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

Nationwide trends and the impact of an oncology hospital network on reducing the burden of thyroid cytology procedures

The diagnostic care pathway of thyroid nodules spans multiple institutions. Collaborative networks are important to deal with such pathways that result from centralization and differentiation of care. Despite the high prevalence of thyroid nodules and the increase in cancer diagnoses, most nodules are benign and attributable to overdiagnosis, leading to an increase in fine-needle aspirations (FNAs). This study assessed the effectiveness of a multi-hospital network that implemented a unified thyroid care pathway in reducing the number of FNAs without compromising malignancy detection. In this nationwide population-based cohort study, Bethesda scores were extracted from all thyroid FNA reports from 2010 to 2021 in the Netherlands using text mining. Trends in the number of FNAs and Bethesda scores were visualized for the network and the rest of the country. Joinpoint analyses with the Davies test determined the statistical significance of observed trend changes. Nationwide, FNAs increased by an average of 5.7% annually from 2010 to 2018. In the network, FNAs started to decrease in 2016-2017, coinciding with the care pathway implementation (p < 0.001). In contrast, in the rest of the Netherlands, a decline was observed in 2020, potentially attributable to the COVID-19 pandemic. In both cases, the reduction mainly involved Bethesda categories 1 and 2, without compromising malignancy detection. High-volume surgical centers seemed to initiate a decline more rapidly compared to non-high-volume surgical centers. This study indicates that a unified care pathway within a multi-hospital network can reduce the number of FNAs without compromising malignancy detection, which could alleviate the burden on both patients and the healthcare system.

Adaptive Dual-Task Deep Learning for Automated Thyroid Cancer Triaging at Screening US

Purpose To develop an adaptive dual-task deep learning model (ThyNet-S) for detection and classification of thyroid lesions at US screening. Materials and Methods This retrospective study used a multicenter dataset comprising 35 008 thyroid US images of 23 294 individual examinations (mean age, 40.4 years ± 13.1 [SD]; 17 587 female) from seven medical centers from January 2009 to December 2021. Of these, 29 004 images were used for model development and 6004 images for validation. The model determined cancer risk for each image and automatically triaged images with normal thyroid and benign nodules by dynamically integrating lesion detection through pixel-level feature analysis and lesion classification through deep semantic features analysis. Diagnostic performance of screening assisted by the model (ThyNet-S triaged screening) and traditional screening (radiologists alone) was assessed by comparing sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve using the McNemar test and DeLong test. The influence of ThyNet-S on radiologist workload and clinical decision-making was also assessed. Results ThyNet-S-assisted triaged screening achieved a higher area under the receiver operating characteristic curve than original screening with six senior and six junior radiologists (0.93 vs 0.91 and 0.92 vs 0.88, respectively; all P < .001). The model improved sensitivity for junior radiologists (88.2% vs 86.8%; P < .001). Notably, the model reduced radiologists' workload by triaging 60.4% of cases as not potentially malignant, which did not require further interpretation. The model simultaneously decreased the unnecessary fine needle aspiration rate from 38.7% to 14.9% and 11.5% when used independently or in combination with the Thyroid Imaging Reporting and Data System, respectively. Conclusion ThyNet-S improved the efficiency of thyroid cancer screening and optimized clinical decision-making. Keywords: Artificial Intelligence, Adaptive, Dual Task, Thyroid Cancer, Screening, Ultrasound Supplemental material is available for this article. © RSNA, 2025.

Multimodality Ultrasound Utilizing Microvascular Flow Imaging and Shear Wave Elastography to Guide Fine-Needle Aspiration of Thyroid Lesions: A Prospective Study Validating Pattern-Based Microvascular Classification

Background: Most current guidelines recommend fine-needle aspiration (FNA) biopsy of thyroid nodules based on grayscale ultrasound (GUS) features, but the biopsy rate for benign nodules remains high. Our aim was to construct a new pattern-based microvascular classification (PBMC) for thyroid nodules to develop and validate predictive multimodality US models based on GUS, microvascular flow imaging, and shear wave elastography, and compare FNA decision accuracy with the American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS). Methods: This prospective study included consecutive patients with thyroid nodules who underwent multimodality US examinations from September 2022 to December 2023. Using PBMC, lesions were divided into three categories: malignant signs (convergence sign, piercing sign, and spoke wheel sign), benign signs (ring sign), and other vascular patterns. Univariate and multivariable logistic regression analyses were conducted to determine the odds ratios (ORs) of US features, including vascular signs, and construct predictive models based on multimodality US. Multimodality US models were validated with internal cross-validation and evaluated based on discrimination, calibration, and decision curve analyses. Results: Overall, 793 thyroid nodules confirmed using pathological analysis (248 benign and 545 malignant) in 599 participants (mean age, 43 years ±11 [SD]) were included. In univariate logistic regression analyses, malignant vascular signs showed a positive association with malignant nodules (OR: 10.43, 95% confidence interval [CI]: 5.76, 18.88; p < 0.01), whereas benign vascular signs were inversely associated with malignancy (OR: 0.10, 95% CI: 0.06, 0.16; p < 0.01). Four multivariable models incorporated GUS features, Young's modulus, and PBMC. The highest area under the receiver operating characteristic curve (AUC) was 0.95 (95% CI: 0.82, 0.97) for the multimodality US model, and the lowest AUC was 0.62 (95% CI: 0.57, 0.66) for ACR TI-RADS based on GUS (p < 0.001). At a 71% risk threshold, multimodality US avoided 27% (95% CI: 21, 34) of FNA procedures, compared with 13% (95% CI: 0, 38) with TI-RADS (p < 0.001). Conclusion: Visual assessment of microvascular morphology patterns may improve differentiation of benign and malignant thyroid nodules and potentially reduce the risk of unnecessary biopsy of benign thyroid nodules.

A comparative analysis of three graph neural network models for predicting axillary lymph node metastasis in early-stage breast cancer

The presence of axillary lymph node metastasis (ALNM) in breast cancer patients is an important factor in deciding whether to have axillary surgery or pursue alternative treatments. Based on axillary ultrasound (US) and histopathologic data, three graph neural network models were compared to predict ALNM in early-stage breast cancer. The patients were randomly divided into two data sets: training (80%) and testing (20%). Predictive performance was measured using accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, and area under the curve (AUC). In the test cohort, the graph convolutional network (GCN) performed the best in predicting ALNM, with an AUC of 0.77 (95% confidence interval [CI]: 0.69-0.84). In conclusion, the GCN model has the potential to provide a noninvasive tool for detecting ALNM and can aid in clinical decision-making. Prospective studies are expected to provide high-level evidence for clinical usage in future investigations.

Thyroid nodule classification in ultrasound imaging using deep transfer learning

BACKGROUND

The accurate diagnosis of thyroid nodules represents a critical and frequently encountered challenge in clinical practice, necessitating enhanced precision in diagnostic methodologies. In this study, we investigate the predictive efficacy of distinguishing between benign and malignant thyroid nodules by employing traditional machine learning algorithms and a deep transfer learning model, aiming to advance the diagnostic paradigm in this field.

METHODS

In this retrospective study, ITK-Snap software was utilized for image preprocessing and feature extraction from thyroid nodules. Feature screening and dimensionality reduction were conducted using the least absolute shrinkage and selection operator (LASSO) regression method. To identify the optimal model, both traditional machine learning and transfer learning approaches were employed, followed by model fusion using post-fusion techniques. The performance of the model was rigorously evaluated through the area under the curve (AUC), calibration curve analysis, and decision curve analysis (DCA).

RESULTS

A total of 1134 images from 630 cases of thyroid nodules were included in this study, comprising 589 benign nodules and 545 malignant nodules. Through comparative analysis, the support vector machine (SVM), which demonstrated the best diagnostic performance among traditional machine learning models, and the Inception V3 convolutional neural network model, based on transfer learning, were selected for model construction. The SVM model achieved an AUC of 0.748 (95% CI: 0.684-0.811) for diagnosing malignant thyroid nodules, while the Inception V3 transfer learning model yielded an AUC of 0.763 (95% CI: 0.702-0.825). Following model fusion, the AUC improved to 0.783 (95% CI: 0.724-0.841). The difference in performance between the fusion model and the traditional machine learning model was statistically significant (p = 0.036). Decision curve analysis (DCA) further confirmed that the fusion model exhibits superior clinical utility, highlighting its potential for practical application in thyroid nodule diagnosis.

