nodule diagnosis, but their study did not enroll patients in a clinical setting and did not compare diagnostic performances between CNN and radiologists.…”