Diagnostic value of a dynamic artificial intelligence-based, ultrasound-assisted diagnostic system in differentiating between benign and malignant thyroid nodules with a diameter greater than 2 cm

Introduction

Thyroid cancer is one the most common malignancies of the head and neck (1). In 2022, there were approximately 821,000 new cases of thyroid cancer worldwide, an increase of 235,000 cases compared to 2020; thus, thyroid cancer poses a serious threat to the life and health of patients (1). The overall prognosis of patients with thyroid cancer is relatively good; however, some patients still experience local invasion in the neck or distant metastasis, which severely affects their quality of life and even directly leads to death. Thus, the risk of death from thyroid cancer cannot be ignored. Early and precise diagnosis and treatment can not only significantly improve the therapeutic effect and patient prognosis but can also reduce the social and healthcare medical burden.

Ultrasound is the preferred imaging examination for the diagnosis of thyroid nodules. By assessing the size, shape, boundary, calcification, and blood flow of the nodules, ultrasound can assist in distinguishing between benign and malignant nodules, and predicting the risk of malignancy. It is also safe, efficient, non-invasive, and convenient. However, in the diagnosis of larger thyroid nodules, the characteristics of the nodule become atypical due to the more complex relationship between the internal structure of the nodule and the surrounding tissues. Additionally, when nodules are located in complex anatomical structures, such as near major blood vessels in the neck or behind the sternum, the tissue structures can interfere with the ultrasound examination, increasing the risk of misdiagnosis or missed diagnosis.

Fine needle aspiration cytology (FNAC) is the gold standard for preoperative diagnosis; however, a related study (2) has shown that due to the uneven distribution of malignant tissue in nodules, and the easy occurrence of necrosis and liquefaction due to the large size of the nodules, the false-negative rate of large nodules, especially those with a maximum diameter >4 cm, is high; thus, the diagnostic accuracy of FNAC is limited. Computed tomography (CT), as a supplementary examination method to ultrasound, is of great value in assessing the size and nature of thyroid nodules, their relationship with surrounding important structures, such as the larynx, trachea, esophagus, neck vessels, sternum, and mediastinum, and the presence of lymph node metastasis and the extent of involvement. It has high spatial resolution and density. In clinical practice, thyroid CT examination should be considered for patients with large nodules, locally advanced disease, or extensive lymph node involvement (3). Despite the important role of this examination in diagnosis, it still has limitations such as radiation, iodine contrast agent allergy, and the risk of acute kidney injury caused by contrast agents.

Based on deep learning, the dynamic artificial intelligence (AI)-based, ultrasound-assisted auxiliary diagnostic system offers a novel approach for distinguishing between benign and malignant thyroid nodules. It can automatically extract key features from ultrasound images, enabling the real-time localization, delineation, and diagnosis of nodules from different sections during the examination process, which not only improves the accuracy and efficacy of diagnosis, but also reduces diagnostic differences among doctors, and promotes the homogenization of diagnosis (4). In previous studies (5,6), we found that the diagnostic accuracy of dynamic AI for thyroid nodules can reach up to 89.97%, with a high consistency with postoperative pathology (kappa =0.737, P<0.001), and have high diagnostic value for the diagnosis of benign and malignant thyroid nodules in the patients with Hashimoto’s thyroiditis (HT) and level 4 nodules of the Thyroid Imaging Reporting and Data System (TI-RADS) of the American College of Radiology (ACR). However, in the previous studies (5,6), the number of nodules with a diameter >2 cm was relatively small. In this study, we specifically applied dynamic AI to the diagnosis of thyroid nodules with a diameter >2 cm. By comparing the diagnostic efficacy of preoperative ultrasound, CT, FNAC, and dynamic AI, we aimed to study the diagnostic efficacy and value of dynamic AI in distinguishing between benign and malignant thyroid nodules with a diameter >2 cm, explore its guiding significance for surgical treatment strategies, and further improve the diagnosis and treatment level of thyroid nodules with a diameter >2 cm. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-253/rc).