CONCLUSION

Our findings demonstrate that the fusion model, which integrates a convolutional neural network (CNN) with traditional machine learning and deep transfer learning techniques, can effectively differentiate between benign and malignant thyroid nodules through the analysis of ultrasound images. This model fusion approach significantly optimizes and enhances diagnostic performance, offering a robust and intelligent tool for the clinical detection of thyroid diseases.

Diagnostic value of deep learning of multimodal imaging of thyroid for TI-RADS category 3-5 classification

BACKGROUND

Thyroid nodules classified within the Thyroid Imaging Reporting and Data Systems (TI-RADS) category 3-5 are typically regarded as having varying degrees of malignancy risk, with the risk increasing from TI-RADS 3 to TI-RADS 5. While some of these nodules may undergo fine-needle aspiration (FNA) biopsy to assess their nature, this procedure carries a risk of false negatives and inherent complications. To avoid the need for unnecessary biopsy examination, we explored a method for distinguishing the benign and malignant characteristics of thyroid TI-RADS 3-5 nodules based on deep-learning ultrasound images combined with computed tomography (CT).

METHODS

Thyroid nodules, assessed as American College of Radiology (ACR) TI-RADS category 3-5 through conventional ultrasound, all of which had postoperative pathology results, were examined using both conventional ultrasound and CT before operation. We investigated the effectiveness of deep-learning models based on ultrasound alone, CT alone, and a combination of both imaging modalities using the following metrics: Area Under Curve (AUC), sensitivity, accuracy, and positive predictive value (PPV). Additionally, we compared the diagnostic efficacy of the combined methods with manual readings of ultrasound and CT.

RESULTS

A total of 768 thyroid nodules falling within TI-RADS categories 3-5 were identified across 768 patients. The dataset comprised 499 malignant and 269 benign cases. For the automatic identification of thyroid TI-RADS category 3-5 nodules, deep learning combined with ultrasound and CT demonstrated a significantly higher AUC (0.930; 95% CI: 0.892, 0.969) compared to the application of ultrasound alone AUC (0.901; 95% CI: 0.856, 0.947) or CT alone AUC (0.776; 95% CI: 0.713, 0.840). Additionally, the AUC of combined modalities surpassed that of radiologists'assessments using ultrasound alone AUCmean (0.725；95% CI:0.677, 0.773), CT alone AUCmean (0.617; 95% CI:0.564, 0.669). Deep learning method combined with ultrasound and CT imaging of thyroid can allow more accurate and precise classification of nodules within TI-RADS categories 3-5.

Model Based on Ultrasound Radiomics and Machine Learning to Preoperative Differentiation of Follicular Thyroid Neoplasm

OBJECTIVES

To evaluate the value of radiomics based on ultrasonography in differentiating follicular thyroid carcinoma (FTC) and follicular thyroid adenoma (FTA) and construct a tool for preoperative noninvasive predicting FTC and FTA.

METHODS

The clinical data and ultrasound images of 389 patients diagnosed with FTC or FTA postoperatively were retrospectively analyzed at 3 institutions from January 2017 to December 2023. Patients in our hospital were randomly assigned in a 7:3 ratio to training cohort and validation cohort. External test cohort consisted of data collected from other 2 hospitals. Radiomics features were used to develop models based on different machine learning classifiers. A combined model was developed combining radiomics features with clinical characteristics and a nomogram was depicted. The performance of the models was assessed by area under the receiver operating characteristic curve (AUC), calibration curve and decision curve.

RESULTS

Radiomics model based on random forest showed best performance in discriminating FTC and FTA, with AUCs 0.880 (95% confidence interval [CI]: 0.8290-0.9308), 0.871 (95% CI: 0.7690-0.9734), and 0.821 (95% CI: 0.7036-0.9389) in training, validation, and test cohort, respectively. The combined model presented better efficacy comparing with clinical model and radiomics model, with AUCs 0.883 (95% CI: 0.8359-0.9295), 0.874 (95% CI: 0.7873-0.9615), and 0.876 (0.7809-0.9714) in training, validation, and test cohort, respectively. The calibration curves suggested good consistency and decision curves showed the highest overall clinical benefit for the combined model.

CONCLUSIONS

Ultrasound radiomics model based on random forest is feasible to differentiate FTC and FTA, and the combined model is an intuitively noninvasive tool for FTC and FTA preoperative identification.

WFUMB Commentary Paper on Artificial intelligence in Medical Ultrasound Imaging

Artificial intelligence (AI) is defined as the theory and development of computer systems able to perform tasks normally associated with human intelligence. At present, AI has been widely used in a variety of ultrasound tasks, including in point-of-care ultrasound, echocardiography, and various diseases of different organs. However, the characteristics of ultrasound, compared to other imaging modalities, such as computed tomography (CT) and magnetic resonance imaging (MRI), poses significant additional challenges to AI. Application of AI can not only reduce variability during ultrasound image acquisition, but can standardize these interpretations and identify patterns that escape the human eye and brain. These advances have enabled greater innovations in ultrasound AI applications that can be applied to a variety of clinical settings and disease states. Therefore, The World Federation of Ultrasound in Medicine and Biology (WFUMB) is addressing the topic with a brief and practical overview of current and potential future AI applications in medical ultrasound, as well as discuss some current limitations and future challenges to AI implementation.

Ultrasound radiomics and genomics improve the diagnosis of cytologically indeterminate thyroid nodules

Background

Increasing numbers of cytologically indeterminate thyroid nodules (ITNs) present challenges for preoperative diagnosis, often leading to unnecessary diagnostic surgical procedures for nodules that prove benign. Research in ultrasound radiomics and genomic testing leverages high-throughput data and image or sequence algorithms to establish assisted models or testing panels for ITN diagnosis. Many radiomics models now demonstrate diagnostic accuracy above 80% and sensitivity over 90%, surpassing the performance of less experienced radiologists and, in some cases, matching the accuracy of experienced radiologists. Molecular testing panels have helped clinicians achieve accurate diagnoses of ITNs, preventing unnecessary diagnostic surgical procedures in 42%-61% of patients with benign nodules.

Objective

In this review, we examined studies on ultrasound radiomics and genomic molecular testing for cytological ITNs conducted over the past 5 years, aiming to provide insights for researchers focused on improving ITN diagnosis.

Conclusion

Radiomics models and molecular testing have enhanced diagnostic accuracy before surgery and reduced unnecessary diagnostic surgical procedures for ITN patients.

Biomedical data analytics for better patient outcomes

Medical professionals today have access to immense amounts of data, which enables them to make decisions that enhance patient care and treatment efficacy. This innovative strategy can improve global health care by bridging the divide between clinical practice and medical research. This paper reviews biomedical developments aimed at improving patient outcomes by addressing three main questions regarding techniques, data sources and challenges. The review includes peer-reviewed articles from 2018 to 2023, found via systematic searches in PubMed, Scopus and Google Scholar. The results show diverse disease-specific applications. Challenges such as data quality and ethics are discussed, underscoring data analytics' potential for patient-focused health care. The review concludes that successful implementation requires addressing gaps, collaboration and innovation in biomedical science and data analytics.

Review of Surgical Interventions in the Thyroid Gland: Recent Advances and Current Considerations

The thyroid gland, located at the base of the neck, regulates metabolism and hormone balance through hormones like T4 and T3, which are essential for growth, neurological development, and energy production. Thyroid diseases affect 10% of the global population, making accurate and up-to-date information on surgical interventions and advancements crucial for improving clinical outcomes. Thyroid gland surgery is a dynamic field that has experienced remarkable advances in diagnosis, surgical techniques, and postoperative management. These include new advances in surgical techniques that improve precision, reduce surgical trauma, and speed up patient recovery, identification of biomarkers, and understanding of the molecular characteristics of tumors that allow for more targeted therapeutic strategies, and incorporation of advanced technologies that improve diagnostic accuracy and efficacy. This review aims to guide healthcare professionals and lay the groundwork for future research and innovative treatments in thyroid surgery.

Multimodal MRI Deep Learning for Predicting Central Lymph Node Metastasis in Papillary Thyroid Cancer

BACKGROUND

Central lymph node metastasis (CLNM) in papillary thyroid cancer (PTC) significantly influences surgical decision-making strategies.

OBJECTIVES

This study aims to develop a predictive model for CLNM in PTC patients using magnetic resonance imaging (MRI) and clinicopathological data.

METHODS

By incorporating deep learning (DL) algorithms, the model seeks to address the challenges in diagnosing CLNM and reduce overtreatment. The results were compared with traditional machine learning (ML) models. In this retrospective study, preoperative MRI data from 105 PTC patients were divided into training and testing sets. A radiologist manually outlined the region of interest (ROI) on MRI images. Three classic ML algorithms (support vector machine [SVM], logistic regression [LR], and random forest [RF]) were employed across different data modalities. Additionally, an AMMCNet utilizing convolutional neural networks (CNNs) was proposed to develop DL models for CLNM. Predictive performance was evaluated using receiver operator characteristic (ROC) curve analysis, and clinical utility was assessed through decision curve analysis (DCA).

