The Diagnostic Value of Artificial Intelligence Ultrasound S-Detect Technology for Thyroid Nodules

This study aimed to evaluate the consistency of ultrasound TI-RADS classification used by sonographers with different ultrasound diagnosis experience in the diagnosis of thyroid nodules and the diagnostic value of using artificial intelligence ultrasound S-Detect technology in the differentiation of benign and malignant thyroid lesions. 100 patients who underwent ultrasound examination of thyroid masses in our hospital from June 2019 to June 2021 and were further punctured or operated on were included in the study. Pathological results were used as the gold standard to evaluate ultrasound S-Detect technology and the value of TI-RADS classification and the combined application of the two in diagnosing benign and malignant thyroid TI-RADS 4 types of nodules, and the consistency of judgments of doctors of different ages is assessed by a Kappa value. There were 128 nodules in 100 patients, 51 benign nodules, and 77 malignant nodules. For senior physicians, the sensitivity of diagnosis using TI-RADS classification combined with ultrasound S-Detect technology is 93.5%, specificity is 94.1%, and accuracy is 93.8%; for middle-aged physicians using TI-RADS classification combined with ultrasound S-Detect technology for diagnosis, the sensitivity is 89.6%, specificity is 92.2%, and accuracy is 90.6%; for junior doctors, the sensitivity of diagnosis using TI-RADS classification combined with ultrasound S-Detect technology is 83.1%, specificity is 88.2%, and accuracy is 85.1%. Regardless of seniority, the combined application of artificial intelligence ultrasound S-Detect technology and TI-RADS classification can improve the diagnostic ability of sonographers for thyroid nodules and at the same time improve the consistency of judgment among physicians, and this is especially important for radiologists.


Introduction
Athyroid nodule is one of the most common nodular lesions in adults, and its incidence is getting higher and higher. Te prevalence rate of Chinese residents is as high as 18.6%, and it is more common in women, and most of them are benign nodules, only 7% of thyroid nodules tend to be malignant, but 5% of patients will change from benign to malignant without timely diagnosis and treatment [1]. Te incidence of thyroid cancer increases with age, and it is the malignant tumor with the largest number of new patients among all cancer types in recent years [2]. According to statistics from the National Cancer Research Center of the United States, there were about 64,300 new cases of thyroid cancer worldwide in 2016, and about 1,980 people died of thyroid cancer [3]. Timely and accurate detection of thyroid nodules and identifcation of benign and malignant thyroid nodules are of great signifcance to improve clinical treatment efects and prognosis of patients [4].
Fine-needle aspiration biopsy (FNAB) or surgical biopsy is the gold standard for diagnosing thyroid cancer, while for incidentally discovered thyroid nodules, ultrasound is the most commonly used noninvasive method to diferentiate benign and malignant thyroid nodules [5]. Ultrasound is a real-time imaging method, which has the advantages of no radiation and no damage, high accuracy, simple and fast operation, and low cost, but it is highly dependent on the operator and has a low repetition rate of diagnostic results [6]. With the development and application of artifcial intelligence (AI) technology in the medical feld, the AI ultrasound intelligent auxiliary diagnosis system based on static ultrasound images has emerged, which can greatly reduce the labor of medical workers while ensuring accuracy. However, the collection of static ultrasound images is still afected by multiple factors, while dynamic AI can perform real-time synchronous dynamic analysis of nodules from multiple levels and angles, determine the nature of nodules, and further improve diagnostic efciency. Tere is no domestic report on the dynamic AI ultrasound intelligently assisted real-time diagnosis system [7][8][9][10][11]. Te application of computer-aided diagnosis (CAD) technology "S-Detect TM " can realize the qualitative and quantitative automatic analysis of ultrasound images and obtain objective, repeatable, and more accurate diagnostic results. In the latest generation of ultrasonic diagnostic equipment, S-Detect TM technology adopts the method of deep learning, which can improve the accuracy of diagnosis. At present, only a few scholars have reported the preliminary application of S-Detect TM technology in the diferential diagnosis of thyroid masses [9][10][11][12][13][14]. Te purpose of this study was to investigate the consistency of ultrasound TI-RADS classifcation in the diagnosis of thyroid nodules by sonographers with diferent ultrasound diagnostic experiences, to explore the diagnostic value of using artifcial intelligence ultrasound S-Detect technology in the diferentiation of benign and malignant thyroid lesions, as well as the clinical value of the combined diagnosis of S-Detect technology and routine thyroid ultrasonography, so as to provide sonographers with a more objective and accurate assessment tests for thyroid tumors.

