Thyroid Nodule Classification for Physician Decision Support Using Machine Learning-Evaluated Geometric and Morphological Features

A thyroid nodule, a common abnormal growth within the thyroid gland, is often identified through ultrasound imaging of the neck. These growths may be solid- or fluid-filled, and their treatment is influenced by factors such as size and location. The Thyroid Imaging Reporting and Data System (TI-RADS) is a classification method that categorizes thyroid nodules into risk levels based on features such as size, echogenicity, margin, shape, and calcification. It guides clinicians in deciding whether a biopsy or other further evaluation is needed. Machine learning (ML) can complement TI-RADS classification, thereby improving the detection of malignant tumors. When combined with expert rules (TI-RADS) and explanations, ML models may uncover elements that TI-RADS misses, especially when TI-RADS training data are scarce. In this paper, we present an automated system for classifying thyroid nodules according to TI-RADS and assessing malignancy effectively. We use ResNet-101 and DenseNet-201 models to classify thyroid nodules according to TI-RADS and malignancy. By analyzing the models' last layer using the Grad-CAM algorithm, we demonstrate that these models can identify risk areas and detect nodule features relevant to the TI-RADS score. By integrating Grad-CAM results with feature probability calculations, we provide a precise heat map, visualizing specific features within the nodule and potentially assisting doctors in their assessments. Our experiments show that the utilization of ResNet-101 and DenseNet-201 models, in conjunction with Grad-CAM visualization analysis, improves TI-RADS classification accuracy by up to 10%. This enhancement, achieved through iterative analysis and re-training, underscores the potential of machine learning in advancing thyroid nodule diagnosis, offering a promising direction for further exploration and clinical application.

Thyroid nodules are a common and prevalent problem encountered in every day clinical practice. Ultrasound (US) is considered to be the initial imaging modality used in the evaluation of thyroid nodules. The TI-RADS (Thyroid Image Reporting and Data system) is an US-based system used to categorize thyroid nodules and stratify their malignant risk in the aim to standardize reporting and simplify communication among practitioners. The aim of this study is to compare the TI-RADS classification of thyroid nodules on ultrasound with the findings of fine-needle aspiration cytology (FNAC) reported using the Bethesda system. A retrospective study of 59 thyroid nodules that underwent single radiologist bedside neck and thyroid US over a period of one year between January 2020 – January 2021 was performed. The US findings were based on the TI-RADS and all the nodules were eventually subjected to US-guided FNAC. Clinico-pathological retrospective analysis was performed by comparing TI-RADS findings to the Bethesda FNAC classification. Comparing TI-RADS results with the Bethesda system of classification, the risk of malignancy for TI-RADS 2, TI-RADS 3, TI-RADS 4, and TI-RADS 5 were 0%, 10%, 50%, and 100%, respectively. A significant association was noted between the TI-RADS and Bethesda system of classification (P < 0.001). We derived 91.3% sensitivity, 88.9% specificity, 84% positive predictive value, and 94.1% negative predictive value for our study. There is a strong correlation between thyroid sonographic reporting using the TI-RADS and Bethesda cytological classification system for thyroid nodules. The TI-RADS classification is an effective method for risk stratification of thyroid nodules into categories predictive of their malignant potential. This aims to guide further management and facilitate the selection process for fine-needle aspiration biopsy, thus avoiding unnecessary procedures.

Background: Thyroid nodules are common, with up to 68% of adults having them. Ultrasonography is commonly used to examine them, but cannot determine if they are benign or malignant. High-res ultrasonography can detect disease before diagnosis but confirmation requires an invasive procedure, fine-needle aspiration cytology (FNAC), which has risks. Thyroid Imaging Reporting and Data System (TI-RADS) classifies nodules based on ultrasound findings, providing objectivity and guidance. Aim: The aim of this study is to evaluate the ultrasound characteristics of thyroid nodules & categorize them using the ACR-TIRADS category based on suspicious ultrasound findings, assess the malignancy risk and co-relate with the pathological(FNAC) findings. Materials and Methods: A study involving 50 participants was conducted in the department of radiodiagnosis at our institute to assess thyroid nodules using ultrasound. The patients underwent ultrasound examination and FNAC, and the results were analysed using Microsoft Excel 2016. Results: Out of the 50 cases, 88% had single nodules and 72% were classified as TI-RADS 3, 20% as TI-RADS 4 and 8% as TI-RADS 5. The FNAC results revealed benign and malignant nodules. All patients with TI-RADS 3 criteria were benign, 80% of those with TI-RADS 4 criteria were benign and all patients with TI-RADS 5 criteria were malignant. Conclusion: The study concludes that the American College of Radiology (ACR) TI-RADS scoring system is an effective and practical way to categorise thyroid nodules using ultrasound. It provides a level of certainty about the likelihood of malignancy and helps healthcare providers make informed decisions about which nodules require biopsy, monitoring or no further action based on their ultrasound characteristics. This system reduces the need for unnecessary biopsies and surgeries, making it a cost-effective solution.