RESULTS

Lesion diameter was identified as an independent risk factor for CLNM. Among ML models, the RF-(T1WI + T2WI, T1WI + T2WI + Clinical) models achieved the highest area under the curve (AUC) at 0.863. The DL fusion model surpassed all ML fusion models with an AUC of 0.891.

CONCLUSIONS

A fusion model based on the AMMCNet architecture using MRI images and clinicopathological data was developed, effectively predicting CLNM in PTC patients.

Radiomics Combined with ACR TI-RADS for Thyroid Nodules: Diagnostic Performance, Unnecessary Biopsy Rate, and Nomogram Construction

RATIONALE AND OBJECTIVES

To develop a radiomics model with enhanced diagnostic performance, reduced unnecessary fine needle aspiration biopsy (FNA) rate, and improved clinical net benefit for thyroid nodules.

METHODS

We conducted a retrospective study of 217 thyroid nodules. Lesions were divided into training (n = 152) and verification (n = 65) cohorts. Three radiomics scores were derived from B-mode ultrasound (B-US) and strain elastography (SE) images, alone and in combination. A radiomics nomogram was constructed by combining high-frequency ultrasonic features and the best-performing radiomics score. The area under the receiver operating characteristic curve (AUC), unnecessary FNA rate, and decision curve analysis (DCA) results for the nomogram were compared to those obtained with the American College of Radiology Thyroid Imaging, Reporting and Data System (ACR TI-RADS) score and the combined TI-RADS+SE+ contrast-enhanced ultrasound (CEUS) advanced clinical score.

RESULTS

The three radiomics scores (B-US, SE, B-US+SE) achieved training AUCs of 0.753 (0.668-0.825), 0.761 (0.674-0.838), and 0.795 (0.715-0.871), and validation AUCs of 0.732 (0.579-0.867), 0.753 (0.609-0.892), and 0.752 (0.592-0.899) respectively. The AUC of the nomogram for the entire patient cohort was 0.909 (0.864-0.954), which was higher than that of the ACR TI-RADS score (P < 0.001) and equivalent to the TI-RADS+SE+CEUS score (P = 0.753). Similarly, the unnecessary FNA rate of the radiomics nomogram was significantly lower than that of the ACR TI-RADS score (P = 0.007) and equivalent to the TI-RADS+SE+CEUS score (P = 0.457). DCA also showed that the radiomics nomogram brought more net clinical benefit than the ACR TI-RADS score but was similar to that of the TI-RADS+SE+CEUS score.

CONCLUSION

The radiomics nomogram developed in this study can be used as an objective, accurate, cost-effective, and noninvasive method for the characterization of thyroid nodules.

Using multimodal ultrasound including full-time-series contrast-enhanced ultrasound cines for identifying the nature of thyroid nodules

Background

Based on the conventional ultrasound images of thyroid nodules, contrast-enhanced ultrasound (CEUS) videos were analyzed to investigate whether CEUS improves the classification accuracy of benign and malignant thyroid nodules using machine learning (ML) radiomics and compared with radiologists.

Materials and methods

The B-mode ultrasound (B-US), real-time elastography (RTE), color doppler flow images (CDFI) and CEUS cines of patients from two centers were retrospectively gathered. Then, the region of interest (ROI) was delineated to extract the radiomics features. Seven ML algorithms combined with four kinds of radiomics data (B-US, B-US + CDFI + RTE, CEUS, and B-US + CDFI + RTE + CEUS) were applied to establish 28 models. The diagnostic performance of ML models was compared with interpretations from expert and nonexpert readers.

Results

A total of 181 thyroid nodules from 181 patients of 64 men (mean age, 42 years +/- 12) and 117 women (mean age, 46 years +/- 12) were included. Adaptive boosting (AdaBoost) achieved the highest area under the receiver operating characteristic curve (AUC) of 0.89 in the test set among 28 models when combined with B-US + CDFI + RTE + CEUS data and an AUC of 0.72 and 0.66 when combined with B-US and B-US + CDFI + RTE data. The AUC achieved by senior and junior radiologists was 0.78 versus (vs.) 0.69 (p > 0.05), 0.79 vs. 0.64 (p < 0.05), and 0.88 vs. 0.69 (p < 0.05) combined with B-US, B-US+CDFI+RTE and B-US+CDFI+RTE+CEUS, respectively.

Conclusion

With the addition of CEUS, the diagnostic performance was enhanced for all seven classifiers and senior radiologists based on conventional ultrasound images, while no enhancement was observed for junior radiologists. The diagnostic performance of ML models was similar to senior radiologists, but superior to those junior radiologists.

A Comparative Study of Two Radiomics-Based Blood Flow Modes with Thyroid Imaging Reporting and Data System in Predicting Malignancy of Thyroid Nodules and Reducing Unnecessary Fine-Needle Aspiration Rate

RATIONALE AND OBJECTIVES

We aimed to compare superb microvascular imaging (SMI)-based radiomics methods, and contrast-enhanced ultrasound (CEUS)-based radiomics methods to the American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS) for classifying thyroid nodules (TNs) and reducing unnecessary fine-needle aspiration biopsy (FNAB) rate.

MATERIALS AND METHODS

This retrospective study enrolled a dataset of 472 pathologically confirmed TNs. Radiomics characteristics were extracted from B-mode ultrasound (BMUS), SMI, and CEUS images, respectively. After eliminating redundant features, four radiomics scores (Rad-scores) were constructed. Using multivariable logistic regression analysis, four radiomics prediction models incorporating Rad-score and corresponding US features were constructed and validated in terms of discrimination, calibration, decision curve analysis, and unnecessary FNAB rate.

RESULTS

The diagnostic performance of the BMUS + SMI radiomics method was better than ACR TI-RADS (area under the curve [AUC]: 0.875 vs. 0.689 for the training cohort, 0.879 vs. 0.728 for the validation cohort) (P < 0.05), and comparable with BMUS + CEUS radiomics method (AUC: 0.875 vs. 0.878 for the training cohort, 0.879 vs. 0.865 for the validation cohort) (P > 0.05). Decision curve analysis showed that the BMUS+SMI radiomics method could achieve higher net benefits than the BMUS radiomics method and ACR TI-RADS when the threshold probability was between 0.13 and 0.88 in the entire cohort. When applying the BMUS+SMI radiomics method, the unnecessary FNAB rate reduced from 43.4% to 13.9% in the training cohort and from 45.6% to 18.0% in the validation cohorts in comparison to ACR TI-RADS.

CONCLUSION

The dual-modal SMI-based radiomics method is convenient and economical and can be an alternative to the dual-modal CEUS-based radiomics method in helping radiologists select the optimal clinical strategy for TN management.

An Improved k-Nearest Neighbor Algorithm for Recognition and Classification of Thyroid Nodules

OBJECTIVES

To complete the task of automatic recognition and classification of thyroid nodules and solve the problem of high classification error rates when the samples are imbalanced.

METHODS

An improved k-nearest neighbor (KNN) algorithm is proposed and a method for automatic thyroid nodule classification based on the improved KNN algorithm is established. In the improved KNN algorithm, we consider not only the number of class labels for various classes of data in KNNs, but also the corresponding weights. And we use the Minkowski distance measure instead of the Euclidean distance measure.

RESULTS

A total of 508 ultrasound images of thyroid nodules, including 415 benign nodules and 93 malignant nodules, were used in the paper. Experimental results show the improved KNN has 0.872549 accuracy, 0.867347 precision, 1 recall, and 0.928962 F1-score. At the same time, we also considered the influence of different distance weights, the value of k, different distance measures on the classification results.

CONCLUSIONS

A comparison result shows that our method has a better performance than the traditional KNN and other classical machine learning methods.

Predicting the malignancy of extremity soft-tissue tumors by an ultrasound-based radiomics signature

BACKGROUND

Accurate differentiation of extremity soft-tissue tumors (ESTTs) is important for treatment planning.

PURPOSE

To develop and validate an ultrasound (US) image-based radiomics signature to predict ESTTs malignancy.

