Artificial intelligence and elastography in diagnostic work-up of thyroid nodules: a systematic review and meta-analysis

Introduction

Thyroid cancer, which manifests as thyroid nodules, ranks as the most frequent endocrine malignancy. Notably, its prevalence is on the rise. A recent study revealed that from 2019 to 2030, the incidence of thyroid cancer will show a steady upward trend, with a 32.4% increase in men and a 13.1% increase in women (1,2). Therefore, it is important to distinguish benign and malignant thyroid lesions for the diagnosis and treatment of such patients. Histopathology is the gold standard for diagnosing thyroid cancer and serves as the definitive method for determining whether thyroid nodule are benign or malignant (3). However, this method is invasive, costly, and time-consuming. Fine-needle aspiration (FNA), a cytopathological technique, plays a crucial role in the diagnosis of thyroid cancer (4). As a minimally invasive approach for evaluating thyroid nodule, FNA is commonly used in preoperative assessment for thyroid cancer. Nevertheless, FNA has notable limitations: its sensitivity and specificity vary significantly, it may cause patient discomfort, and it carries certain risks of complications. Importantly, approximately 5–19% of lesions cannot be definitively classified as benign or malignant through FNA (5). Therefore, noninvasive methods are needed to evaluate thyroid nodule. Conventional ultrasound is the primary imaging method for identifying benign and malignant thyroid nodule, but it is operator-dependent and has variable diagnostic accuracy.

According to recent research, elastography is an ultrasound technique used to assess nodule stiffness (6). It offers additional information on tissue hardness compared to conventional ultrasound, enabling the detection of deeply located nodules. This addresses certain limitations of routine ultrasound, thus enhancing diagnostic objectivity. Elastography was first proposed in the United States in 1991 by Ophir et al. (7) and first applied to the thyroid in 2005 by Lyshchik et al. (8). Subsequently, shear wave elastography (SWE) was first reported for the diagnosis of thyroid nodule by Sebag et al. in 2010 (9). Currently, two main types of elastography are used in clinical practice, categorized into strain elastography (SE) and SWE based on their underlying physical principles. Substantial research has confirmed the clinical significance of SWE in distinguishing benign from malignant thyroid nodule (10,11). Zhang et al. (12) showed that combining contrast-enhanced ultrasound (CEUS) with SWE significantly improved diagnostic performance for Chinese Thyroid Imaging Reporting and Data System (C-TIRADS) category 4 nodules, with sensitivity reaching 94.6% and accuracy 92.2%. Additionally, combining SWE with grayscale imaging in multimodal examinations has achieved notable outcomes and markedly enhanced diagnostic accuracy, surpassing the capabilities of grayscale imaging alone (13,14).

Artificial intelligence (AI), a branch of computer science that encompasses machine learning (ML) and deep learning (DL) algorithms, is becoming increasingly prevalent in the medical field. Wang et al. (15) conducted a study on nodules larger than 2 cm in diameter and reported that the dynamic AI system achieved a diagnostic accuracy of 86.2%, with a high level of agreement with pathological results (kappa =0.723). This system enables real-time, multi-angle analysis of internal nodule features, offering an objective and non-invasive assessment tool for larger nodules, thereby helping to address the limitations of conventional ultrasound in the diagnosis of such lesions. Ma et al. (16) were the first to propose a thyroid nodule classification method based on convolutional neural network (CNN) fusion, achieving a diagnostic accuracy of 83.02%. As a core component of AI, the rapid development of ML, especially the most advanced DL frameworks, has opened up unprecedented possibilities for automatic medical image analysis tasks such as segmentation, detection, and classification (17). Tan et al. (18) developed a dual-stage DL framework that not only segments thyroid nodule but also classifies them into four specific pathological types (papillary carcinoma, medullary carcinoma, nodular goiter with adenomatous hyperplasia, and chronic lymphocytic thyroiditis), achieving an overall accuracy of 90.27% on the test set. S-Detect is a computer-aided diagnosis (CAD) system based on DL, integrated into ultrasound equipment. It uses CNN to automatically detect and analyze the features of thyroid nodule, such as their boundaries, shapes, and internal echoes, thereby reducing the reliance on operators and achieving an objective assessment of malignant thyroid nodule.