Instruments and Methods.
We used SamsungRS80A (South Korea, Seoul, Samsung Madison Co., Ltd., L3-12A linear array high-frequency probe, frequency 5-13 MHz, equipped with S-Detect intelligent detection system) highend color Doppler ultrasound diagnostic apparatus to scan the thyroid, to understand the overall situation of the thyroid, longitudinal and transverse scanning of nodules (select 2-3 nodules with pathological results and the most suspicious of malignancy) and then analyzed the color Doppler signal of the nodule, longitudinal section and transverse section of the nodule were saved. We entered the S-Detect mode, started the S-Detect automatic analysis program to automatically draw the region of interest (ROI) on the largest longitudinal section and the largest transverse section of the thyroid nodule, and output the S-Detect diagnostic report, and the ultrasound images and evaluation report records automatically analyzed by S-Detect were stored on the hard disk for later data sorting and data analysis.

Tyroid Nodule Ultrasound TI-RADS Classifcation
Evaluation. All cases were routinely examined by two senior doctors with more than 10 years of experience in thyroid ultrasound diagnosis. Tyroid ultrasound TI-RADS classifcation assessment was performed by 6 sonographers with diferent experience in low, medium, and high grades and divided into three groups (there are 2 people in the low-level group, divided into low-level A and low-level B, physicians with 2 years of experience in thyroid disease ultrasound diagnosis, 2 people in the middle-level group, divided into middle-level A and middle-level senior B, physicians with 5 years of experience in thyroid disease ultrasound diagnosis, and there are 2 seniors in the senior group, divided into senior A and senior B, physicians with 10 years of experience in thyroid disease ultrasound diagnosis). According to the 2017 version of the ultrasound TI-RADS classifcation criteria [13], a double-blind method was used to evaluate the TI-RADS classifcation of thyroid nodules. Te diagnostic cutof between TI-RADS4 and TI-RADS5 was used as the diagnostic cut-of point for benign and malignant tumors, and TI-RADS4 was assessed as a possible benign tumor and TI-RADS5 was assessed as a possible malignant tumor. All sonographers were unaware of the clinical information of the cases, the number of benign and malignant cases, and the fnal diagnosis of the cases.

Evaluation of Artifcial Intelligence Ultrasound S-Detect Technology.
We activate the S-Detect automatic analysis program, make precise identifcation of lesions based on various characteristics of lesions, and use the ultrasonic TI-RADS classifcation report to issue recommended diagnostic results for thyroid nodules, which are automatically generated S-Detect diagnostic reports, judge the lesion as "probably benign" or "probably malignant" and prompt the result of "binary classifcation," and fnally keep a record of images and evaluation results. If the prompting results of the two sections are diferent, it is regarded as "probably malignant." Te enumeration data were expressed as a percentage, and the chi-square test of paired data was used to compare the between-group comparison between the ultrasonic features of thyroid tumors judged by artifcial intelligence ultrasound S-Detect technology and the results judged by sonographers. When P < 0.05, the diference was statistically signifcant. Consistency assessment: 0.6-0.8 is fair, 0.8-0.9 is good, and 0.9-1.0 is excellent.

Diagnostic Results of Senior Physicians and Diferent
Diagnostic Methods. For senior physicians, 63 benign nodules and 65 malignant nodules were diagnosed by TI-RADS classifcation, 57 benign nodules and 71 malignant nodules were diagnosed by ultrasound S-Detect technology, and when the two were used together, 55 benign nodules and 73 malignant nodules were diagnosed, and the kappa value was between 0.6 and 1; the sensitivity, specifcity, and accuracy of TI-RADS for diagnosing thyroid nodules were 83.1%, 88.2%, and 85.1%, respectively, and AUC � 0.796, and the diagnostic sensitivity of ultrasonic S-Detect technology was 89.6%, specifcity was 88.2%, accuracy was 89.1%, and AUC � 0.869. When the two were combined, diagnostic sensitivity was 93.5%, specifcity was 94.1%, accuracy was 93.8%, and AUC � 0.923 (Table 1, Figure 1).  Figure 2).  Figure 3).