MATERIAL AND METHODS

A dataset of US images from 108 ESTTs were retrospectively enrolled and divided into the training cohort (78 ESTTs) and validation cohort (30 ESTTs). A total of 1037 radiomics features were extracted from each US image. The most useful predictive radiomics features were selected by the maximum relevance and minimum redundancy method, least absolute shrinkage, and selection operator algorithm in the training cohort. A US-based radiomics signature was built based on these selected radiomics features. In addition, a conventional radiologic model based on the US features from the interpretation of two experienced radiologists was developed by a multivariate logistic regression algorithm. The diagnostic performances of the selected radiomics features, the US-based radiomics signature, and the conventional radiologic model for differentiating ESTTs were evaluated and compared in the validation cohort.

RESULTS

In the validation cohort, the area under the curve (AUC), sensitivity, and specificity of the US-based radiomics signature for predicting ESTTs malignancy were 0.866, 84.2%, and 81.8%, respectively. The US-based radiomics signature had better diagnostic predictability for predicting ESTT malignancy than the best single radiomics feature and the conventional radiologic model (AUC = 0.866 vs. 0.719 vs. 0.681 for the validation cohort, all P <0.05).

CONCLUSION

The US-based radiomics signature could provide a potential imaging biomarker to accurately predict ESTT malignancy.

Deep learning to assist composition classification and thyroid solid nodule diagnosis: a multicenter diagnostic study

OBJECTIVES

This study aimed to propose a deep learning (DL)-based framework for identifying the composition of thyroid nodules and assessing their malignancy risk.

METHODS

We conducted a retrospective multicenter study using ultrasound images from four hospitals. Convolutional neural network (CNN) models were constructed to classify ultrasound images of thyroid nodules into solid and non-solid, as well as benign and malignant. A total of 11,201 images of 6784 nodules were used for training, validation, and testing. The area under the receiver-operating characteristic curve (AUC) was employed as the primary evaluation index.

RESULTS

The models had AUCs higher than 0.91 in the benign and malignant grading of solid thyroid nodules, with the Inception-ResNet AUC being the highest at 0.94. In the test set, the best algorithm for identifying benign and malignant thyroid nodules had a sensitivity of 0.88, and a specificity of 0.86. In the human vs. DL test set, the best algorithm had a sensitivity of 0.93, and a specificity of 0.86. The Inception-ResNet model performed better than the senior physicians (p < 0.001). The sensitivity and specificity of the optimal model based on the external test set were 0.90 and 0.75, respectively.

CONCLUSIONS

This research demonstrates that CNNs can assist thyroid nodule diagnosis and reduce the rate of unnecessary fine-needle aspiration (FNA).

CLINICAL RELEVANCE STATEMENT

High-resolution ultrasound has led to increased detection of thyroid nodules. This results in unnecessary fine-needle aspiration and anxiety for patients whose nodules are benign. Deep learning can solve these problems to some extent.

KEY POINTS

• Thyroid solid nodules have a high probability of malignancy. • Our models can improve the differentiation between benign and malignant solid thyroid nodules. • The differential performance of one model was superior to that of senior radiologists. Applying this could reduce the rate of unnecessary fine-needle aspiration of solid thyroid nodules.

Ultrasound radiomics based XGBoost model to differential diagnosis thyroid nodules and unnecessary biopsy rate: Individual application of SHapley additive exPlanations

OBJECTIVES

Radiomics-based eXtreme gradient boosting (XGBoost) model was developed to differentiate benign thyroid nodules from malignant thyroid nodules and to prevent unnecessary thyroid biopsies, including positive and negative effects.

METHODS

The study evaluated a data set of ultrasound images of thyroid nodules in patients retrospectively, who initially received ultrasound-guided fine-needle aspiration biopsy (FNAB) for diagnostic purposes. According to ACR TI-RADS, a total of five ultrasound feature categories and the maximum size of the nodule were determined by four radiologists. A radiomics score was developed by the LASSO algorithm from the ultrasound-based radiomics features. An interpretative method based on Shapley additive explanation (SHAP) was developed. XGBoost was compared with ACR TI-RADS for its diagnostic performance and FNAB rate and was compared with six other machine learning models to evaluate the model performance.

RESULTS

Finally, 191 thyroid nodules were examined from 177 patients. The radiomics score were calculated using 8 features, which were selected among 789 candidate features generated from the ultrasound images. The model yielded an AUC of 93% in the training cohort and 92% in the test cohort. It outperformed traditional machine learning models in assessing the nature of thyroid nodules. Compared with ACR TI-RADS, the FNAB rate decreased from 34% to 30% in training and from 35% to 41% in test.

CONCLUSIONS

The radiomics-based XGBoost model proposed could distinguish benign and malignant thyroid nodules, thereby reduced significantly the number of unnecessary FNAB. It was effective in making preoperative decisions and managing selected patients using the SHAP visual interpretation tools.

Thyroid ultrasound and its ancillary techniques

Ultrasound (US) of the thyroid has been used as a diagnostic tool since the late 1960s. US is the most important imaging tool for diagnosing thyroid disease. In the majority of cases a correct diagnosis can already be made in synopsis of the sonographic together with clinical findings and basal thyroid hormone parameters. However, the characterization of thyroid nodules by US remains challenging. The introduction of Thyroid Imaging Reporting and Data Systems (TIRADSs) has improved diagnostic accuracy of thyroid cancer significantly. Newer techniques such as elastography, superb microvascular imaging (SMI), contrast enhanced ultrasound (CEUS) and multiparametric ultrasound (MPUS) expand diagnostic options and tools further. In addition, the use of artificial intelligence (AI) is a promising tool to improve and simplify diagnostics of thyroid nodules and there is evidence that AI can exceed the performance of humans. Combining different US techniques with the introduction of new software, the use of AI, FNB as well as molecular markers might pave the way for a completely new area of diagnostic accuracy in thyroid disease. Finally, interventional ultrasound using US-guided thermal ablation (TA) procedures are increasingly proposed as therapy options for benign as well as malignant thyroid diseases.

A novel approach to quantify calcifications of thyroid nodules in US images based on deep learning: predicting the risk of cervical lymph node metastasis in papillary thyroid cancer patients

OBJECTIVE

Based on ultrasound (US) images, this study aimed to detect and quantify calcifications of thyroid nodules, which are regarded as one of the most important features in US diagnosis of thyroid cancer, and to further investigate the value of US calcifications in predicting the risk of lymph node metastasis (LNM) in papillary thyroid cancer (PTC).

METHODS

Based on the DeepLabv3+ networks, 2992 thyroid nodules in US images were used to train a model to detect thyroid nodules, of which 998 were used to train a model to detect and quantify calcifications. A total of 225 and 146 thyroid nodules obtained from two centers, respectively, were used to test the performance of these models. A logistic regression method was used to construct the predictive models for LNM in PTCs.

RESULTS

Calcifications detected by the network model and experienced radiologists had an agreement degree of above 90%. The novel quantitative parameters of US calcification defined in this study showed a significant difference between PTC patients with and without cervical LNM (p < 0.05). The calcification parameters were beneficial to predicting the LNM risk in PTC patients. The LNM prediction model using these calcification parameters combined with patient age and other US nodular features showed a higher specificity and accuracy than the calcification parameters alone.

CONCLUSIONS

Our models not only detect the calcifications automatically, but also have value in predicting cervical LNM risk of PTC patients, thereby making it possible to investigate the relationship between calcifications and highly invasive PTC in detail.

CLINICAL RELEVANCE STATEMENT

Due to the high association of US microcalcifications with thyroid cancers, our model will contribute to the differential diagnosis of thyroid nodules in daily practice.

KEY POINTS

• We developed an ML-based network model for automatically detecting and quantifying calcifications within thyroid nodules in US images. • Three novel parameters for quantifying US calcifications were defined and verified. • These US calcification parameters showed value in predicting the risk of cervical LNM in PTC patients.

Risk stratification of gallbladder masses by machine learning-based ultrasound radiomics models: a prospective and multi-institutional study

OBJECTIVE

This study aimed to evaluate the diagnostic performance of machine learning (ML)-based ultrasound (US) radiomics models for risk stratification of gallbladder (GB) masses.

METHODS

We prospectively examined 640 pathologically confirmed GB masses obtained from 640 patients between August 2019 and October 2022 at four institutions. Radiomics features were extracted from grayscale US images and germane features were selected. Subsequently, 11 ML algorithms were separately used with the selected features to construct optimum US radiomics models for risk stratification of the GB masses. Furthermore, we compared the diagnostic performance of these models with the conventional US and contrast-enhanced US (CEUS) models.