Background: Thyroid nodules are increasingly common and pose a diagnostic challenge, particularly in distinguishing benign from malignant lesions. In low-resource settings like Pakistan, effective and accessible diagnostic tools are essential to guide appropriate clinical management and avoid unnecessary invasive procedures. The Thyroid Imaging Reporting and Data System (TIRADS) offers a standardized ultrasonographic approach for risk stratification of thyroid nodules. However, its validation in local populations remains limited. Objective: To assess the diagnostic performance of TIRADS in differentiating benign and malignant thyroid nodules in a Pakistani population, using fine-needle aspiration cytology (FNAC) as the reference standard. Methods: A cross-sectional validation study was conducted at Jinnah Postgraduate Medical Centre, Karachi, from April to September 2024. A total of 100 patients with thyroid nodules were enrolled. Ultrasound examinations were performed by an experienced radiologist using a 7–15 MHz transducer. Nodules were categorized according to TIRADS, and all participants underwent FNAC for histological confirmation. Diagnostic metrics—sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and overall accuracy—were calculated. Stratified analysis was conducted based on gender, age, location, and thyroid function status. Results: The mean age of participants was 46.91 ± 18.25 years, with a slight female predominance (56%). Malignancy was identified in 64% of cases on ultrasound and 78% on FNAC. TIRADS demonstrated 80.8% sensitivity, 95.5% specificity, 98.4% PPV, 58.3% NPV, and 84.0% overall accuracy. TIRADS 5 was the most frequent category (24%). Higher sensitivity was noted among rural residents (85.7%) and females (84.1%), while specificity declined in patients over 60 years (80%). Conclusion: TIRADS offers high diagnostic accuracy in confirming thyroid malignancy and is a valuable triage tool in resource-limited settings. However, its limited NPV warrants cautious interpretation of low-risk classifications. Local validation and radiologist training are essential for optimized use. Background: Thyroid nodules are increasingly common and pose a diagnostic challenge, particularly in distinguishing benign from malignant lesions. In low-resource settings like Pakistan, effective and accessible diagnostic tools are essential to guide appropriate clinical management and avoid unnecessary invasive procedures. The Thyroid Imaging Reporting and Data System (TIRADS) offers a standardized ultrasonographic approach for risk stratification of thyroid nodules. However, its validation in local populations remains limited. Objective: To assess the diagnostic performance of TIRADS in differentiating benign and malignant thyroid nodules in a Pakistani population, using fine-needle aspiration cytology (FNAC) as the reference standard. Methods: A cross-sectional validation study was conducted at Jinnah Postgraduate Medical Centre, Karachi, from April to September 2024. A total of 100 patients with thyroid nodules were enrolled. Ultrasound examinations were performed by an experienced radiologist using a 7–15 MHz transducer. Nodules were categorized according to TIRADS, and all participants underwent FNAC for histological confirmation. Diagnostic metrics—sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and overall accuracy—were calculated. Stratified analysis was conducted based on gender, age, location, and thyroid function status. Results: The mean age of participants was 46.91 ± 18.25 years, with a slight female predominance (56%). Malignancy was identified in 64% of cases on ultrasound and 78% on FNAC. TIRADS demonstrated 80.8% sensitivity, 95.5% specificity, 98.4% PPV, 58.3% NPV, and 84.0% overall accuracy. TIRADS 5 was the most frequent category (24%). Higher sensitivity was noted among rural residents (85.7%) and females (84.1%), while specificity declined in patients over 60 years (80%). Conclusion: TIRADS offers high diagnostic accuracy in confirming thyroid malignancy and is a valuable triage tool in resource-limited settings. However, its limited NPV warrants cautious interpretation of low-risk classifications. Local validation and radiologist training are essential for optimized use.