Discussion
Te combined application of medical image data and AI based on deep learning has revolutionized the expression of medical images. Many centers have carried out static AI diagnosis of thyroid nodules in clinical practice. A large number of research data at home and abroad show that AI is efective in judging benign and malignant thyroid nodules. It has high diagnostic value, is convenient and quick, and improves clinical examination and diagnostic efciency [8][9][10][11][12][13]. On this basis, dynamic AI uses an ultra-large-scale convolutional neural network and deep learning technology to realize real-time localization, the real-time outline of thyroid nodules, and real-time auxiliary diagnosis of benign and malignant nodules during the inspection process, making ultrasonography more efcient and precise.   S-Detect TM technology is an emerging computer-aided diagnostic method, based on Korea and Russ TI-RADS (Tyroid Image Reporting and Data System) classifcation and ATA (American Tyroid Association) guideline classifcation, and using the deep learning model CNN (convolutional neural network), it automatically detects and analyzes the internal structure, echo level, boundary, direction, shape, and other information of thyroid tumors to realize the diagnosis of benign and malignant lesions [15,16]. Te application of S-Detect TM can not only improve diagnostic accuracy and repeatability but also greatly reduce the work pressure of ultrasound doctors, which has important reference value for beginners and clinicians.
Among the 128 nodules in the 100 patients in this study, 51 were benign nodules and 77 were malignant nodules confrmed by pathology. For senior physicians, the sensitivity, specifcity, and accuracy of TI-RADS classifcation for diagnosing thyroid nodules were 89.6%, 89.6%, and 89.1%, respectively (kappa value 0.701); for middle-aged physicians, the sensitivity, specifcity, and accuracy of TI-RADS classifcation for diagnosing thyroid nodules were 80.5%, 78.4%, and 80.4%, respectively (kappa value 0.604); for junior physicians, the sensitivity of TI-RADS classifcation for thyroid nodules was 72.7%, specifcity was 76.5%, and accuracy was 74.2% (kappa value 0.440). With the increase in the physician seniority level, their diagnostic ability for thyroid nodules was signifcantly better, and they showed better evaluation consistency, confrming the infuence of physician experience on the diagnosis of thyroid nodules.
Artifcial intelligence ultrasound S-Detect technology is the frst commercial image analysis program based on deep learning algorithms newly developed by Samsung Medical in recent years. It is installed on the high-end ultrasound diagnostic equipment of Samsung Madison in South Korea. Te clinical application is mostly concentrated in the diagnosis of breast tumors [16][17][18][19], while the application in the diagnosis of thyroid nodules is less. In this study, the application of ultrasound S-Detect technology improved the specifcity, sensitivity, and accuracy of thyroid nodule diagnosis among sonographers of diferent seniority, especially for middle and low-level doctors, and when ultrasound S-Detect technology is combined with the TI-RADS classifcation and the diagnostic method, it can signifcantly improve the diagnostic ability of middle and low-level  Computational Intelligence and Neuroscience 5 physicians for thyroid nodules. Tis is especially important for junior physicians, because their diagnostic sensitivity for benign and malignant thyroid nodules using TI-RADS classifcation is only 72.7%, and the specifcity (76.5%) and accuracy (74.2%) also need to be improved urgently. When the ultrasound S-Detect technique was used in combination with the TI-RADS classifcation, the diagnostic sensitivity of junior physicians for benign and malignant thyroid nodules increased to 83.1%, which was the same as that of senior physicians using the TI-RADS classifcation (88.2% specifcity and 85.1% accuracy), at the same time, due to the application of artifcial intelligence ultrasound S-Detect technology, the judgment of physicians tends to be correct, resulting in a signifcant improvement in the consistency of assessment among junior physicians, from general diagnostic consistency (kappa � 0.44) to stronger diagnostic consistency (kappa � 0.696), and this improvement in diagnostic consistency also occurred among middle-aged and senior physicians.
In conclusion, the results of this study show that regardless of their seniority, the combined application of artifcial intelligence ultrasound S-Detect technology and TI-RADS classifcation can improve the diagnostic ability and diagnostic consistency of sonographers for thyroid nodules. It is especially obvious for junior doctors, which helps improve the diagnostic confdence of junior sonographers and avoid unnecessary needle biopsy or surgery, which is worthy of clinical promotion. However, this study also has certain limitations; for example, it is only a single-center retrospective study, and the number of patients is small.

Data Availability
Te data used to support the fndings of this study are included within the article.

Conflicts of Interest
Te authors declare that they have no conficts of interest.