RESULTS

The optimal XGBoost-based US radiomics model for discriminating neoplastic from non-neoplastic GB lesions showed higher diagnostic performance in terms of areas under the curves (AUCs) than the conventional US model (0.822-0.853 vs. 0.642-0.706, p < 0.05) and potentially decreased unnecessary cholecystectomy rate in a speculative comparison with performing cholecystectomy for lesions sized over 10 mm (2.7-13.8% vs. 53.6-64.9%, p < 0.05) in the validation and test sets. The AUCs of the XGBoost-based US radiomics model for discriminating carcinomas from benign GB lesions were higher than the conventional US model (0.904-0.979 vs. 0.706-0.766, p < 0.05). The XGBoost-US radiomics model performed better than the CEUS model in discriminating GB carcinomas (AUC: 0.995 vs. 0.902, p = 0.011).

CONCLUSIONS

The proposed ML-based US radiomics models possess the potential capacity for risk stratification of GB masses and may reduce the unnecessary cholecystectomy rate and use of CEUS.

CLINICAL RELEVANCE STATEMENT

The machine learning-based ultrasound radiomics models have potential for risk stratification of gallbladder masses and may potentially reduce unnecessary cholecystectomies.

KEY POINTS

• The XGBoost-based US radiomics models are useful for the risk stratification of GB masses. • The XGBoost-based US radiomics model is superior to the conventional US model for discriminating neoplastic from non-neoplastic GB lesions and may potentially decrease unnecessary cholecystectomy rate for lesions sized over 10 mm in comparison with the current consensus guideline. • The XGBoost-based US radiomics model could overmatch CEUS model in discriminating GB carcinomas from benign GB lesions.

Machine Learning Model as a Useful Tool for Prediction of Thyroid Nodules Histology, Aggressiveness and Treatment-Related Complications

Thyroid nodules are very common, 5-15% of which are malignant. Despite the low mortality rate of well-differentiated thyroid cancer, some variants may behave aggressively, making nodule differentiation mandatory. Ultrasound and fine-needle aspiration biopsy are simple, safe, cost-effective and accurate diagnostic tools, but have some potential limits. Recently, machine learning (ML) approaches have been successfully applied to healthcare datasets to predict the outcomes of surgical procedures. The aim of this work is the application of ML to predict tumor histology (HIS), aggressiveness and post-surgical complications in thyroid patients. This retrospective study was conducted at the ENT Division of Eastern Piedmont University, Novara (Italy), and reported data about 1218 patients who underwent surgery between January 2006 and December 2018. For each patient, general information, HIS and outcomes are reported. For each prediction task, we trained ML models on pre-surgery features alone as well as on both pre- and post-surgery data. The ML pipeline included data cleaning, oversampling to deal with unbalanced datasets and exploration of hyper-parameter space for random forest models, testing their stability and ranking feature importance. The main results are (i) the construction of a rich, hand-curated, open dataset including pre- and post-surgery features (ii) the development of accurate yet explainable ML models. Results highlight pre-screening as the most important feature to predict HIS and aggressiveness, and that, in our population, having an out-of-range (Low) fT3 dosage at pre-operative examination is strongly associated with a higher aggressiveness of the disease. Our work shows how ML models can find patterns in thyroid patient data and could support clinicians to refine diagnostic tools and improve their accuracy.

Value of Artificial Intelligence in Improving the Accuracy of Diagnosing TI-RADS Category 4 Nodules

OBJECTIVE

Considerable heterogeneity is observed in the malignancy rates of thyroid nodules classified as category 4 according to the Thyroid Imaging Reporting and Data System (TI-RADS). This study was aimed at comparing the diagnostic performance of artificial intelligence algorithms and radiologists with different experience levels in distinguishing benign and malignant TI-RADS 4 (TR4) nodules.

METHODS

Between January 2019 and September 2022, 1117 TR4 nodules with well-defined pathological findings were collected for this retrospective study. An independent external data set of 125 TR4 nodules was incorporated for testing purposes. Traditional feature-based machine learning (ML) models, deep convolutional neural networks (DCNN) models and a fusion model that integrated the prediction outcomes from all models were used to classify benign and malignant TR4 nodules. A fivefold cross-validation approach was employed, and the diagnostic performance of each model and radiologists was compared.

RESULTS

In the external test data set, the area under the receiver operating characteristic curve (AUROC) of the three DCNN-based secondary transfer learning models-InceptionV3, DenseNet121 and ResNet50-were 0.852, 0.837 and 0.856, respectively. These values were higher than those of the three traditional ML models-logistic regression, multilayer perceptron and random forest-at 0.782, 0.790, and 0.767, respectively, and higher than that of an experienced radiologist (0.815). The fusion diagnostic model we developed, with an AUROC of 0.880, was found to outperform the experienced radiologist in diagnosing TR4 nodules.

CONCLUSION

The integration of artificial intelligence algorithms into medical imaging studies could improve the accuracy of identifying high-risk TR4 nodules pre-operatively and have significant clinical application potential.

Application and prospects of AI-based radiomics in ultrasound diagnosis

Artificial intelligence (AI)-based radiomics has attracted considerable research attention in the field of medical imaging, including ultrasound diagnosis. Ultrasound imaging has unique advantages such as high temporal resolution, low cost, and no radiation exposure. This renders it a preferred imaging modality for several clinical scenarios. This review includes a detailed introduction to imaging modalities, including Brightness-mode ultrasound, color Doppler flow imaging, ultrasound elastography, contrast-enhanced ultrasound, and multi-modal fusion analysis. It provides an overview of the current status and prospects of AI-based radiomics in ultrasound diagnosis, highlighting the application of AI-based radiomics to static ultrasound images, dynamic ultrasound videos, and multi-modal ultrasound fusion analysis.

A novel wavelet-transform-based convolution classification network for cervical lymph node metastasis of papillary thyroid carcinoma in ultrasound images

Preoperative assessment of cervical lymph nodes metastasis (CLNM) for accurate qualitative and locating diagnosis is important for choosing the best treatment option for patients with papillary thyroid cancer. Non-destructive, non-invasive ultrasound is currently the imaging method of choice for lymph node metastatic assessment. For lymph node characteristics and ultrasound images, this paper proposes a multitasking network framework for diagnosing metastatic lymph nodes in ultrasound images, in which localization module not only provides information on the location of lymph nodes to focus on the peripheral and self regions of lymph nodes, but also provides structural features of lymph nodes for subsequent classification module. In the classification module, we design a novel wavelet-transform-based convolution network. Wavelet transform is introduced into the deep learning convolution module to analyze ultrasound images in both spatial and frequency domains, which effectively enriches the feature information and improves the classification performance of the model without increasing the model parameters. We collected 510 patient data (N = 1376) from Shanghai Sixth People's Hospital regarding ultrasound lymph nodes in the neck, as well as used three publicly available ultrasound datasets, including SCUI2020 (N = 2914), DDTI (N = 480), and BUSI (N = 780). Compared to the optimal two-stage model, our model has improved its accuracy and AUC indexes by 5.83% and 4%, which outperforms the two-stage architectures and also surpasses the latest classification networks.

Combined morphologic-metabolic biomarkers from [18F]FDG-PET/CT stratify prognostic groups in low-risk NSCLC

AIM

The aim of this study was to derive prognostic parameters from 2-[18F]fluoro-2-deoxy-D-glucose ([18F]FDG-PET/CT) in patients with low-risk NSCLC and determine their prognostic value.

METHODS

81 (21 female, mean age 66 a) therapy-naive patients that underwent [18F]FDG-PET/CT before histologic confirmation of NSCLC with stadium I and II between 2008-2016 were included. A mean follow-up time of 58 months (13-176), overall and progression free survival (OS, PFS) were registered. A volume of interest for the primary tumor was defined on PET and CT images. Parameters SUVmax, PET-solidity, PET-circularity, and CT-volume were analyzed. To evaluate the prognostic value of each parameter for OS, a minimum p-value approach was used to define cutoff values, survival analysis, and log-rank tests were performed, including subgroup analysis for combinations of parameters.

RESULTS

Mean OS was 58±28 months. Poor OS was associated with a tumor CT-volume >14.3 cm3 (p=0.02, HR=7.0, CI 2.7-17.7), higher SUVmax values >12.2 (p=0.003; HR=3.0, CI 1.3-6.7) and PET-solidity >0.919 (p=0.004; HR=3.0, CI 1.0-8.9). Combined parameter analysis revealed worse prognosis in larger volume/high SUVmax tumors compared to larger volume/lower SUVmax (p=0.028; HR=2.5, CI 1.1-5.5), high PET-solidity/low volume (p=0.01; HR=2.4, CI 0.8-6.6) and low SUVmax/high PET-solidity (p=0.02, HR=4.0, CI 0.8-19.0).