Alongside the increasing use of elastography as an adjunct to conventional ultrasound, there is a growing trend toward the use of AI for analyzing elastography images. However, the combined potential of elastography and AI in enhancing diagnostic outcomes remains unclear, and more research is needed to clarify the advantages and limitations of this integration. Therefore, this study aimed to synthesize findings from multiple studies to provide more comprehensive and reliable, evidence-based conclusions, using histopathologic diagnosis as the reference standard. This study aimed to evaluate, through a systematic review and meta-analysis, the diagnostic value of the combined application of AI and elastography in the differentiation of benign and malignant thyroid nodule. We present this article in accordance with the PRISMA-DTA reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1161/rc).

Methods

Before initiation, this study was registered in the International Prospective Register of Systematic Reviews (PROSPERO) with the ID CRD42024595399.

Search strategy and inclusion criteria

The databases of PubMed, Embase, Web of Science, and the China National Knowledge Infrastructure (CNKI) were searched from inception to 20 March 2025. The key Chinese search terms were “artificial intelligence”, “elastography”, “thyroid nodule”, “benign”, and “malignant”. The English search terms included “artificial intelligence”, “elastography”, “thyroid nodule”, “benign”, and “malignant”, along with their synonyms and free-text variants. For details, please refer to the complete PubMed search strategy provided in Table S1.

The inclusion criteria were as follows: (I) patients had histopathologic as the gold standard for diagnosis. (II) Data could be extracted directly or indirectly from the four-frame table: true positive (TP), false positive (FP), false negative (FN), and true negative (TN). (III) Patients were diagnosed by high-resolution thyroid ultrasound. (IV) Combined diagnosis by AI and elastography. The exclusion criteria were as follows: (I) secondary publications (e.g., reviews, meta-analyses, and conference abstracts). (II) Full text was not available. (III) Incomplete data provided. (IV) Studies with an inappropriate study design. (V) Repeatedly published literature.

Data extraction and quality assessment

Two reviewers independently performed the literature selection and data extraction. Any disagreements encountered during these processes were resolved through discussion or, when necessary, by consulting a third senior researcher. The extracted data included the first author, year of publication, country, instrument, diagnostic method, sample size, number of thyroid nodules, and the values for TP, FP, FN, and TN. The methodological quality of the included studies was assessed independently by the same two reviewers using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. The QUADAS-2 scale is divided into four sections, including “risk of bias in case selection”, “risk of bias in diagnostic tests to be evaluated”, “risk of bias in gold standard”, and “risk of bias in case flow and progression”. For each domain, the risk of bias was judged as ‘high’, ‘low’, or ‘unclear’. Clinical concerns regarding applicability were assessed for the first three domains only. A summary of the risk of bias was generated using Review Manager 5.4 software (Cochrane Collaboration, London, UK).

Statistical analysis

The statistical analysis was performed using Stata 17 (StataCorp., College Station, TX, USA). Based on the extracted data of TP, FP, FN, and TN, we calculated the following pooled effect sizes along with their 95% confidence intervals (CIs): sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR). The accuracy of AI combined with elastography for thyroid nodule diagnosis was evaluated by drawing the receiver operating characteristic (ROC) curve and calculating the area under the curve (AUC) using Stata software. When the AUC ≥0.5, the closer the AUC is to 1, the higher the accuracy of the diagnostic test; on the contrary, the closer the AUC is to 0.5, the lower the accuracy of the diagnostic test. To explore potential sources of heterogeneity, we performed univariable meta-regression and subgroup analyses based on publication year, total sample size, and elastography type. For each subgroup, we calculated the pooled effect size with its 95% CI. A P value of less than 0.05 for the interaction term was considered indicative of a statistically significant subgroup effect. Sensitivity analysis was conducted by sequentially excluding each study. The results demonstrated that the pooled effect size and its significance (P<0.05) remained consistent before and after each exclusion, confirming the robustness and stability of our primary findings. Stata software was used to draw Deeks’ funnel plot to analyze the potential publication bias of the included literature, and P>0.05 suggested that there was no publication bias.