Methods

Patients

In total, 276 patients with thyroid nodules who underwent surgery at the Department of Thyroid & Hernia Surgery, Senior Department of General Surgery, The First Medical Center of Chinese People’s Liberation Army (PLA) General Hospital, between June 2022 and May 2023 were enrolled in the study. Data, including age, sex, nodule dimensions, histopathology, and findings from dynamic AI ultrasound, CT, conventional ultrasound, and FNAC diagnosis, were retrospectively reviewed.

Patients were included in the study if they met the following inclusion criteria: (I) met the diagnostic criteria for thyroid nodules as defined in the “Guidelines for the Diagnosis and Treatment of Thyroid Nodules and Differentiated Thyroid Cancer” (2nd edition) (7); (II) had nodules with the following surgical indications: (i) thyroid carcinoma or follicular neoplasm as indicated by preoperative FNAC; (ii) benign thyroid nodules with compressive symptoms involving the trachea, esophagus, or recurrent laryngeal nerve; (iii) substernal goiter; (iv) nodules associated with hyperthyroidism refractory to pharmacologic or radioiodine therapy; or (v) benign thyroid nodules showing a significant increase in volume in the short term, for which malignant transformation could not be excluded; (III) had nodule with at least one diameter >2 cm as measured by ultrasound using three diameters on the transverse and longitudinal sections; (IV) had a definitive postoperative histopathological diagnosis; and (V) had complete clinical-pathologic data and ultrasonographic imaging records. Patients were excluded from the study if they met any of the following exclusion criteria: (I) were allergic to the iodinated contrast agent used in enhanced CT; (II) had severe organ dysfunction such as heart, liver, brain, and kidney dysfunction; (III) had other types of malignant tumors; (IV) had a mental illness or were unable to cooperate with the examination; (V) had incomplete clinical pathological or imaging data; (VI) had postoperative pathology that did not yield a clear diagnosis; (VII) had Class I (unsatisfactory or undiagnostic material) or Class III (atypical lesions of undetermined significance) based on the FNAC results; and/or (VIII) did not agree to participate in this study and did not sign the informed consent. The flowchart of subject enrollment is shown in Figure 1.

Figure 1 Study flowchart of patient enrollment. CT, computed tomography; FNAC, fine needle aspiration cytology.

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Chinese PLA General Hospital (No. S2021-082-04) and informed consent was obtained from all patients.

Dynamic AI examination equipment and methods

The Ian Thyroid Solution 100 ultrasound imaging intelligent system (version 1.2.8.0, MedAI Technology Co. Ltd., Wuxi, China) was used for the dynamic AI examination. This system includes a host computer, GE LOGIQ e ultrasound diagnostic device, GE L4-12t-RS linear array transducer, and AI auxiliary display screen. All the patients underwent dynamic AI examination before surgery. For the examination, the patients were placed in a supine position with their neck extended and fully exposed. Linear array transducer parameters (e.g., gain, focus, and depth) were adjusted appropriately for each patient based on their thyroid nodule characteristics. The thyroid gland was scanned three times consecutively in both transverse and longitudinal planes, covering the left and right lobes, systematically from top to bottom, and inside out, for a comprehensive assessment of the nodules.

CT examination equipment and methods

The CT examination was performed using the SOMATOM Force dual-source CT scanner (Siemens, Munich, Germany), and the tube voltage and tube current were set at 120 kV and 116 mA, respectively. The rotation time was set at 1.0 s, the matrix setting was 512×512, the pitch was 0.8, the scan layer thickness was 3.0 mm, and the iteration weight was 3. All the patients underwent enhanced CT examination before surgery. For the CT examination, the patients were placed in a supine position with their neck extended and fully exposed, and each arm placed horizontally along the corresponding side of the body. The scanning range was from the base of the skull to the level of the aortic arch, with a slice thickness of 3.0 mm. Iodinated contrast medium (iodine phenol injection, national drug approval number: H20113430) was injected at a flow rate of 2.5–3 mL/s through the elbow vein, and enhanced scanning was performed after a delay of 30 seconds.