CONCLUSION

Even in this group of low-risk NSCLC patients, we identified a subgroup with a significantly worse prognosis by combining morphologic-metabolic biomarkers from [18F]FDG-PET/CT. The combination of SUVmax and CT-volume performed best. Based on these preliminary data, future prospective studies to validate this combined morphologic-metabolic imaging biomarker for potential therapeutic decisions seem promising.

Performance of ACR TI-RADS and the Bethesda System in Predicting Risk of Malignancy in Thyroid Nodules at a Large Children's Hospital and a Comprehensive Review of the Pediatric Literature

While thyroid nodules are less common in children than in adults, they are more frequently malignant. However, pediatric data are scarce regarding the performance characteristics of imaging and cytopathology classification systems validated to predict the risk of malignancy (ROM) in adults and select those patients who require fine-needle aspiration (FNA) and possibly surgical resection. We retrospectively reviewed the electronic medical records of all patients 18 years of age or younger who underwent thyroid FNA at our institution from 1 July 2015 to 31 May 2022. Based on surgical follow-up from 74 of the 208 FNA cases, we determined the ROM for the American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS) ultrasound risk stratification system and The Bethesda System for Reporting Thyroid Cytopathology and added our results to those of pediatric cohorts from other institutions already published in the literature. We found the following ROMs for 1458 cases using ACR TI-RADS (TR): TR1. Benign: 2.2%, TR2. Not Suspicious: 9.3%, TR3. Mildly Suspicious: 16.6%, TR4. Moderately Suspicious: 27.0%, and TR5. Highly Suspicious 76.5%; and for 5911 cases using the Bethesda system: Bethesda I. Unsatisfactory: 16.8%, Bethesda II. Benign: 7.2%, Bethesda III: Atypia of Undetermined Significance: 29.6%, Bethesda IV. Follicular Neoplasm: 42.3%, Bethesda V. Suspicious for Malignancy: 90.8%, and Bethesda VI. Malignant: 98.8%. We conclude that ACR TI-RADS levels imply higher ROMs for the pediatric population than the corresponding suggested ROMs for adults, and, in order to avoid missing malignancies, we should consider modifying or altogether abandoning size cutoffs for recommending FNA in children and adolescents whose thyroid glands are smaller than those of adults. The Bethesda categories also imply higher ROMs for pediatric patients compared to adults.

Predicting BRAFV600E mutations in papillary thyroid carcinoma using six machine learning algorithms based on ultrasound elastography

The most common BRAF mutation is thymine (T) to adenine (A) missense mutation in nucleotide 1796 (T1796A, V600E). The BRAFV600E gene encodes a protein-dependent kinase (PDK), which is a key component of the mitogen-activated protein kinase pathway and essential for controlling cell proliferation, differentiation, and death. The BRAFV600E mutation causes PDK to be activated improperly and continuously, resulting in abnormal proliferation and differentiation in PTC. Based on elastography ultrasound (US) radiomic features, this study seeks to create and validate six distinct machine learning algorithms to predict BRAFV6OOE mutation in PTC patients prior to surgery. This study employed routine US strain elastography image data from 138 PTC patients. The patients were separated into two groups: those who did not have the BRAFV600E mutation (n = 75) and those who did have the mutation (n = 63). The patients were randomly assigned to one of two data sets: training (70%), or validation (30%). From strain elastography US images, a total of 479 radiomic features were retrieved. Pearson's Correlation Coefficient (PCC) and Recursive Feature Elimination (RFE) with stratified tenfold cross-validation were used to decrease the features. Based on selected radiomic features, six machine learning algorithms including support vector machine with the linear kernel (SVM_L), support vector machine with radial basis function kernel (SVM_RBF), logistic regression (LR), Naïve Bayes (NB), K-nearest neighbors (KNN), and linear discriminant analysis (LDA) were compared to predict the possibility of BRAFV600E. The accuracy (ACC), the area under the curve (AUC), sensitivity (SEN), specificity (SPEC), positive predictive value (PPV), negative predictive value (NPV), decision curve analysis (DCA), and calibration curves of the machine learning algorithms were used to evaluate their performance. ① The machine learning algorithms' diagnostic performance depended on 27 radiomic features. ② AUCs for NB, KNN, LDA, LR, SVM_L, and SVM_RBF were 0.80 (95% confidence interval [CI]: 0.65-0.91), 0.87 (95% CI 0.73-0.95), 0.91(95% CI 0.79-0.98), 0.92 (95% CI 0.80-0.98), 0.93 (95% CI 0.80-0.98), and 0.98 (95% CI 0.88-1.00), respectively. ③ There was a significant difference in echogenicity,vertical and horizontal diameter ratios, and elasticity between PTC patients with BRAFV600E and PTC patients without BRAFV600E. Machine learning algorithms based on US elastography radiomic features are capable of predicting the likelihood of BRAFV600E in PTC patients, which can assist physicians in identifying the risk of BRAFV600E in PTC patients. Among the six machine learning algorithms, the support vector machine with radial basis function (SVM_RBF) achieved the best ACC (0.93), AUC (0.98), SEN (0.95), SPEC (0.90), PPV (0.91), and NPV (0.95).

Artificial Intelligence in Thyroidology: A Narrative Review of the Current Applications, Associated Challenges, and Future Directions

Background: The use of artificial intelligence (AI) in health care has grown exponentially with the promise of facilitating biomedical research and enhancing diagnosis, treatment, monitoring, disease prevention, and health care delivery. We aim to examine the current state, limitations, and future directions of AI in thyroidology. Summary: AI has been explored in thyroidology since the 1990s, and currently, there is an increasing interest in applying AI to improve the care of patients with thyroid nodules (TNODs), thyroid cancer, and functional or autoimmune thyroid disease. These applications aim to automate processes, improve the accuracy and consistency of diagnosis, personalize treatment, decrease the burden for health care professionals, improve access to specialized care in areas lacking expertise, deepen the understanding of subtle pathophysiologic patterns, and accelerate the learning curve of less experienced clinicians. There are promising results for many of these applications. Yet, most are in the validation or early clinical evaluation stages. Only a few are currently adopted for risk stratification of TNODs by ultrasound and determination of the malignant nature of indeterminate TNODs by molecular testing. Challenges of the currently available AI applications include the lack of prospective and multicenter validations and utility studies, small and low diversity of training data sets, differences in data sources, lack of explainability, unclear clinical impact, inadequate stakeholder engagement, and inability to use outside of the research setting, which might limit the value of their future adoption. Conclusions: AI has the potential to improve many aspects of thyroidology; however, addressing the limitations affecting the suitability of AI interventions in thyroidology is a prerequisite to ensure that AI provides added value for patients with thyroid disease.

Artificial intelligence - based ultrasound elastography for disease evaluation - a narrative review

Ultrasound elastography (USE) provides complementary information of tissue stiffness and elasticity to conventional ultrasound imaging. It is noninvasive and free of radiation, and has become a valuable tool to improve diagnostic performance with conventional ultrasound imaging. However, the diagnostic accuracy will be reduced due to high operator-dependence and intra- and inter-observer variability in visual observations of radiologists. Artificial intelligence (AI) has great potential to perform automatic medical image analysis tasks to provide a more objective, accurate and intelligent diagnosis. More recently, the enhanced diagnostic performance of AI applied to USE have been demonstrated for various disease evaluations. This review provides an overview of the basic concepts of USE and AI techniques for clinical radiologists and then introduces the applications of AI in USE imaging that focus on the following anatomical sites: liver, breast, thyroid and other organs for lesion detection and segmentation, machine learning (ML) - assisted classification and prognosis prediction. In addition, the existing challenges and future trends of AI in USE are also discussed.

Artificial intelligence in thyroid ultrasound

Artificial intelligence (AI), particularly deep learning (DL) algorithms, has demonstrated remarkable progress in image-recognition tasks, enabling the automatic quantitative assessment of complex medical images with increased accuracy and efficiency. AI is widely used and is becoming increasingly popular in the field of ultrasound. The rising incidence of thyroid cancer and the workload of physicians have driven the need to utilize AI to efficiently process thyroid ultrasound images. Therefore, leveraging AI in thyroid cancer ultrasound screening and diagnosis cannot only help radiologists achieve more accurate and efficient imaging diagnosis but also reduce their workload. In this paper, we aim to present a comprehensive overview of the technical knowledge of AI with a focus on traditional machine learning (ML) algorithms and DL algorithms. We will also discuss their clinical applications in the ultrasound imaging of thyroid diseases, particularly in differentiating between benign and malignant nodules and predicting cervical lymph node metastasis in thyroid cancer. Finally, we will conclude that AI technology holds great promise for improving the accuracy of thyroid disease ultrasound diagnosis and discuss the potential prospects of AI in this field.