Background The detection of thyroid nodules has recently increased due to widespread use of ultrasonography. The main concern in the evaluation of thyroid nodules is the detection of malignancy. Ultrasonography is a widely accepted imaging modality for the initial assessment of thyroid nodules and helps in differentiating benign from malignant nodules. FNAC is considered as a reliable and effective method in differentiating benign and malignant nodules. Thyroid nodules show a highly diverse pattern on ultrasonography and there is an overlap between the ultrasound features of various thyroid lesions which causes difculty in classication of these nodules into benign and malignant categories. There is lack of a unied system to classify thyroid lesions clearly into benign and malignant categories. Horvath et al proposed an evaluation system for thyroid nodules called TIRADS (Thyroid Imaging Reporting and Data System). Objectives 1. To evaluate thyroid nodules using B-Mode and color Doppler ultrasound. 2. To classify these nodules based on TIRADS. 3. To correlate the ndings of ultrasound with ne needle aspiration cytology / histopathology. Materials and methods Patients who were referred to department of radiodiagnosis for USG investigation of thyroid in Pacic Institute of Medical Sciences, Udaipur, Rajasthan, over the time period from 1st March, 2021 to 30th September, 022.Thyroid nodules were evaluated for various ultrasound features and categorized using the TIRADS. These ndings were correlated with histopathology/ cytopathology ndings. Results Out of 118 patients, 101 were benign and 17 nodules were malignant. The nodules were evaluated for ultrasound features like hypo echogenicity, lobulated or poorly dened margins, micro calcications, taller than wider shape, central or predominant, central pattern of vascularity and solid composition; which showed a statistically signicant association with malignancy (p<0.05). The TIRADS classication has high sensitivity, specicity, and diagnostic accuracy in differentiating benign and malignant thyroid nodules. The risk of malignancy in our study for the TIRADS categories 2, 3, 4A, 4B, 4C and 5 were 0%, 1.5%, 4.8%, 63%, 100% and 100% respectively. Conclusion B mode ultrasound with colour Doppler is the most sensitive and specic imaging modality for initial assessment of thyroid nodules. Using ultrasound; TIRADS helps in differentiating benign and malignant thyroid nodules and standardizes the evaluation of thyroid nodules.

Background Risk stratification systems for thyroid nodules are often complicated and affected by low specificity. Continual improvement of these systems is necessary to reduce the number of unnecessary thyroid biopsies. Purpose To use artificial intelligence (AI) to optimize the American College of Radiology (ACR) Thyroid Imaging Reporting and Data System (TI-RADS). Materials and Methods A total of 1425 biopsy-proven thyroid nodules from 1264 consecutive patients (1026 women; mean age, 52.9 years [range, 18-93 years]) were evaluated retrospectively. Expert readers assigned points based on five ACR TI-RADS categories (composition, echogenicity, shape, margin, echogenic foci), and a genetic AI algorithm was applied to a training set (1325 nodules). Point and pathologic data were used to create an optimized scoring system (hereafter, AI TI-RADS). Performance of the systems was compared by using a test set of the final 100 nodules with interpretations from the expert reader, eight nonexpert readers, and an expert panel. Initial performance of AI TI-RADS was calculated by using a test for differences between binomial proportions. Additional comparisons across readers were conducted by using bootstrapping; diagnostic performance was assessed by using area under the receiver operating curve. Results AI TI-RADS assigned new point values for eight ACR TI-RADS features. Six features were assigned zero points, which simplified categorization. By using expert reader data, the diagnostic performance of ACR TI-RADS and AI TI-RADS was area under the receiver operating curve of 0.91 and 0.93, respectively. For the same expert, specificity of AI TI-RADS (65%, 55 of 85) was higher (P < .001) than that of ACR TI-RADS (47%, 40 of 85). For the eight nonexpert radiologists, mean specificity for AI TI-RADS (55%) was also higher (P < .001) than that of ACR TI-RADS (48%). An interactive AI TI-RADS calculator can be viewed at http://deckard.duhs.duke.edu/∼ai-ti-rads . Conclusion An artificial intelligence-optimized Thyroid Imaging Reporting and Data System (TI-RADS) validates the American College of Radiology TI-RADS while slightly improving specificity and maintaining sensitivity. Additionally, it simplifies feature assignments, which may improve ease of use. © RSNA, 2019 Online supplemental material is available for this article.