Diagnostic criteria

Dynamic AI diagnosis

Dynamic AI automatically identifies the lesion, uses deep learning technology to automatically extract image features, constructs a convolutional neural network to sample the pixel points on the image, and derives global features characterizing the thyroid nodule region in the ultrasound image. This process constructs a high-dimensional, multi-level feature space. Subsequently, the diagnostic model processes the input nodule image’s features and outputs two probability values. The two values represent the probabilities that the diagnostic model assigns to the nodule being malignant and benign, respectively. When the predicted malignant probability value was greater than or equal to the benign probability value, the model prediction result was malignant, and displayed as a red “M” mark, otherwise it was benign, and displayed as a green “B” mark. The percentage represents the probabilities. Based on the results of three scans, a comprehensive analysis was conducted to obtain the final dynamic AI diagnosis result. This operation and process was supervised and guided by an application engineer from an AI company. All nodule properties were confirmed by postoperative pathological results (Figure 2).

Figure 2 Dynamic AI diagnosis, CT, and postoperative pathology of typical cases. (A) Dynamic AI diagnosis of malignant thyroid nodules; (B) CT diagnosis of malignant thyroid nodules; (C) postoperative pathology of malignant nodules showed papillary thyroid carcinoma (HE staining, 200×). (D) dynamic AI diagnosis of benign thyroid nodules; (E) CT diagnosis of benign thyroid nodules; (F) postoperative pathology of benign nodules suggested thyroid follicular adenoma (HE staining, 40×). AI, artificial intelligence; CT, computed tomography; HE, hematoxylin and eosin.

Routine ultrasound diagnosis

The ultrasound examination and diagnosis were performed by one senior physician specializing in ultrasound, using the ACR’s TI-RADS as the diagnostic criteria (8). Under the TI-RADS, grades 1–3 nodules are classified as benign, and grades 4–5 as malignant. Additionally, a senior physician specializing in ultrasound reviewed the films for verification.

CT diagnosis

The CT diagnosis was performed by one associate senior physician specializing in radiology, using the “Expert Consensus on Thyroid Nodule Imaging Examination Process” as the diagnostic criteria (9). According to these criteria, the nodules were classified as: (I) benign (characterized by clear boundaries, a regular shape, the presence of cystic degeneration, with enhanced scans showing clearer margins and higher enhancement compared to non-enhanced scans); or (II) malignant (characterized by blurred borders, an irregular shape, the presence of the “bite-cake” sign and microcalcifications, with enhanced scans showing more blurred margins compared to non-enhanced scans). Additionally, a senior physician specializing in radiology reviewed the films for verification.

FNAC diagnosis

The Bethesda reporting system (10) was used as the diagnostic standard for thyroid nodule pathology. Under this system, the nodules were classified as follows—Category I: unsatisfactory or nondiagnostic sample; Category II: benign; Category III: atypical lesion of undetermined significance; Category IV: follicular neoplasm or suspected follicular neoplasm; Category V: suspicious for malignancy; Category VI: malignant. Due to the limitations of FNAC in diagnosing follicular tumors, Categories II and IV were classified as benign, while Categories V and VI were classified as malignant for the purposes of this study. Additionally, a senior physician specializing in pathology reviewed the films for verification.

Clinical pathological diagnosis

The “Endocrine and Neuroendocrine Tumor Classification” (5th edition), published by the World Health Organization, was used as the diagnostic standard for clinical pathological diagnosis (11). One senior pathologist specializing in this field from our hospital performed the film review and diagnosis.