Results

Search results and study characteristics

The literature search across four databases yielded 188 articles (32 Chinese, 156 English). The screening process was as follows: 60 duplicates were removed using EndNote (Clarivate, Ann Arbor, MI, USA); 45 records (e.g., reviews, conference abstracts) were excluded based on title and abstract; and 64 articles were excluded after full-text assessment for not meeting inclusion criteria or lacking complete diagnostic data. This process resulted in the final inclusion of 19 articles (six Chinese, 13 English). For details, see the flowchart of literature screening in Figure 1. Data extraction from the 19 studies included: first author, publication year, country, instrument, diagnostic method, sample size, number of nodules, and TP, FP, FN, and TN values. See the details of included articles in Table 1.

Figure 1 Flowchart of literature screening.

Assessment of study quality

The methodological quality of the included studies was assessed using the QUADAS-2 tool. In the patient selection domain, 15 studies were rated as low risk of bias and low concern regarding applicability, and four studies were rated as high risk. For the index test, 17 studies demonstrated low risk of bias and low applicability concerns, with the remaining two studies posing unclear risk. Regarding the reference standard, all 19 studies were judged as low risk of bias and low concern. In the flow and timing domain, 17 studies were rated as low risk, and two studies were rated as high risk. For details, see Figure 2 risk of bias and applicability of included literature.

Figure 2 Risk of bias and applicability of included literature: summary of QUADAS-2 assessments of included literature; proportion of studies with low, high, or unclear risk of bias. (A) Each row represents a study, each column an assessment domain (patient selection, index test, reference standard, flow and timing), and colors indicate risk levels (green=low, red=high, yellow=unclear); (B) Proportion of bias risk and applicability in each domain: bar chart shows the percentage of studies with different risk levels in each domain. QUADAS-2, Quality Assessment of Diagnostic Accuracy Studies-2.

Assessment of diagnostic accuracy and heterogeneity

Diagnostic performance results from the 19 included studies demonstrated that the combination of AI and elastography for thyroid nodule assessment achieved a pooled sensitivity of 0.88 (95% CI: 0.84–0.91) and specificity of 0.91 (95% CI: 0.88–0.94). The PLR was 9.78 (95% CI: 6.89–13.89), the NLR was 0.13 (95% CI: 0.10–0.18), and the DOR was 73.89 (95% CI: 40.49–134.85). The overall discriminative ability, quantified by the summary receiver operating characteristic (SROC) curve, was excellent, with an AUC of 0.95 (95% CI: 0.93–0.97). Notable heterogeneity was present in these pooled estimates, as indicated by the I² statistics: 80.63% for sensitivity, 85.94% for specificity, 81.52% for PLR, 82.69% for NLR, and 100% for DOR. The forest plots are presented in Figures 3-5 and the SROC in Figure 6.

Figure 3 Forest plot of combined sensitivity and specificity. CI, confidence interval.

Figure 4 Forest plot of combined positive and negative likelihood ratios. CI, confidence interval; DLR, diagnostic likelihood ratio.

Figure 5 Forest plot of combined diagnostic advantage ratio. CI, confidence interval.

Figure 6 Composite subject job characterization curve. AUC, area under the curve; CI, confidence interval; Sens, sensitivity; Spec, specificity; SROC, summary receiver operating characteristic.

Meta-regression and subgroup analysis

Meta-regression and subgroup analyses were conducted to explore the potential effects of publication year, total sample size, and elastography type on the pooled results. Analysis based on publication year showed a statistically significant difference between subgroups (P<0.01). The heterogeneity (I²) substantially decreased to 59% in the subgroup of studies published before 2024, indicating that that publication year was a potential source of heterogeneity in this meta-analysis. In contrast, subgroup analyses based on total sample size and elastography type did not show significant differences (P>0.05). The detailed results of the meta-regression and subgroup analyses are provided in Table 2.

Sensitivity analysis and publication bias

To evaluate the robustness of the meta-analysis results, a sensitivity analysis was conducted by sequentially excluding each study. The pooled sensitivity and specificity estimates remained stable throughout this process, indicating that no single study disproportionately influenced the overall results (Table 3). Stata software was used to draw Deeks’ funnel plot to analyze the potential publication bias of the included literature, and P>0.05 suggested that there was no publication bias. Deeks’ funnel plot of publication bias is displayed in Figure 7.