Automatic detection and diagnosis of thyroid ultrasound images based on attention mechanism

Incidents of thyroid cancer have dramatically increased in recent years; however, early ultrasound diagnosis can reduce morbidity and mortality. The work in clinical situations relies heavily on the subjective experience of the sonographer. Numerous computer-aided diagnostic techniques exist, but most consider how good the results are, ignoring the pre-image collecting and its usefulness in post-clinical practise. To address these issues, this study proposes a computer-aided diagnosis method based on an attentional mechanism. Due to its lightweight properties, the model can rapidly identify nodules and distinguish between benign and malignant ones without using much hardware. The model uses a bounding box to locate the thyroid nodule and determines whether it is benign or cancerous, and outputs the diagnostic result of the thyroid nodule ultrasound images. The latest attention mechanisms are used to get better results at a fraction of the cost. Additionally, ultrasound images with different features of benign and malignant thyroid nodules were collected following the Thyroid Imaging Reporting and Data System standards. The experimental results showed that the approach identifies and classifies thyroid nodules rapidly and effectively; the mAP value of the results reached 0.89, and the mAP value of malignant nodules reached 0.94, with detection rate of single image reached 7 ms. Young physicians and small hospitals with limited resources can benefit from using this method to assist with thyroid ultrasound examination diagnosis.

Endocrine Tumor Classification via Machine-Learning-Based Elastography: A Systematic Scoping Review

Elastography complements traditional medical imaging modalities by mapping tissue stiffness to identify tumors in the endocrine system, and machine learning models can further improve diagnostic accuracy and reliability. Our objective in this review was to summarize the applications and performance of machine-learning-based elastography on the classification of endocrine tumors. Two authors independently searched electronic databases, including PubMed, Scopus, Web of Science, IEEEXpress, CINAHL, and EMBASE. Eleven (n = 11) articles were eligible for the review, of which eight (n = 8) focused on thyroid tumors and three (n = 3) considered pancreatic tumors. In all thyroid studies, the researchers used shear-wave ultrasound elastography, whereas the pancreas researchers applied strain elastography with endoscopy. Traditional machine learning approaches or the deep feature extractors were used to extract the predetermined features, followed by classifiers. The applied deep learning approaches included the convolutional neural network (CNN) and multilayer perceptron (MLP). Some researchers considered the mixed or sequential training of B-mode and elastographic ultrasound data or fusing data from different image segmentation techniques in machine learning models. All reviewed methods achieved an accuracy of ≥80%, but only three were ≥90% accurate. The most accurate thyroid classification (94.70%) was achieved by applying sequential training CNN; the most accurate pancreas classification (98.26%) was achieved using a CNN-long short-term memory (LSTM) model integrating elastography with B-mode and Doppler images.

Models of ultrasonic radiomics and clinical characters for lymph node metastasis assessment in thyroid cancer: a retrospective study

Background

Preoperative prediction of cervical lymph node metastasis in papillary thyroid carcinoma provided a basis for tumor staging and treatment decision. This study aimed to investigate the utility of machine learning and develop different models to preoperatively predict cervical lymph node metastasis based on ultrasonic radiomic features and clinical characteristics in papillary thyroid carcinoma nodules.

Methods

Data from 400 papillary thyroid carcinoma nodules were included and divided into training and validation group. With the help of machine learning, clinical characteristics and ultrasonic radiomic features were extracted and selected using randomforest and least absolute shrinkage and selection operator regression before classified by five classifiers. Finally, 10 models were built and their area under the receiver operating characteristic curve, accuracy, sensitivity, specificity, positive predictive value and negative predictive value were measured.

Results

Among the 10 models, RF-RF model revealed the highest area under curve (0.812) and accuracy (0.7542) in validation group. The top 10 variables of it included age, seven textural features, one shape feature and one first-order feature, in which eight were high-dimensional features.

Conclusions

RF-RF model showed the best predictive performance for cervical lymph node metastasis. And the importance features selected by it highlighted the unique role of higher-dimensional statistical methods for radiomics analysis.

Ultrasound Elastography: Basic Principles and Examples of Clinical Applications with Artificial Intelligence—A Review

Ultrasound elastography (USE) or elastosonography is an ultrasound-based, non-invasive imaging method for assessing tissue elasticity. The different types of elastosonography are distinguished according to the mechanisms used for estimating tissue elasticity and the type of information they provide. In strain imaging, mechanical stress is applied to the tissue, and the resulting differential strain between different tissues is used to provide a qualitative assessment of elasticity. In shear wave imaging, tissue elasticity is inferred through quantitative parameters, such as shear wave velocity or longitudinal elastic modulus. Shear waves can be produced using a vibrating mechanical device, as in transient elastography (TE), or an acoustic impulse, which can be highly focused, as in point-shear wave elastography (p-SWE), or directed to multiple zones in a two-dimensional area, as in 2D-SWE. A general understanding of the basic principles behind each technique is important for clinicians to improve data acquisition and interpretation. Major clinical applications include chronic liver disease, breast lesions, thyroid nodules, lymph node malignancies, and inflammatory bowel disease. The integration of artificial intelligence tools could potentially overcome some of the main limitations of elastosonography, such as operator dependence and low specificity, allowing for its effective integration into clinical workflow.

Incorporation of a machine learning pathological diagnosis algorithm into the thyroid ultrasound imaging data improves the diagnosis risk of malignant thyroid nodules

OBJECTIVE

This study aimed at establishing a new model to predict malignant thyroid nodules using machine learning algorithms.

METHODS

A retrospective study was performed on 274 patients with thyroid nodules who underwent fine-needle aspiration (FNA) cytology or surgery from October 2018 to 2020 in Xianyang Central Hospital. The least absolute shrinkage and selection operator (lasso) regression analysis and logistic analysis were applied to screen and identified variables. Six machine learning algorithms, including Decision Tree (DT), Extreme Gradient Boosting (XGBoost), Gradient Boosting Machine (GBM), Naive Bayes Classifier (NBC), Random Forest (RF), and Logistic Regression (LR), were employed and compared in constructing the predictive model, coupled with preoperative clinical characteristics and ultrasound features. Internal validation was performed by using 10-fold cross-validation. The performance of the model was measured by the area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, F1 score, Shapley additive explanations (SHAP) plot, feature importance, and correlation of features. The best cutoff value for risk stratification was identified by probability density function (PDF) and clinical utility curve (CUC).

RESULTS

The malignant rate of thyroid nodules in the study cohort was 53.2%. The predictive models are constructed by age, margin, shape, echogenic foci, echogenicity, and lymph nodes. The XGBoost model was significantly superior to any one of the machine learning models, with an AUC value of 0.829. According to the PDF and CUC, we recommended that 51% probability be used as a threshold for determining the risk stratification of malignant nodules, where about 85.6% of patients with malignant nodules could be detected. Meanwhile, approximately 89.8% of unnecessary biopsy procedures would be saved. Finally, an online web risk calculator has been built to estimate the personal likelihood of malignant thyroid nodules based on the best-performing ML-ed model of XGBoost.

CONCLUSIONS

Combining clinical characteristics and features of ultrasound images, ML algorithms can achieve reliable prediction of malignant thyroid nodules. The online web risk calculator based on the XGBoost model can easily identify in real-time the probability of malignant thyroid nodules, which can assist clinicians to formulate individualized management strategies for patients.

Advancements in Thyroidectomy: A Mini Review

Demand for minimally invasive surgery has driven the development of new gadgets and surgical techniques. Yet, questions about safety and skeptical views on new technology have prevented proliferation of new modes of surgery. This skepticism is perhaps due to unfamiliarity of new fields. Likewise, there are currently various remote-access techniques available for thyroid surgeons that only few regions in the world have adapted. This review will explore the history of minimally invasive techniques in thyroid surgery and introduce new technology to be implemented.

Multimodal ultrasound imaging: A method to improve the accuracy of diagnosing thyroid TI-RADS 4 nodules

Thyroid nodule is a common and frequently occurring disease in the neck in recent years, and ultrasound has become the preferred imaging diagnosis method for thyroid nodule due to its advantages of noninvasive, nonradiation, real-time, and repeatable. The thyroid imaging, reporting and data system (TI-RADS) classification standard scores suspicious nodules that are difficult to determine benign and malignant as grade 4, and further pathological puncture is recommended clinically, which may lead to a large number of unnecessary biopsies and operations. Including conventional ultrasound, ACR TI-RADS, shear wave elastography, super microvascular imaging, contrast enhanced ultrasound, "firefly," artificial intelligence, and multimodal ultrasound imaging used in combination. In order to identify the most clinically significant malignant tumors when reducing invasive operations. This article reviews the application and research progress of multimodal ultrasound imaging in the diagnosis of TI-RADS 4 thyroid nodules.