For the ultrasound, CT, and FNAC pathological diagnoses, three professional senior physicians in this field conducted a retrospective analysis of the enrolled cases. In cases of disagreement, the final confirmation result was determined by the agreement of two of the three physicians.

Statistical analysis

Using postoperative pathology as the gold standard, we calculated the diagnostic efficacy of dynamic AI, including the sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), missed diagnosis rate, and misdiagnosis rate. Agreement was assessed using the kappa statistic, interpreted as follows—kappa value <0: no significant consistency; kappa value ≤0.40: poor consistency; 0.40< kappa value ≤0.60: moderate consistency; 0.60< kappa value ≤0.80: high consistency; kappa value >0.80: extremely high consistency. The χ² test was used to compare the differences between groups. SPSS 23.0 software (IBM, Chicago, IL, USA) was used for the statistical analysis. A P value <0.05 was considered statistically significant.

Results

Postoperative pathology examination

A total of 276 patients (74 males and 202 females) were included in this study. The age of the patients ranged from 14 to 77 years (average age: 45.36±0.84 years). There were 189 patients under the age of 55, and 87 patients aged 55 and above. In total, there were 297 thyroid nodules with diameters ranging from 2 to 10.7 cm, and an average diameter of 3.37±0.07 cm. Among them, there were 138 malignant nodules and 159 benign nodules. Among the malignant nodules, there were 115 papillary thyroid carcinomas, 22 follicular thyroid carcinomas, and one medullary thyroid carcinoma. Among the benign nodules, there were 34 nodular goiters and 125 follicular adenomas.

Diagnostic efficacy analysis of dynamic AI for thyroid nodules with a diameter >2 cm

Of the 297 nodules analyzed in this study, dynamic AI identified 141 as malignant and 156 as benign. The postoperative pathological results confirmed that there were 119 malignant nodules and 137 benign nodules, demonstrating a high consistency (kappa =0.723, P<0.001). The diagnostic efficacy results for dynamic AI are detailed in Tables 1,2.

Stability analysis of dynamic AI for thyroid nodules with a diameter >2 cm

Accuracy analysis of dynamic AI in different genders

As set out in Table 3, among the 276 patients, there were 74 males and 202 females. For the males, 69 of 82 nodules were consistent with the postoperative pathological results (accuracy: 84.15%); while for the females 187 of 215 nodules were consistent with the postoperative pathological results (accuracy: 86.98%). There was no significant difference between the two gender groups in terms of the diagnostic accuracy of the dynamic AI (χ²=0.400, P=0.527).

Accuracy analysis of dynamic AI in different ages

Among the patients aged <55 years (n=189, 200 nodules), dynamic AI had an accuracy of 86.5% (173 of 200 nodules confirmed by postoperative pathology). Among those aged ≥55 years (n=87, 97 nodules), dynamic AI had an accuracy of 85.57% (83 of 97 nodules confirmed by postoperative pathology). No significant difference in accuracy was observed between the age groups (χ²=0.048, P=0.827).

Accuracy analysis of dynamic AI in different nature of nodules

Dynamic AI accurately identified 119 of 138 malignant nodules, and 137 of 159 benign nodules, with an accuracy of 86.23% and 86.16%, respectively. There was no significant difference in the accuracy of dynamic AI in terms of the nature of the nodules (χ²=0.000, P=0.986).

Accuracy analysis of dynamic AI in HT and non-HT patients

In this study, 49 patients with HT had 50 nodules, of which the diagnoses of 45 were consistent with the postoperative pathology (accuracy: 90%). There were 227 non-HT patients with 247 nodules, of which the diagnoses of 211 were consistent with the postoperative pathology (accuracy: 85.43%). The difference between the two groups was not statistically significant (χ²=0.731, P=0.392).