Figure 7 Publication bias Deeks’ funnel plot. ESS, effective sample size.

Discussion

This study provides a comprehensive assessment, integrating 19 studies via meta-analysis, to evaluate the diagnostic performance of AI combined with elastography in differentiating benign and malignant thyroid nodule. The pooled results showed excellent diagnostic accuracy, with a sensitivity of 0.88 (95% CI: 0.84–0.91), a specificity of 0.91 (95% CI: 0.88–0.94), and an AUC of 0.95 (95% CI: 0.93–0.97). This study showed heterogeneity, and the results indicated that the year 2024 was a potential source of heterogeneity.

The high diagnostic efficacy observed in this study, with an AUC of 0.95, is consistent with findings in recent literature. Zhao et al. (35) reported that an AI model incorporating SWE parameters significantly reduced the unnecessary biopsy rate from 30% to 4.5%, highlighting its considerable clinical utility. Namsena et al. (38) demonstrated in a multicenter retrospective study that an AI system achieved a sensitivity of 80% and specificity of 71.4% in interpreting thyroid nodule on ultrasound images, significantly outperforming experienced radiologists in sensitivity (40%). This highlights AI’s potential to reduce subjective interpretation and decrease unnecessary FNA procedures. Similarly, Qin et al. (25) developed an end-to-end CNN model that transferred feature parameters from VGG16 (pre-trained on ImageNet) to ultrasound images, utilizing hybrid features from both B-mode and SE images. Their method achieved an accuracy of 0.947, surpassing the performance of models based on a single data source.

Although the overall findings are consistent with existing evidence, the pooled sensitivity in our analysis is notably higher than the 72.26% reported by Zhou et al. (39) for elastography alone. This discrepancy can be attributed to two key factors.

First, a fundamental difference lies in the elastography techniques. Zhou et al. likely employed operator-dependent SE, whereas the studies in our meta-analysis predominantly used SWE, which provides quantitative and more objective parameters (e.g., shear wave velocity). Second, the integration of AI is crucial. Although earlier studies often relied on traditional ML, our included literature predominantly features advanced DL architectures, which excel at automatically extracting complex features. These include characteristics such as subtle infiltrative patterns at nodule margins, a capability that directly enhances diagnostic sensitivity. The clinical applicability of such AI assistance is further validated by Szczepanek-Parulska et al. (40), who demonstrated that the S-Detect system outperformed experienced sonographers in specificity, positive predictive value (PPV), and accuracy. This aligns with our findings, indicating that DL models can effectively support physicians in nodule identification and characterization, thereby streamlining the diagnostic process.

This study provides the first systematic evidence and a validated diagnostic framework for integrating AI with multimodal elastography to address the challenge of inconsistent imaging manifestations in thyroid nodule. The high diagnostic accuracy of this approach directly supports the American Thyroid Association (ATA) initiative to reduce unnecessary procedures for thyroid nodule. Practically, the AI system performs nodule segmentation and risk stratification within seconds. By standardizing assessments with quantitative elastography parameters, it demonstrates significant potential to reduce inter-observer and inter-institutional variability, thereby elevating diagnostic standards, particularly in primary care settings. Consequently, this work offers high-level evidence for a field that currently lacks comprehensive synthesis and robust clinical validation.

There are some limitations of our meta-analysis that need to be addressed. First, the included elastic imaging devices vary in models, and their device-specific calibration models require further validation. Second, as the meta-analysis encompassed both retrospective and prospective studies; uniform patient selection could not be guaranteed. In the future, determining the diagnostic accuracy of combination of AI and elastography could be improved by only including prospective randomized controlled trials in the analysis.

Conclusions

This meta-analysis evaluated the combination of AI and ultrasound elastography to improve thyroid diagnostic accuracy. The findings indicate that integrating any ultrasound elastography technique with AI can markedly enhance diagnostic sensitivity and specificity. Consequently, we advocate for this combination as a noninvasive diagnostic approach. It boosts clinical diagnostic precision, lessens tissue damage from unwarranted biopsies, diminishes the clinical risks tied to FNA, and aids in the early diagnosis and treatment of malignant thyroid nodule.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the PRISMA-DTA reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1161/rc

Funding: This study was funded by Science and Technology Department of Sichuan Province Project (grant/award No. 2023NSFSC0038).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1161/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.

