AI diagnosis of Bethesda category IV thyroid nodules

Beyond genomics: artificial intelligence-powered diagnostics for indeterminate thyroid nodules-a systematic review and meta-analysis

Introduction

In recent years, artificial intelligence (AI) tools have become widely studied for thyroid ultrasonography (USG) classification. The real-world applicability of these developed tools as pre-operative diagnostic aids is limited due to model overfitting, clinician trust, and a lack of gold standard surgical histology as ground truth class label. The ongoing dilemma within clinical thyroidology is surgical decision making for indeterminate thyroid nodules (ITN). Genomic sequencing classifiers (GSC) have been utilised for this purpose; however, costs and availability preclude universal adoption creating an inequity gap. We conducted this review to analyse the current evidence of AI in ITN diagnosis without the use of GSC.

Methods

English language articles evaluating the diagnostic accuracy of AI for ITNs were identified. A systematic search of PubMed, Google Scholar, and Scopus from inception to 18 February 2025 was performed using comprehensive search strategies incorporating MeSH headings and keywords relating to AI, indeterminate thyroid nodules, and pre-operative diagnosis. This systematic review and meta-analysis was conducted in accordance with methods recommended by the Cochrane Collaboration (PROSPERO ID CRD42023438011).

Results

The search strategy yielded 134 records after the removal of duplicates. A total of 20 models were presented in the seven studies included, five of which were radiological driven, one utilised natural language processing, and one focused on cytology. The pooled meta-analysis incorporated 16 area under the curve (AUC) results derived from 15 models across three studies yielding a combined estimate of 0.82 (95% CI: 0.81-0.84) indicating moderate-to-good classification performance across machine learning (ML) and deep learning (DL) architectures. However, substantial heterogeneity was observed, particularly among DL models (I² = 99.7%, pooled AUC = 0.85, 95% CI: 0.85-0.86). Minimal heterogeneity was observed among ML models (I² = 0.7%), with a pooled AUC of 0.75 (95% CI: 0.70-0.81). Meta-regression analysis performed suggests potential publication bias or systematic differences in model architectures, dataset composition, and validation methodologies.

Conclusion

This review demonstrated the burgeoning potential of AI to be of clinical value in surgical decision making for ITNs; however, study-developed models were unsuitable for clinical implementation based on performance alone at their current states or lacked robust independent external validation. There is substantial capacity for further development in this field.

Systematic Review Registration

https://www.crd.york.ac.uk/PROSPERO/, identifier CRD42023438011.

Multimodal GPT model for assisting thyroid nodule diagnosis and management

Although using artificial intelligence (AI) to analyze ultrasound images is a promising approach to assessing thyroid nodule risks, traditional AI models lack transparency and interpretability. We developed a multimodal generative pre-trained transformer for thyroid nodules (ThyGPT), aiming to provide a transparent and interpretable AI copilot model for thyroid nodule risk assessment and management. Ultrasound data from 59,406 patients across nine hospitals were retrospectively collected to train and test the model. After training, ThyGPT was found to assist in reducing biopsy rates by more than 40% without increasing missed diagnoses. In addition, it detects errors in ultrasound reports 1,610 times faster than humans. With the assistance of ThyGPT, the area under the curve for radiologists in assessing thyroid nodule risks improved from 0.805 to 0.908 (p < 0.001). As an AI-generated content-enhanced computer-aided diagnosis (AIGC-CAD) model, ThyGPT has the potential to revolutionize how radiologists use such tools.

A model based on Chinese thyroid imaging reporting and data systems for predicting Bethesda III/IV thyroid nodules

Objectives

This study aimed to explore the performance of a model based on Chinese Thyroid Imaging Reporting and Data Systems (C-TIRADS), clinical characteristics, and other ultrasound characteristics for the prediction of Bethesda III/IV thyroid nodules before fine needle aspiration (FNA).

Materials and methods

A total of 855 thyroid nodules from 810 patients were included. All nodules underwent ultrasound examination before FNA. All nodules were categorized according to the C-TIRADS criteria and classified into two groups, Bethesda III/IV and non-III/IV thyroid nodules, using cytologic diagnosis as the gold standard. The clinical and ultrasonographic characteristics of the nodules in the two groups were compared, and independent predictors of Bethesda III/IV nodules were determined by univariate and multivariate logistic regression analyses, based on which a prediction model was constructed. The predictive efficacy of the model was compared with that of C-TIRADS alone by sensitivity, specificity, and area under the curve (AUC).

Results

Our study found that the C-TIRADS category, homogeneous echotexture, blood flow signal present, and posterior echo unchanged were independent predictors for Bethesda III/IV thyroid nodules. Based on multiple logistic regression, a predictive model was established: Logit (p)= - 4.213 + 0.965 × homogeneous echotexture+ 1.050 × blood flow signal present + 0.473 × posterior echo unchanged+ 2.859 × C-TIRADS 3 + 2.804 × C-TIRADS 4A + 1.824 × C-TIRADS 4B + 0.919 × C-TIRADS 4C. The AUC of the model among all nodules was 0.746 (95%CI: 0.710-0.782), 0.779 (95%CI: 0.730-0.829) among nodules with a diameter (D) > 10mm, and 0.718 (95%CI: 0.667-0.769) among nodules with D ≤ 10mm, which were significantly higher than that of the C-TIRADS alone.