Comparison of diagnostic efficacy between dynamic AI and ACR TI-RADS

Among the 297 nodules analyzed, dynamic AI identified 141 as malignant and 156 as benign. Postoperative pathology confirmed 119 malignant and 137 benign AI-classified nodules, demonstrating a high consistency (kappa =0.723, P<0.001). In comparison, the preoperative TI-RADS classified 169 nodules as malignant and 128 as benign; postoperative pathology confirmed 121 malignant and 111 benign TI-RADS classified nodules (kappa =0.567, P<0.001). Dynamic AI showed higher consistency with postoperative pathology than the TI-RADS. A comparison revealed significant differences in the specificity (χ²=12.383, P<0.001), accuracy (χ²=6.614, P=0.010), PPV (χ²=7.204, P=0.007), and misdiagnosis rate (χ²=12.383, P<0.001) between the dynamic AI examination and TI-RADS. There were no significant differences in the sensitivity, NPV, and missed diagnosis rate between the dynamic AI examination and TI-RADS (Table 4).

Comparison of diagnostic efficacy between dynamic AI and FNAC

Of the 297 nodules analyzed, 162 underwent both dynamic AI and FNAC before surgery. Dynamic AI classified 123 nodules as malignant and 39 nodules as benign; postoperative pathology confirmed 109 of the malignant and 30 of the benign AI classifications. FNAC diagnosed 107 malignant and 55 benign nodules, postoperative pathology confirmed 106 of the malignant and 43 of the benign classifications. There were significant differences in the specificity (χ²=13.582, P<0.001), PPV (χ²=10.245, P=0.001), and misdiagnosis rate (χ²=13.582, P<0.001) between dynamic AI and FNAC. There were no significant differences in the sensitivity, accuracy, NPV, and missed diagnosis rate between dynamic AI and FNAC (Table 5).

Comparison of diagnostic efficacy between dynamic AI and CT

In total, 217 patients with 238 nodules underwent preoperative CT examination. Dynamic AI identified 117 malignant nodules and 121 benign nodules, of which 98 malignant nodules and 105 benign nodules were consistent with the postoperative pathological results, indicating a high consistency (kappa =0.706, P<0.001). CT identified 121 malignant nodules and 117 benign nodules, of which 76 malignant nodules and 79 benign nodules were consistent with the postoperative pathological results, indicating a high consistency (kappa =0.303, P<0.001). There were significant differences in the sensitivity (χ²=11.745, P=0.001), specificity (χ²=14.236, P<0.001), accuracy (χ²=25.961, P<0.001), PPV (χ²=13.281, P<0.001), NPV (χ²=12.573, P<0.001), missed diagnosis rate (χ²=11.745, P=0.001), and misdiagnosis rate (χ²=14.236, P<0.001) between dynamic AI and CT (Table 6).

Accuracy analysis of dynamic AI, ACR TI-RADS, and CT (three non-invasive diagnostic methods) in nodules with different diameters

In total, 217 patients were diagnosed with 238 nodules using preoperative dynamic AI, the TI-RADS, and CT. The nodules were divided into two groups based on their size: the 2.0< diameter ≤4.0 cm group, and the diameter >4.0 cm group. There were 169 nodules in the 2.0< diameter ≤4.0 cm group, and 69 nodules in the diameter >4.0 cm group. There were no significant differences in the three non-invasive diagnostic methods between the two groups for different sizes of nodules (Table 7).

Discussion

Ultrasound is the preferred imaging examination for thyroid nodules, but traditional ultrasound is often limited by the resolution of instruments, the subjective experience of doctors, and the complexity of the nodules themselves, which may lead to misdiagnosis or missed diagnosis. AI plays a crucial role in the accurate diagnosis of thyroid nodules. It integrates core technologies such as AI algorithms and machine vision to achieve the high-throughput analysis of ultrasound imaging data, quickly and accurately identifying and annotating the location, size, shape, and other characteristics of thyroid nodules. It has significant advantages in improving diagnostic accuracy and efficacy, promoting precision medicine, and providing a more reliable basis for clinical decision making.