This study evaluates the value of ultrasound in classifying hypoechoic nodules of thyroid of 70 patients at the Department of Diagnostic Imaging and Oncology of E Hospital from January 2019 to December 2019. The study results show that the patients’ average age was 55.29 ± 13.7; the female/male ratio was approximately 6.5/1; patients with a history of radiotherapy to the neck area and chronic thyroid disease had a very high risk of developing cancer; homogeneous hypoechoic nodules had a higher risk of malignancy than the heterogeneous; the sign of height ≥ width had the diagnostic value of thyroid malignant nodule with Se = 57.8%, Sp = 89.6%, PPV = 88%, NPV = 89.6%, Acc = 41.4%, OR = 10.03; the sign of border irregularity had the diagnostic value of thyroid malignant nodule with Se = 77.8%, Sp = 80%, PPV = 87.5%, NPV = 66.7%, Acc = 57.1%, OR = 14; microcalcification has a diagnostic value of thyroid malignant nodule with Se = 71.1%, Sp = 56%, PPV = 17.4%, NPV = 51.9%, Acc = 57.1%, OR = 3.1. TIRADS 4B, 5 actually had a higher risk of malignancy than TIRADS 4A. The study concludes that ultrasound has a great value in orienting malignant lesions of hypoechonic thyroid nodules.&#x0D; Keywords: Thyroid nodule, thyroid hypoechoic nodule, TIRADS.&#x0D; References&#x0D; [1] F. Bray, J. Ferrlay, I. Soerjomataram et al., Global Cancer statistics 2018: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries, CA: A Cancer Journal for Clinicians, Vol. 68, No. 6, 2018, pp. 394-424. https://doi.org/full/10.3322/caac.21492.[2] E. Papini, R. Guglielmi, A. Bianchini, A. Crescenzi et al., Risk of Malignancy in Nonpalpable Thyroid Nodules: Predictive Value of Ultrasound and Color-Doppler Features, The Journal of Clinical Endocrinology &amp; Metabolism, Vol. 87, No. 5, 2002, pp. 1941-1946, https://doi.org/10.1210/jcem.87.5.8504.[3] T. T. Hong, Image Characteristics and Value of Ultrasound in the Diagnosis of Localized Thyroid Lesions, Master of Medicine Thesis Majoring in Diagnostic Imaging of Hanoi Medical University, 2013 (in Vietnamese).[4] V. T. Giao, Research on Ultrasound Values in TIRADS Classification of Localized Thyroid Lesions, Master of Medicine Thesis Majoring in Diagnostic Imaging, Hanoi Medical University, 2015 (in Vietnamese).[5] B. Y. Cho, H. S. Choi, Y. J. Park, J. A. Lim, H. Y. Ahn et al., Changes in the Clinicopathological Characteristics and Outcomes of Thyroid Cancer in Korea Over the Past Four Decades, Thyroid, Vol. 23, No. 7, 2013, pp. 797-804, https://doi.org/10.1089/thy.2012.0329.[6] C. M. Oh, S. Park, J. Y. Lee et al., Increased Prevalence of Chronic Lymphocytic Thyroiditis in Korean Patients With Papillary Thyroid Cancer, PLoS One, Vol. 9, No. 6, 2014, pp. 1-7, https://doi.org/10.1371/journal.pone.0099054.[7] G. Grani, L. Lamartina, V. Ramundo, R. Falcone, et al., Taller-Than-Wide Shape: A New Definition Improves the Specificity of TIRADS Systems, Eur Thyroid J, Vol. 9, No. 2, 2020, pp. 85-91, https://doi.org/10.1159/000504219.[8] B. Moifo, E. O. Takoeta, J. Tambe et al., Reliability of Thyroid Imaging Reporting and Data System (TIRADS) Classification in Differentiating Benign from Malignant Thyroid Nodules, Open Journal of Radiology, Vol. 3 No. 3, 2013, pp. 103-107, http://dx.doi.org/10.4236/ojrad.2013.33016.[9] H. J. Moon, J. Y. Kwak, M. J. Kim et al, Can Vascularity at Power Doppler US Help Predict Thyroid Malignancy?, Radiology, Vol. 255, 2010, pp. 260-269, https://doi.org/10.1148/radiol.09091284.[10] E. Horvath, S. Majlis, R. Rossi et al., An Ultrasonogram Reporting System for Thyroid Nodules Stratifying Cancer Risk for Clinical Management, J Clin Endocrinol Metab, Vol. 94 No. 5, 2009, pp. 1748-1751, https://doi.org/10.1210/jc.2008-1724.[11] G. Russ, B. Royer, C. Bigorgne et al., Prospective Evaluation of Thyroid Imaging Reporting and Data System on 4550 Nodules With and Without Elastography, Eur J Endocrinol, Vol. 168, No. 5, 2013, pp. 649-655, https://doi.org/10.1530/EJE-12-0936.[12] J. H. Shin, J. H. Baek, J. Chung et al., Ultrasonography Diagnosis and Imaging-Based Management of Thyroid Nodules: Revised Korean Society of Thyroid Radiology Consensus Statement and Recommendations, Korean J. Radiol., Vol. 17, No. 3, 2016, pp. 370-395, https://dx.doi.org/10.3348%2Fkjr.2016.17.3.370.[13] F. N. Tessler, W. D. Middlenton, E. G. Grant et al., ACR Thyroid Imaging, Reporting and Data System (TI-RADS): White Paper of the ACR TI-RADS Committee, J Am Coll Radiol., Vol. 14, No. 5, 2017, pp. 587-595, https://doi.org/10.1016/j.jacr.2017.01.046.[14] N. T. T. Thao, N. L. Lam, V. D. Luu, Research Into the Value in the Diagnosis of Thyroid Cancer of the EU-TIRADS 2017 Classification, Vietnamese Journal of Radiology, No 33, 2019, pp. 12-18, https://radiology.com.vn/bao-cao-khoa-hoc/nghien-cuu-gia-tri-chan-doan-ung-thu-tuyen-giap-cua-phan-do-eu-tirads-2017-n125.html/, (accessed on: Deceber 4th, 2019) (in Vietnamese).