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. Shank JB, Are C, Wenos CD. Thyroid Cancer: Global Burden and Trends. Indian J Surg Oncol 2022;13:40-5. [Crossref] [PubMed]
2. Wu J, Zhao X, Sun J, Cheng C, Yin C, Bai R. The epidemic of thyroid cancer in China: Current trends and future prediction. Front Oncol 2022;12:932729. [Crossref] [PubMed]
3. Bernet VJ, Chindris AM. Update on the Evaluation of Thyroid Nodules. J Nucl Med 2021;62:13S-9S. [Crossref] [PubMed]
4. Tanaka A, Hirokawa M, Suzuki A, Higuchi M, Kanematsu R, Yamao N, Kuma S, Hayashi T, Miyauchi A. Clinical significance and cytological detection of tracheal puncture following thyroid fine-needle aspiration: A retrospective study. Diagn Cytopathol 2021;49:1116-21. [Crossref] [PubMed]
5. Liu Y, Liu H, Zhan J, Chai Q, Zhu J, Ding S, Chen L. Contrast-Enhanced Ultrasound for Diagnosing Thyroid Nodules With Indeterminate Cytology: A Retrospective Study. Clin Endocrinol (Oxf) 2025;102:223-31. [Crossref] [PubMed]
6. Yu L, Che M, Wu X, Luo H. Research on ultrasound-based radiomics: a bibliometric analysis. Quant Imaging Med Surg 2024;14:4520-39. [Crossref] [PubMed]
7. Ophir J, Céspedes I, Ponnekanti H, Yazdi Y, Li X. Elastography: a quantitative method for imaging the elasticity of biological tissues. Ultrason Imaging 1991;13:111-34. [Crossref] [PubMed]
8. Lyshchik A, Higashi T, Asato R, Tanaka S, Ito J, Mai JJ, Pellot-Barakat C, Insana MF, Brill AB, Saga T, Hiraoka M, Togashi K. Thyroid gland tumor diagnosis at US elastography. Radiology 2005;237:202-11. [Crossref] [PubMed]
9. Sebag F, Vaillant-Lombard J, Berbis J, Griset V, Henry JF, Petit P, Oliver C. Shear wave elastography: a new ultrasound imaging mode for the differential diagnosis of benign and malignant thyroid nodules. J Clin Endocrinol Metab 2010;95:5281-8. [Crossref] [PubMed]
10. Cosgrove D, Barr R, Bojunga J, Cantisani V, Chammas MC, Dighe M, Vinayak S, Xu JM, Dietrich CF. WFUMB Guidelines and Recommendations on the Clinical Use of Ultrasound Elastography: Part 4. Thyroid. Ultrasound Med Biol 2017;43:4-26. [Crossref] [PubMed]
11. Săftoiu A, Gilja OH, Sidhu PS, Dietrich CF, Cantisani V, Amy D, et al. The EFSUMB Guidelines and Recommendations for the Clinical Practice of Elastography in Non-Hepatic Applications: Update 2018. Ultraschall Med 2019;40:425-53. [Crossref] [PubMed]
12. Zhang WB, Xu W, He BL, Chen Z, Liu H, Deng WF. Contrast-enhanced ultrasound combined with shear wave elastography in the diagnosis of C-TIRADS category 4 thyroid nodules. Quant Imaging Med Surg 2025;15:4113-21. [Crossref] [PubMed]
13. Kim H, Kim JA, Son EJ, Youk JH. Quantitative assessment of shear-wave ultrasound elastography in thyroid nodules: diagnostic performance for predicting malignancy. Eur Radiol 2013;23:2532-7. [Crossref] [PubMed]
14. Hang J, Li F, Qiao XH, Ye XH, Li A, Du LF. Combination of Maximum Shear Wave Elasticity Modulus and TIRADS Improves the Diagnostic Specificity in Characterizing Thyroid Nodules: A Retrospective Study. Int J Endocrinol 2018;2018:4923050. [Crossref] [PubMed]
15. Wang B, Dong X, Ding J, Wan Z, Miao X, Yang Z, Jian Y, Zhang L, Li C, Zhang M, Yao J, Tian W. 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. Quant Imaging Med Surg 2025;15:9644-55. [Crossref] [PubMed]