The dynamic AI-based ultrasound-assisted diagnosis system demonstrated good diagnostic efficacy in the diagnosis of thyroid nodules, and can locate and outline thyroid nodules in real time. Its multi-section real-time diagnosis feature can comprehensively and intuitively present nodule characteristics and benign and malignant diagnosis results. It has an accuracy of 89.97%, and a high consistency (kappa =0.737, P<0.001) (5). In previous studies, the number of nodules with a diameter >2 cm was relatively small. However, large nodules have more complex internal structures and surrounding tissues, making their characteristics atypical and often prone to misdiagnosis and missed diagnosis in clinical practice. To further explore the efficacy of dynamic AI in the diagnosis of larger nodules, we applied dynamic AI examination to the diagnosis of thyroid nodules with a diameter >2 cm, and studied its ability to differentiate between benign and malignant nodules, and its guiding significance for surgical treatment strategies.

This study applied dynamic AI to examine the benignity and malignancy of 297 thyroid nodules with a diameter >2 cm. Dynamic AI still maintained good stability, and there was no significant difference in the diagnostic accuracy of benign and malignant nodules among different age groups, regardless of whether HT was present or not. In the preliminary research, there was a statistically significant difference in the ability of dynamic AI to diagnose benign and malignant nodules, such that it was more accurate in diagnosing malignant nodules than benign nodules. This may be because patients with benign nodules often undergo follow-up treatment; thus, dynamic AI collected and learned less data from benign nodules. Subsequently, we strengthened the deep learning training of dynamic AI on benign nodules. Currently, the deep learning data of dynamic AI has reached 5 million cases, and dynamic AI showed good diagnostic stability in this study.

This study demonstrated that dynamic AI achieved high diagnostic performance (sensitivity: 86.23%, specificity: 86.16%, accuracy: 86.20%), showing high consistency with postoperative pathology (kappa =0.723, P<0.001), and a high diagnostic value for thyroid nodules >2 cm in diameter. Dynamic AI had a higher consistency with postoperative pathology than the TI-RADS, which relies on preoperative traditional ultrasound (kappa =0.567, P<0.001). Significant differences were found between dynamic AI and the TI-RADS in terms of specificity, accuracy, the PPV, and the misdiagnosis rate, indicating that dynamic AI is better in the diagnosis of benign thyroid nodules with a diameter >2 cm than preoperative traditional ultrasound. Changes in the internal structure of nodules (e.g., calcification, necrosis, and liquefaction) may occur during the process of nodule enlargement (12,13). These changes can affect the diagnoses of ultrasound doctors as to the benignity or malignancy of nodules. Dynamic AI can deeply analyze the subtle features inside large nodules from multiple sections and angles. Deep learning and algorithm optimization not only improves the accuracy of large nodule diagnosis but also reduces human errors due to individual subjective experience and operation, promoting the homogenization of diagnosis.

In this study, 162 of 297 nodules were diagnosed by FNAC before surgery. There were no significant differences between FNAC and dynamic AI in terms of the sensitivity, accuracy, NPV, and missed diagnosis rate, indicating that dynamic AI can achieve the same diagnostic accuracy as FNAC. The accurate diagnosis of malignant nodules by dynamic AI, along with the reduction of misdiagnosis and missed diagnosis, could aid in the early diagnosis and treatment of thyroid cancer. Compared with FNAC, dynamic AI is safe and non-invasive, which not only reduces medical costs but also shortens pre-hospital examinations and waiting times. Dynamic AI could help improve the diagnosis and treatment of thyroid nodules in areas or primary hospitals with relatively limited medical technology resources, where biopsies are difficult to perform. There were statistically significant differences between dynamic AI and FNAC in terms of specificity, the PPV, and the misdiagnosis rate. There may be two reasons for this finding. First, fewer patients with benign nodules undergo FNAC. In this study, 44 benign nodules underwent FNAC before surgery. Thus, large scale studies need to be conducted in the future. Second, when nodules are located in complex anatomical structures such as the clavicle and sternum, ultrasound examination may be affected by tissue structure interference. These nodules are often benign and larger in diameter, which also affects the diagnostic efficacy of dynamic AI. Therefore, if clinical doctors cannot comprehensively evaluate the benignity and malignancy of large nodules with complex anatomies through dynamic AI, multiple imaging examinations such as FNAC, CT, and magnetic resonance imaging should be combined to fully evaluate and formulate individualized diagnoses and treatment plans.