This study retrospectively assessed the ultrasound parameters in Bethesda category III thyroid nodules accompanied by a diagnostic Afirma genomic sequencing classifier (GSC) testing result and surgical/clinical follow-up, and investigated whether combining ultrasound and GSC classifications could improve risk stratification of the atypia of undetermined significance (AUS) nodules. Data including radiologic Thyroid Imaging Report and Data System (TIRADS) score, concurrent GSC test results, and surgical/clinical follow-up were collected from patients with thyroid nodules meeting the study criteria. The distribution of TIRADS scores was compared between GSC-suspicious versus GSC-benign cohorts, as well as between malignant and benign nodules. The diagnostic performance of the GSC was compared between TIRADS 5 and TIRADS < 5 cohorts. The study consisted of 322 AUS nodules. Compared to the GSC-benign cohort, the GSC-suspicious cohort had a greater proportion of TIRADS 5 nodules (34% vs. 24%, P = 0.045), a lower proportion of TIRADS 1-3 nodules (20% vs. 30%, P = 0.048), and a higher average TIRADS score (4.12 vs. 3.92, P = 0.03). Compared to the benign cohort (histologically and/or clinically benign), the malignant cohort (histologically malignant) showed a higher proportion of TIRADS 5 nodules (52% vs. 24%, P = 0.004). The mean TIRADS score was also significantly higher in the malignant cohort compared to the benign cohort (4.36 vs. 3.92, P = 0.015). GSC testing demonstrated a significantly higher positive predictive value in the TIRADS 5 AUS nodules than that of TIRADS< 5 nodules (39% vs. 17%, P = 0.016). A GSC-suspicious result combined with TIRADS 5 significantly improved the positive predictive value, thereby enhancing risk stratification in AUS thyroid nodules, while other diagnostic parameters remained unchanged.