16. Ma J, Wu F, Zhu J, Xu D, Kong D. A pre-trained convolutional neural network based method for thyroid nodule diagnosis. Ultrasonics 2017;73:221-30. [Crossref] [PubMed]
17. Chartrand G, Cheng PM, Vorontsov E, Drozdzal M, Turcotte S, Pal CJ, Kadoury S, Tang A. Deep Learning: A Primer for Radiologists. Radiographics 2017;37:2113-31. [Crossref] [PubMed]
18. Tan D, Zhai Y, Hu Z, Xu B, Zheng T, Chen Y, Sun D. Dual-stage artificial intelligence-powered screening for accurate classification of thyroid nodules: enhancing fine needle aspiration biopsy precision. Quant Imaging Med Surg 2025;15:5719-38. [Crossref] [PubMed]
19. Barzegar-Golmoghani E, Mohebi M, Gohari Z, Aram S, Mohammadzadeh A, Firouznia S, Shakiba M, Naghibi H, Moradian S, Ahmadi M, Almasi K, Issaiy M, Anjomrooz M, Tavangar SM, Javadi S, Bitarafan-Rajabi A, Davoodi M, Sharifian H, Mohammadzadeh M. ELTIRADS framework for thyroid nodule classification integrating elastography, TIRADS, and radiomics with interpretable machine learning. Sci Rep 2025;15:8763. [Crossref] [PubMed]
20. Chen Z, Chambara N, Lo X, Liu SYW, Gunda ST, Han X, Ying MTC. Improving the diagnostic strategy for thyroid nodules: a combination of artificial intelligence-based computer-aided diagnosis system and shear wave elastography. Endocrine 2025;87:744-57. [Crossref] [PubMed]
21. Chu X, Wang T, Chen M, Li J, Wang L, Wang C, Wang H, Wong ST, Chen Y, Li H. Deep learning model for malignancy prediction of TI-RADS 4 thyroid nodules with high-risk characteristics using multimodal ultrasound: A multicentre study. Comput Med Imaging Graph 2025;124:102576. [Crossref] [PubMed]
22. Cui S, Liu Q, Wang H, Li H, Li W, Li C, Bi L, Mu Y, Guo W, Yao J, Zhang Z. The value of a combined model based on ultra-radiomics and multi-modal ultrasound in the benign-malignant differentiation of C-TIRADS 4A thyroid nodules: a prospective multicenter study. Front Oncol 2025;15:1543020. [Crossref] [PubMed]
23. Li N, Zhang H. Study on the AIAS Score and Strain Rate Ratio in the Clinical Diagnosis of Benign and Malignant Thyroid Nodules With the Application of Artificial Intelligence-Assisted Scoring System Combined With Ultrasound Elastography Diagnosis. Imaging Res Med Appl 2022;6:73-5.
24. Liu T, Duan X, Niu H, Wang W. Clinical Value of Artificial Intelligence Assisted Score Combined With Shear Wave Elastography in the Differential Diagnosis of Benign and Malignant Thyroid Nodules. Journal of Clinical Ultrasound in Medicine 2024;26:828-32.
25. Qin P, Wu K, Hu Y, Zeng J, Chai X. Diagnosis of Benign and Malignant Thyroid Nodules Using Combined Conventional Ultrasound and Ultrasound Elasticity Imaging. IEEE J Biomed Health Inform 2020;24:1028-36. [Crossref] [PubMed]
26. Song L, Jiang J, Wang J, Zhou Q. Value of Artificial Intelligence Assisted Scoring System Combined With Ultrasound Elastography in the Diagnosis of Benign and Malignant Thyroid Nodules. Journal of Clinical Medicine in Practice 2021;25:7-10.
27. Sun H, Yu F, Xu H. Discriminating the Nature of Thyroid Nodules Using the Hybrid Method. Math Probl Eng 2020;2020:6147037.