Conclusion

We developed a predictive model for Bethesda III/IV thyroid nodules that is better for nodules with D > 10mm FNA operators can choose the optimal puncture strategy based on the prediction results to improve the rate of definitive diagnosis of the first FNA of Bethesda III/IV nodules and thus reduce repeat FNA.

Global research trends in the application of artificial intelligence in oncology care: a bibliometric study

Objective

To use bibliometric methods to analyze the prospects and development trends of artificial intelligence(AI) in oncology nursing from 1994 to 2024, providing guidance and reference for oncology nursing professionals and researchers.

Methods

The core set of the Web of Science database was searched for articles from 1994 to 2024. The R package "Bibliometrix" was used to analyze the main bibliometric features, creating a three-domain chart to display relationships among institutions, countries, and keywords. VOSviewer facilitated co-authorship analysis and its visualization was used for co- occurrence analysis. CiteSpace calculated citation bursts and keyword occurrences.

Results

A total of 517 articles were retrieved, representing 80 countries/regions. The United States had the highest number of publications, with 188 articles (36.4%), followed by China with 79 articles (15.3%). The top 10 institutions in terms of publication output were all U.S.-based universities or cancer research institutes, with Harvard University ranking first. Prominent research teams, such as those led by Repici, Aerts, and Almangush, have made significant contributions to studies on AI in tumor risk factor identification and symptom management. In recent years, the keywords with the highest burst strength were "model" and "human papillomavirus." The most studied tumor type was breast cancer. While Cancers published the highest number of articles, journals such as CA: A Cancer Journal for Clinicians and PLOS ONE had higher impact and citation rates.

Conclusion

By analyzing the volume of AI literature in oncology nursing, combined with the statistical analysis of institutions, core authors, journals, and keywords, the research hotspots and trends in the application of AI in oncology nursing over the past 30 years are revealed. AI in oncology nursing is entering a stage of rapid development, providing valuable reference for scholars and professionals in the field.

Deep learning model for diagnosis of thyroid nodules with size less than 1 cm: A multicenter, retrospective study

Objective

To develop a ultrasound images based dual-channel deep learning model to achieve accurate early diagnosis of thyroid nodules less than 1 cm.

Methods

A dual-channel deep learning model called thyroid nodule transformer network (TNT-Net) was proposed. The model has two input channels for transverse and longitudinal ultrasound images of thyroid nodules, respectively. A total of 9649 nodules from 8455 patients across five hospitals were retrospectively collected. The data were divided into a training set (8453 nodules, 7369 patients), an internal test set (565 nodules, 512 patients), and an external test set (631 nodules, 574 patients).

Results

TNT-Net achieved an area under the curve (AUC) of 0.953 (95 % confidence interval (CI): 0.934, 0.969) on the internal test set and 0.941 (95 % CI: 0.921, 0.957) on the external test set, significantly outperforming traditional deep convolutional neural network models and single-channel swin transformer model, whose AUCs ranged from 0.800 (95 % CI: 0.759, 0.837) to 0.856 (95 % CI: 0.819, 0.881). Furthermore, feature heatmap visualization showed that TNT-Net could extract richer and more energetic malignant nodule patterns.

Conclusion

The proposed TNT-Net model significantly improved the recognition capability for thyroid nodules with size less than 1 cm. This model has the potential to reduce overdiagnosis and overtreatment of such nodules, providing essential support for precise management of thyroid nodules while complementing fine-needle aspiration biopsy.

Optimizing thyroid AUS nodules malignancy prediction: a comprehensive study of logistic regression and machine learning models

Background

The accurate diagnosis of thyroid nodules with indeterminate cytology, particularly in the atypia of undetermined significance (AUS) category, remains challenging. This study aims to predict the risk of malignancy in AUS nodules by comparing two machine learning (ML) and three conventional logistic regression (LR) models.

Methods

A retrospective study on 356 AUS nodules in 342 individuals from 6728 patients who underwent thyroid surgery in 2021. All the clinical, ultrasonographic, and molecular data were collected and randomly separated into training and validation cohorts at a ratio of 7: 3. ML (random forest and XGBoost) and LR (lasso regression, best subset selection, and backward stepwise regression) models were constructed and evaluated using area under the curve (AUC), calibration, and clinical utility metrics.

Results

Approximately 90% (321/356) of the AUS nodules were malignant, predominantly papillary thyroid carcinoma with 68.6% BRAF V600E mutations. The final LR prediction model based on backward stepwise regression exhibited superior discrimination with AUC values of 0.83 (95% CI: 0.73-0.92) and 0.80 (95% CI: 0.67-0.94) in training and validation, respectively. Well calibration, and clinical utility were also confirmed. The ML models showed moderate performance. A nomogram was developed on the final LR model.