Thyroid cancer is one of the most common cancers and is especially common in young patients. Therefore, effective recognition and treatment of thyroid cancer are essential for patient survival. To compare the effectiveness of standard guidelines for predicting thyroid malignancy. To do so, thyroid nodules were classified according to the categories of the American Thyroid Association (ATA) and Thyroid Imaging Reporting and Data System (TI-RADS) guidelines, and compared with fine-needle aspiration biopsy (FNAB) results. The study included 1741 thyroid nodules with a final diagnosis in 1121 consecutive patients. The FNAB was recommended for all patients according to ATA guidelines and subsequently performed. The nodules were reclassified according to TI-RADS guidelines. Comparing nodules classified according to ATA and TI-RADS in terms of ultrasonography (US) features with the Bethesda cytological diagnosis classification System for Reporting Thyroid Cytopathology, 37.6% of the nodules classified in the high-risk category according to the ATA classification were found to be malignant cytology, 10.4% suspicious for malignancy, 4% non-diagnostic, 9.6% indeterminant cytology, and 38.4% benign. According to the TI-RADS risk category, 50% of those with high suspicion were malignant, 13.3% suspicious for malignancy cytology and 36.7% were benign. For the TI-RADS guidelines, the best cutoff value for differentiating benign and malignant nodules was found to be 4.5 (area under the curve (AUC) = 0.962, 95% CI = 0.943-0.981, p < 0.001). For the ATA guidelines, the best cutoff value for separating benign and malignant nodules was 4.5 (AUC = 0.917, 95% CI = 0.875-0.959, p < 0.001). The diagnostic performances of the TI-RADS and ATA score systems were evaluated using highly suspicious nodules. The sensitivity and specificity of highly suspicious nodules, according to both TI-RADS and ATA guidelines, were both high. Sensitivity and specificity of ATA classification were 80% and 96.3%, respectively. Sensitivity and specificity of TI-RADS classification were 76% and 97.5%, respectively, but positive predictive value was low (63.3% compared to 55.5%). Both, the ATA and TI-RADS classifications can effectively predict malignancy risk of thyroid nodules and may thus decrease unnecessary FNAB.

To explore the value of color-coded virtual touch tissue imaging (CCV) using acoustic radiation force pulse technology (ARFI) in diagnosing malignant thyroid nodules. Images including 189 thyroid nodules were collected as training samples and a binary logistic regression analysis was used to calculate regression coefficients for Thyroid Imaging Reporting and Data System (TI-RADS) and CCV. An integrated prediction model (TI-RADS+CCV) was then developed based on the regression coefficients. Another testing dataset involving 40 thyroid nodules was used to validate and compare the diagnostic performance of TI-RADS, CCV, and the integrated predictive models using the receiver operating characteristic (ROC) curves. Both TI-RADS and CCV are independent predictors. The diagnostic performance advantage of CCV is insignificant compared to TI-RADS (P = 0.61). However, the diagnostic performance of the integrated prediction model is significantly higher than that of TI-RADS or CCV (all P < 0.05). Applying to the validation image dateset, the integrated predictive model yields an area under the curve (AUC) of 0.880. Developing a new predictive model that integrates the regression coefficients calculated from TI-RADS and CCV enables to achieve the superior performance of thyroid nodule diagnosis to that of using TI-RADS or CCV alone.

ObjectiveTo describe the prevalence of thyroid nodules (TNs), their ultrasonographic characteristics, and the cancer risk assessment using the Thyroid Imaging Reporting and Data System (TIRADS) in a natural population, while also exploring their association with multiple metabolic disorders. This study aims to provide insight into the disease burden of TNs in the coastal area of Southeast China.DesignA cross-sectional study.Setting and participantsA total of 6221 participants from the baseline survey of the Fuqing Cohort Study, an ongoing population-based study in a coastal city of Southeast China, were included.Primary and secondary outcome measuresTNs and its detailed ultrasonographic characteristics, risk grading in TIRADS, and metabolic factors.ResultsThe age- and sex-standardized prevalence of TNs was found to be 67.2%, with a higher prevalence observed in females and an increase with age. Additionally, multiple, solid, and < 10 mm TNs were common. Approximately 39.2% of participants were diagnosed with grade 3, while 2.3% were classified as grade ≥ 4a according to TIRADS. Metabolic syndrome was associated with TNs, but this association was significant only in females. The relationship between metabolic disorders and the characteristics and grade of TNs in TIRADS varied by sex.ConclusionThe prevalence of TNs in the coastal area of Southeast China is notable, with 2.3% of the population classified as grade ≥ 4a in TIRADS, indicating a high risk of cancer and necessitating further assessment for thyroid cancer. The association between TNs and metabolic disorders varies by sex and requires further confirmation.