In this study, 238 nodules were examined by preoperative CT, with a sensitivity, specificity, and accuracy of 66.67%, 63.71%, and 65.13%, respectively. The consistency between CT and the postoperative pathological results was low (kappa =0.303, P<0.001), which reflects the “Guidelines for the Diagnosis and Treatment of Thyroid Nodules and Differentiated Thyroid Cancer” (2nd edition), which state that CT is not superior to ultrasound in evaluating the benignity and malignancy of thyroid nodules (7). Related studies have shown that the use of CT alone has significant limitations in determining the benignity and malignancy of thyroid nodules; however, when combined with ultrasound examination, CT can provide a more comprehensive assessment of nodule properties and improve diagnostic accuracy (14,15). Dynamic AI is superior to CT in the diagnosis of benign and malignant nodules, and can perform continuous scans at multiple sections and angles for the real-time diagnosis of nodules. This feature is similar to CT, but dynamic AI has a significant advantage in terms of its reduced radiation. Although CT has limited diagnostic value for benign and malignant nodules, preoperative CT examination is recommended for larger thyroid nodules requiring surgical treatment, particularly those with a complex anatomical structure and surrounding tissues, those that extend behind the sternum, and those at an advanced stage. CT can help evaluate the size and nature of the nodule, its relationship with important neck structures, and the presence of suspicious lymph nodes, which can aid in preoperative clinical staging, surgical plan formulation, and prognosis assessment.

In this study, we also examined the effect of nodule size on the accuracy of three non-invasive examinations: dynamic AI, preoperative ultrasound, and CT. There was no statistically significant difference in the diagnostic accuracy of the three non-invasive examinations in the 2.0< diameter ≤4.0 cm and diameter >4.0 cm groups, indicating that the three non-invasive examinations are not affected by nodule size in the diagnosis of thyroid nodules with diameter >2 cm. Among them, dynamic AI had the highest diagnostic accuracy of 88.17% in the 2.0< diameter ≤4.0 cm group, while dynamic AI and preoperative ultrasound had similar diagnostic accuracy for the diameter >4.0 cm group. In this study, we did not further group nodules larger than 4 cm, as only nine nodules were larger than 6.0 cm. We intend to increase the study sample size to conduct more in-depth research on this issue in the future.

Conclusions

Dynamic AI examination is non-invasive, safe, and objectively accurate. Enabling real-time diagnosis from multiple sections and angles, dynamic AI can deeply analyze the subtle features of large nodules, comprehensively evaluate thyroid nodules with a diameter >2 cm, achieve accurate diagnosis of benign and malignant thyroid nodules with a diameter >2 cm, promote diagnostic homogenization, and provide important guidance for individualized diagnosis and treatment strategies, and the development of surgical resection ranges. As learning data continues to increase, dynamic AI will demonstrate better diagnostic performance, and in the future, dynamic AI will have broader application prospects in the diagnosis of difficult and complex cases.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-253/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-253/dss