Conclusions

The LR model developed using the backward stepwise regression, outperformed ML models in predicting malignancy in AUS thyroid nodules. The corresponding nomogram based on this model provides a valuable and practical tool for personalized risk assessment, potentially reducing unnecessary surgeries and enhancing clinical decision-making.

A model based on C-TIRADS combined with SWE for predicting Bethesda I thyroid nodules

Objectives

This study aimed to explore the performance of a model based on Chinese Thyroid Imaging Reporting and Data Systems (C-TIRADS), clinical characteristics, and shear wave elastography (SWE) for the prediction of Bethesda I thyroid nodules before fine needle aspiration (FNA).

Materials and methods

A total of 267 thyroid nodules from 267 patients were enrolled. Ultrasound and SWE were performed for all nodules before FNA. The nodules were scored according to the 2020 C-TIRADS, and the ultrasound and SWE characteristics of Bethesda I and non-I thyroid nodules were compared. The independent predictors were determined by univariate analysis and multivariate logistic regression analysis. A predictive model was established based on independent predictors, and the sensitivity, specificity, and area under the curve (AUC) of the independent predictors were compared with that of the model.

Results

Our study found that the maximum diameter of nodules that ranged from 15 to 20 mm, the C-TIRADS category <4C, and E max <52.5 kPa were independent predictors for Bethesda I thyroid nodules. Based on multiple logistic regression, a predictive model was established: Logit (p) = -3.491 + 1.630 × maximum diameter + 1.719 × C-TIRADS category + 1.046 × E max (kPa). The AUC of the model was 0.769 (95% CI: 0.700-0.838), which was significantly higher than that of the independent predictors alone.

Conclusion

We developed a predictive model for predicting Bethesda I thyroid nodules. It might be beneficial to the clinical optimization of FNA strategy in advance and to improve the accurate diagnostic rate of the first FNA, reducing repeated FNA.

Utilizing machine learning for early screening of thyroid nodules: a dual-center cross-sectional study in China

Background

Thyroid nodules, increasingly prevalent globally, pose a risk of malignant transformation. Early screening is crucial for management, yet current models focus mainly on ultrasound features. This study explores machine learning for screening using demographic and biochemical indicators.

Methods

Analyzing data from 6,102 individuals and 61 variables, we identified 17 key variables to construct models using six machine learning classifiers: Logistic Regression, SVM, Multilayer Perceptron, Random Forest, XGBoost, and LightGBM. Performance was evaluated by accuracy, precision, recall, F1 score, specificity, kappa statistic, and AUC, with internal and external validations assessing generalizability. Shapley values determined feature importance, and Decision Curve Analysis evaluated clinical benefits.

Results

Random Forest showed the highest internal validation accuracy (78.3%) and AUC (89.1%). LightGBM demonstrated robust external validation performance. Key factors included age, gender, and urinary iodine levels, with significant clinical benefits at various thresholds. Clinical benefits were observed across various risk thresholds, particularly in ensemble models.

Conclusion

Machine learning, particularly ensemble methods, accurately predicts thyroid nodule presence using demographic and biochemical data. This cost-effective strategy offers valuable insights for thyroid health management, aiding in early detection and potentially improving clinical outcomes. These findings enhance our understanding of the key predictors of thyroid nodules and underscore the potential of machine learning in public health applications for early disease screening and prevention.

Establishment and validation of a multivariate logistic model for risk factors of thyroid nodules using lasso regression screening

Objective

This study aims to analyze the association between the occurrence of thyroid nodules and various factors and to establish a risk factor model for thyroid nodules.

Methods

The study population was divided into two groups: a group with thyroid nodules and a group without thyroid nodules. Regression with the least absolute shrinkage and selection operator (Lasso) was applied to the complete dataset for variable selection. Binary logistic regression was used to analyze the relationship between various influencing factors and the prevalence of thyroid nodules.

Results

Based on the screening results of Lasso regression and the subsequent establishment of the Binary Logistic Regression Model on the training dataset, it was found that advanced age (OR=1.046, 95% CI: 1.033-1.060), females (OR = 1.709, 95% CI: 1.342-2.181), overweight individuals (OR = 1.546, 95% CI: 1.165-2.058), individuals with impaired fasting glucose (OR = 1.590, 95% CI: 1.193-2.122), and those with dyslipidemia (OR = 1.588, 95% CI: 1.197-2.112) were potential risk factors for thyroid nodule disease (p<0.05). The area under the curve (AUC) of the receiver operating characteristic (ROC) curve for the Binary Logistic Regression Model is 0.68 (95% CI: 0.64-0.72).

Conclusions

advanced age, females, overweight individuals, those with impaired fasting glucose, and individuals with dyslipidemia are potential risk factors for thyroid nodule disease.