Assessment of thyroid nodules is usually relied on the experience of the radiologist and is time-consuming. Classification model of thyroid nodules cannot only reduce the burden on physicians but also provide objective recommendations. However, most classification models based on deep learning simply give a prediction result of the benignity or malignancy of nodules; thus, physicians have no way of knowing how the deep learning gets the prediction result due to the black-box nature of neural networks. In this work, we integrate the explainability directly into the outputs generated by the model through combining thyroid imaging reporting and data system (TI-RADS). The inference process of the proposed method is consistent with doctor's clinical diagnosis process; therefore, doctors can better explain the diagnosis results of the model to thepatient. A multitask network based on TI-RADS (MTN-TI-RADS) for the classification of thyroid nodules is proposed. In this network, a set of TI-RADS classifications of nodules is first obtained by multitask learning, then the TI-RADS points and the corresponding risk levels are calculated, and finally, nodules are classified as benign and malignant. The classification process through the network is consistent with the diagnostic process of physician; thus, the results of classification can be easily understood by physicians. In addition, the attention modules are introduced to the spatial and channel domains to let the network focus more on criticalfeatures. To verify the classification performance of our method, we compared the results obtained through our method with the results of the radiologist's evaluation. For the 781 test nodules in the internal dataset and the 886 test nodules in the external dataset, the sensitivity and specificity of MTN-TI-RADS were 0.988, 0.912 in internal dataset, 0.949, 0.930 in external dataset, versus the senior radiologist of 0.925 ( ), 0.816 ( ), and 0.910 ( ), 0.836 ( ), respectively. And the area under the receiver operating characteristic curve of MTN-TI-RADS was 0.981 in internal dataset, 0.973 in external dataset, versus the senior radiologist of 0.905, 0.923. For the internal dataset, we also computed the accuracy of the risk level (TR1 to TR5) and the mean absolute error (MAE). The accuracy of the risk level of the proposed method is 78%, and the MAE is 1.30. The MAE of the total points (0-14 points) is 1.30. An effective and result-interpretable end-to-end thyroid nodule classification network (MTN-TI-RADS) is proposed. MTN-TI-RADS has superior ability to classify malignant and benign thyroid nodules compared to senior radiologists. Based on MTN-TI-RADS, a classification model with strong interpretation and a high degree of physician trust is constructed. The proposed classification network is consistent with the diagnosis process of physicians, thus is more reliable and interpretable, and has great potential for clinicalapplication.

Introduction: Thyroid ultrasound (US) should be performed in the initial assessment of the gland. The fine-needle aspiration cytology (FNAC) is an inexpensive and useful tool for detecting thyroid cancer. In the management of thyroid nodule patients, recommending who should be submitted to FNAC is still controversial. Aim: The aim of the study was to evaluate whether the US-based Thyroid Imaging Reporting and Data System (TI-RADS) scoring system is reliable in differentiating between benign and malignant thyroid nodules when compared to FNAC and its practical application in clinical decision-making. Materials and Methods: A total of 59 patients having solitary thyroid nodules were studied over 20 months. Thyroid function tests, US and FNAC of the thyroid were done. Using US, the thyroid nodule was evaluated for five sonological features: internal composition, echogenicity, margins, presence and type of calcification and shape of the lesion. The nature of the lesion was assessed based on the TI-RADS score calculated from the five US parameters. This result, along with FNAC, which was reported using The Bethesda System for Reporting Thyroid Cytopathology was finally compared with the histopathology reports. Results: When the suspicious sonological features were evaluated, the presence of solid components and taller-than-wide shapes was found to be maximum sensitive but less specific. Specificity, positive predictive value and negative predictive value were found to be maximum for the presence of microcalcification and presence of hypoechogenicity. The sensitivity of TI-RADS to detect malignancy was higher (100%) than FNAC (66.7%). The specificity of TI-RADS to detect malignancy was lower (36.2%) as compared to FNAC. Conclusion: TI-RADS can be used to avoid unnecessary fine-needle aspirations. It helps in decreasing confusion and assists in deciding when FNAC should be performed. TI-RADS is cost-effective and has a vital role in the initial evaluation and investigation of patients presenting with solitary nodules of the thyroid. Moreover, it is simple to use, reliable and easily reproducible.