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-253/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Chinese PLA General Hospital (No. S2021-082-04) and informed consent was obtained from all patients.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Li K, Zhang H. Effect of Thyroid Nodule Size on the Diagnostic Accuracy of Ultrasound-Guided Fine-Needle Aspiration Cytology. Clinical Medicine & Engineering 2017;24:1355-7.
3. Tao W, Qingjun Z, Wei Z, Fang Z, Lei Z, Yuanyuan N, Kefu H. Computed tomography versus ultrasound/fine needle aspiration biopsy in differential diagnosis of thyroid nodules: a retrospective analysis. Braz J Otorhinolaryngol 2021;87:402-9. [Crossref] [PubMed]
4. Ding J, Zhang Y, Yan N. Research progress of real-time dynamic artificial intelligence in ultrasound diagnosis of thyroid cancer. Modern Medical Journal 2023;51:1645-50.
5. Wang B, Wan Z, Li C, Zhang M, Shi Y, Miao X, Jian Y, Luo Y, Yao J, Tian W. Identification of benign and malignant thyroid nodules based on dynamic AI ultrasound intelligent auxiliary diagnosis system. Front Endocrinol (Lausanne) 2022;13:1018321. [Crossref] [PubMed]
6. Wang B, Wan Z, Zhang M, Gong F, Zhang L, Luo Y, Yao J, Li C, Tian W. Diagnostic value of a dynamic artificial intelligence ultrasonic intelligent auxiliary diagnosis system for benign and malignant thyroid nodules in patients with Hashimoto thyroiditis. Quant Imaging Med Surg 2023;13:3618-29. [Crossref] [PubMed]
7. Chinese Society of Endocrinology; Thyroid and Metabolism Surgery Group of the Chinese Society of Surgery; China Anti-Cancer Association, Chinese Association of Head and Neck Oncology; Chinese Society of Nuclear Medicine; China Anti-Cancer Association, Chinese Association of Thyroid Oncology; Chinese Thyroid Association, Chinese Collage of Surgeons, Chinese Medical Doctor Association; Chinese Society of Ultrasound in Medicine. Guidelines for the diagnosis and management of thyroid nodules and differentiated thyroid cancer (Second edition). Chinese Journal of Endocrinology and Metabolism 2023;9:181-226.
8. Tessler FN, Middleton WD, Grant EG, Hoang JK, Berland LL, Teefey SA, Cronan JJ, Beland MD, Desser TS, Frates MC, Hammers LW, Hamper UM, Langer JE, Reading CC, Scoutt LM, Stavros AT. ACR Thyroid Imaging, Reporting and Data System (TI-RADS): White Paper of the ACR TI-RADS Committee. J Am Coll Radiol 2017;14:587-95. [Crossref] [PubMed]
9. Head and Neck Group of the Radiology Branch of the Chinese Medical Association. Expert consensus on the imaging examination process of thyroid nodules. Chin J Radiol 2016;50:911-5.
10. Cibas ES, Ali SZ. The 2017 Bethesda System for Reporting Thyroid Cytopathology. Thyroid 2017;27:1341-6. [Crossref] [PubMed]
11. Baloch ZW, Asa SL, Barletta JA, Ghossein RA, Juhlin CC, Jung CK. LiVolsi VA, Papotti MG, Sobrinho-Simões M, Tallini G, Mete O. Overview of the 2022 WHO Classification of Thyroid Neoplasms. Endocr Pathol 2022;33:27-63. [Crossref] [PubMed]
12. Ferreira LB, Gimba E, Vinagre J, Sobrinho-Simões M, Soares P. Molecular Aspects of Thyroid Calcification. Int J Mol Sci 2020;21:7718. [Crossref] [PubMed]
13. Zhu D, Chen W, Xiang J, Han Z. Diagnostic value of CT artifacts for solitary coarse calcifications in thyroid nodules. Int J Clin Exp Med 2015;8:5800-5.
14. Duan T, Yin J. The Imaging Findings and Diagnostic Value of Ultrasonography and CT for Patients with Thyroid Cancer. Chinese Journal of CT and MRI 2017;15:42-4.
15. Chen S. Value analysis of ultrasonography combined with CT in the diagnosis of nodular goiter complicated with thyroid cancer. Chinese Journal of Modern Drug Application 2023;17:94-6.