28. Tao Y, Yu Y, Wu T, Xu X, Dai Q, Kong H, Zhang L, Yu W, Leng X, Qiu W, Tian J. Deep learning for the diagnosis of suspicious thyroid nodules based on multimodal ultrasound images. Front Oncol 2022;12:1012724. [Crossref] [PubMed]
29. Tuo J, Si X, Song H. Artificial intelligence technology enhances the performance of shear wave elastography in thyroid nodule diagnosis. Am J Transl Res 2023;15:6226-33.
30. Wu AD, Wang J, Song YM, Chen R, Li WJ. Application of Ultrasound Multimodal Artificial Intelligence Technology in the Intelligent Diagnosis of Thyroid Cancer. Journal of Cancer Control and Treatment 2024;37:220-5.
31. Wang X, Chen L, Song J. Value of Ultrasonic Intelligent Diagnosis Technology Combined With Elastic Imaging in Identification of Benign and Malignant Thyroid Nodules. Chin Community Dr 2022;38:79-81.
32. Wang T, Liu Y. Value of Artificial Intelligence S-Detect Combined With Ultrasound Elastography for Differential Diagnosis of Benign and Malignant Thyroid Nodules. Journal of Aerospace Medicine 2023;34:916-8.
33. Zhang B, Tian J, Pei S, Chen Y, He X, Dong Y, Zhang L, Mo X, Huang W, Cong S, Zhang S. Machine Learning-Assisted System for Thyroid Nodule Diagnosis. Thyroid 2019;29:858-67. [Crossref] [PubMed]
34. Zhang P, Xu Q, Jiang F. The diagnostic value of convolutional neural networks in thyroid cancer detection using ultrasound images. Front Oncol 2025;15:1534228. [Crossref] [PubMed]
35. Zhao CK, Ren TT, Yin YF, Shi H, Wang HX, Zhou BY, Wang XR, Li X, Zhang YF, Liu C, Xu HX. A Comparative Analysis of Two Machine Learning-Based Diagnostic Patterns with Thyroid Imaging Reporting and Data System for Thyroid Nodules: Diagnostic Performance and Unnecessary Biopsy Rate. Thyroid 2021;31:470-81. [Crossref] [PubMed]
36. Zhao HN, Liu JY, Lin QZ, He YS, Luo HH, Peng YL, Ma BY. Partially cystic thyroid cancer on conventional and elastographic ultrasound: a retrospective study and a machine learning-assisted system. Ann Transl Med 2020;8:495. [Crossref] [PubMed]
37. Zou D, Lyu F, Pan Y, Fan X, Du J, Mai X. Enhancing diagnostic precision for thyroid C-TIRADS category 4 nodules: a hybrid deep learning and machine learning model integrating grayscale and elastographic ultrasound features. Quant Imaging Med Surg 2025;15:7951-63. [Crossref] [PubMed]
38. Namsena P, Songsaeng D, Keatmanee C, Klabwong S, Kunapinun A, Soodchuen S, Tarathipayakul T, Tanasoontrarat W, Ekpanyapong M, Dailey MN. Diagnostic performance of artificial intelligence in interpreting thyroid nodules on ultrasound images: a multicenter retrospective study. Quant Imaging Med Surg 2024;14:3676-94. [Crossref] [PubMed]
39. Zhou Y, Chen H, Qiang J, Wang D. Systematic review and meta-analysis of ultrasonic elastography in the diagnosis of benign and malignant thyroid nodules. Gland Surg 2021;10:2734-44. [Crossref] [PubMed]
40. Szczepanek-Parulska E, Wolinski K, Dobruch-Sobczak K, Antosik P, Ostalowska A, Krauze A, Migda B, Zylka A, Lange-Ratajczak M, Banasiewicz T, Dedecjus M, Adamczewski Z, Slapa RZ, Mlosek RK, Lewinski A, Ruchala M. S-Detect Software vs. EU-TIRADS Classification: A Dual-Center Validation of Diagnostic Performance in Differentiation of Thyroid Nodules. J Clin Med 2020;9:2495. [Crossref] [PubMed]