To compare the clinical value of superb microvascular imaging (SMI) and contrast enhanced ultrasound (CEUS) in the differential diagnosis of thyroid nodules, and to further study whether the combination of SMI and/or CEUS with thyroid imaging reporting and data system (TI-RADS) can improve the diagnostic value of TI-RADS. Methods: SMI, CEUS, TI-RADS, TI-RADS combined with CEUS, TI-RADS combined with SMI, TI-RADS combined with SMI and CEUS were used to differentiate thyroid nodules. TI-RADS 4b, 5 categories in two-dimensional ultrasound, hypoenhancement in CEUS, and peripheral blood flow with penetrating vessels in SMI were considered malignant signs. The diagnostic efficacy of these methods was compared according to post-operative pathology or fine needle aspiration cytology. Results: A total of 237 patients with 296 thyroid nodules were included in the study. The sensitivity, specificity and accuracy for TI-RADS were 78.4%, 77.7% and 78.0% respectively, 75.5%, 86.6% and 81.4% in SMI, 82.0%, 88.5% and 85.5% in CEUS, 92.1%, 90.4% and 91.2% in TI-RADS combined with SMI and CEUS. Area under the receiver operating characteristic curve (AUROC) showed no significant difference between SMI, CEUS, TI-RADS (χ2=4.29, P=0.117). AUROC of both TI-RADS combined with CEUS (χ2=39.62, P<0.001), TI-RADS combined with SMI (χ2=36.61, P<0.001) were higher than TI-RADS. AUROC of TI-RADS combined with SMI and CEUS was higher than TI-RADS combined with SMI (χ2=4.50, P=0.033) or TI-RADS combined with CEUS (χ2=4.24, P=0.039). Conclusion: Peripheral blood flow with penetrating vessels in SMI possesses high diagnostic value in differentiating thyroid cancer. The combination of SMI and/or CEUS with TI-RADS can be used to improve the diagnostic efficacy of TI-RADS in differential diagnosis of thyroid nodules.

To compare malignancy risk stratification of thyroid nodules with the 2014 American Thyroid Association (ATA) management guidelines and the Thyroid Imaging Reporting and Data System (TIRADS). This retrospective study was approved by the institutional review board. The requirement to obtain informed consent was waived. From November 2013 to July 2014, 1293 thyroid nodules in 1241 patients (mean age, 50.8 years ± 13.5) were included in this study. All nodules measured at least 10 mm. Solidity, hypoechogenicity or marked hypoechogenicity, microlobulated to irregular margin, microcalcifications or mixed calcifications, and nonparallel shape were considered suspicious features at ultrasonography (US). A TIRADS category and the US pattern as determined with ATA guidelines were assigned to each nodule. The correlation between the TIRADS category or ATA pattern and the malignancy rate was evaluated with the Spearman rank test. Of the 1293 thyroid nodules, 1059 (81.9%) were benign and 234 (18.1%) were malignant. Forty-four of the 1293 nodules (3.4%) did not meet the criteria for the ATA patterns and were classified as "not specified." The malignancy rates of TIRADS category 3, 4a, 4b, 4c, and 5 nodules were 1.9% (six of 316 nodules), 4.2% (17 of 408 nodules), 12.9% (33 of 256 nodules), 49.8% (130 of 261 nodules), and 92.3% (48 of 52 nodules), respectively, with significant differences between categories (P < .001). Malignancy rates of nodules with very low, low, intermediate, and high suspicion for malignancy with the ATA guidelines and not specified patterns were 2.7% (11 of 407 nodules), 3.1% (10 of 323 nodules), 16.7% (39 of 233 nodules), 58.0% (166 of 286 nodules), and 18.2% (eight of 44 nodules), respectively, with significant differences between patterns (P < .001). There was high correlation between classification with TIRADS (r = 1.000, P < .001) and ATA guidelines (r = 0.900, P = 0.037), without statistically significant differences (P = .873). Both TIRADS and the ATA guidelines provide effective malignancy risk stratification for thyroid nodules. Nodules that do not meet the criteria for a specific pattern with the ATA guidelines have a relatively high risk of malignancy (18.2%).