Malignancy risk stratification of thyroid nodules according to echotexture and degree of hypoechogenicity: a retrospective multicenter validation study

Various risk stratification systems show discrepancies in the ultrasound lexicon of nodule echotexture and hypoechogenicity. This study aimed to determine the malignancy risk of thyroid nodules according to their echotexture and degree of hypoechogenicity. From June to September 2015, we retrospectively evaluated 5601 thyroid nodules with final diagnoses from 26 institutions. Nodules were stratified according to the echotexture (homogeneous vs. heterogeneous) and degree of hypoechogenicity (mild, moderate, or marked). We calculated the malignancy risk according to composition and suspicious features. Heterogeneous hypoechoic nodules showed a significantly higher malignancy risk than heterogeneous isoechoic nodules (P ≤ 0.017), except in partially cystic nodules. Malignancy risks were not significantly different between homogeneous versus heterogeneous nodules in both hypoechoic (P ≥ 0.086) and iso- hyperechoic nodules (P ≥ 0.05). Heterogeneous iso-hyperechoic nodules without suspicious features showed a low malignancy risk. The malignancy risks of markedly and moderately hypoechoic nodules were not significantly different in all subgroups (P ≥ 0.48). Marked or moderately hypoechoic nodules showed a significantly higher risk than mild hypoechoic (P ≤ 0.016) nodules. The predominant echogenicity effectively stratifies the malignancy risk of nodules with heterogeneous echotexture. The degree of hypoechogenicity could be stratified as mild versus moderate to marked hypoechogenicity.

Risk factors and diagnostic prediction models for papillary thyroid carcinoma

Thyroid nodules (TNs) represent a common scenario. More accurate pre-operative diagnosis of malignancy has become an overriding concern. This study incorporated demographic, serological, ultrasound, and biopsy data and aimed to compare a new diagnostic prediction model based on Back Propagation Neural Network (BPNN) with multivariate logistic regression model, to guide the decision of surgery. Records of 2,090 patients with TNs who underwent thyroid surgery were retrospectively reviewed. Multivariate logistic regression analysis indicated that Bethesda category (OR=1.90, P<0.001), TIRADS (OR=2.55, P<0.001), age (OR=0.97, P=0.002), nodule size (OR=0.53, P<0.001), and serum levels of Tg (OR=0.994, P=0.004) and HDL-C (OR=0.23, P=0.001) were statistically significant independent differentiators for patients with PTC and benign nodules. Both BPNN and regression models showed good accuracy in differentiating PTC from benign nodules (area under the curve [AUC], 0.948 and 0.924, respectively). Notably, the BPNN model showed a higher specificity (88.3% vs. 73.9%) and negative predictive value (83.7% vs. 45.8%) than the regression model, while the sensitivity (93.1% vs. 93.9%) was similar between two models. Stratified analysis based on Bethesda indeterminate cytology categories showed similar findings. Therefore, BPNN and regression models based on a combination of demographic, serological, ultrasound, and biopsy data, all of which were readily available in routine clinical practice, might help guide the decision of surgery for TNs.

Automated classification of cervical lymph-node-level from ultrasound using Depthwise Separable Convolutional Swin Transformer

There are few studies on cervical ultrasound lymph-node-level classification which is very important for qualitative diagnosis and surgical treatment of diseases. Currently, ultrasound examination relies on the subjective experience of physicians to judge the level of the cervical lymph nodes, which is easily misclassified. Unlike other automated diagnostic tasks, lymph-node-level classification needs to focus on global structural information. Besides, there is a large range of sternocleidomastoid muscles in levels II, III and IV, which leads to small inter-class differences in these levels, so it also needs to focus on key local areas to extract strong distinguishable features. In this paper, we propose the Depthwise Separable Convolutional Swin Transformer, introducing the deepwise separable convolution branch into the self-attention mechanism to capture discriminative local features. Meanwhile, to address the problem of data imbalance, a new loss function is proposed to improve the performance of the classification network. In addition, for the ultrasound data collected by different devices, low contrast and blurring problems of ultrasound imaging, a unified pre-processing algorithm is designed. The model was validated on 1146 cases of cervical ultrasound lymph node collected from the Sixth People's Hospital of Shanghai. The average accuracy precision, sensitivity, specificity, and F1 value of the model for the valid dataset after five-fold cross-validation were 80.65%, 80.68%, 78.73%, 95.99% and 79.42%, respectively. It has been verified by visualization methods that the Region of Interest (ROI) of the model is similar or consistent with the observed region of the experts.

SuperSonic shear imaging for the differentiation between benign and malignant thyroid nodules: a meta-analysis

PURPOSE

To assess the diagnostic value of SuperSonic shear imaging (SSI) for the differentiation between benign and malignant thyroid nodules through meta-analysis.

METHODS

Online database searches were performed on PubMed, EMBASE, the Cochrane Library, and the Web of Science until 31 July 2021. The Quality Assessment of Diagnostic Accuracy Studies-2 tool was used to assess the quality of the included studies. Three measures of diagnostic test performance were used to examine the value of SSI, including the summary area under the receiver operating characteristic curve (AUROC), the summary diagnostic odds ratio (DOR), and the summary sensitivity and specificity. Heterogeneity was explored using meta-regression and subgroup analyses.

RESULTS

Finally, 21 studies with 3376 patients were included in this study. There were a total of 4296 thyroid nodules, in which 1806 malignant nodules and 2490 benign ones were involved. Thyroid nodules exhibited a malignancy rate of 42.0% (range 5.6-79.8%), 95.1% of which were of papillary variant. SSI showed a summary sensitivity of 74% [95% confidence interval (CI) 67-79%], specificity of 82% (95% CI 77-87%) and AUROC of 0.85 (95% CI 0.82-0.88) for the differentiation between benign and malignant thyroid nodules. The summary positive likelihood ratio (LR), negative LR, and DOR were 4.2 (95% CI 3.3-5.3), 0.32 (95% CI 0.26-0.40), and 13 (95% CI 9-18), respectively.

CONCLUSIONS

SSI showed high accuracy in the diagnostic differentiation between benign and malignant thyroid nodules and can be served as a noninvasive and important adjunct for thyroid nodule evaluation.

Differential Diagnosis of Thyrotoxicosis by Machine Learning Models with Laboratory Findings

Differential diagnosis of thyrotoxicosis is essential because therapeutic approaches differ based on disease etiology. We aimed to perform differential diagnosis of thyrotoxicosis using machine learning algorithms with initial laboratory findings. This is a retrospective study through medical records. Patients who visited a single hospital for thyrotoxicosis from June 2016 to December 2021 were enrolled. In total, 230 subjects were analyzed: 124 (52.6%) patients had Graves’ disease, 65 (28.3%) suffered from painless thyroiditis, and 41 (17.8%) were diagnosed with subacute thyroiditis. In consideration that results for the thyroid autoantibody test cannot be immediately confirmed, two different models were devised: Model 1 included triiodothyronine (T3), free thyroxine (FT4), T3 to FT4 ratio, erythrocyte sediment rate, and C-reactive protein (CRP); and Model 2 included all Model 1 variables as well as thyroid autoantibody test results, including thyrotropin binding inhibitory immunoglobulin (TBII), thyroid-stimulating immunoglobulin, anti-thyroid peroxidase antibody, and anti-thyroglobulin antibody (TgAb). Differential diagnosis accuracy was calculated using seven machine learning algorithms. In the initial blood test, Graves’ disease was characterized by increased thyroid hormone levels and subacute thyroiditis showing elevated inflammatory markers. The diagnostic accuracy of Model 1 was 65–70%, and Model 2 accuracy was 78–90%. The random forest model had the highest classification accuracy. The significant variables were CRP and T3 in Model 1 and TBII, CRP, and TgAb in Model 2. We suggest monitoring the initial T3 and CRP levels with subsequent confirmation of TBII and TgAb in the differential diagnosis of thyrotoxicosis.

Development of Artificial Intelligence for Parathyroid Recognition During Endoscopic Thyroid Surgery

OBJECTIVE

We aimed to establish an artificial intelligence (AI) model to identify parathyroid glands during endoscopic approaches and compare it with senior and junior surgeons' visual estimation.

METHODS

A total of 1,700 images of parathyroid glands from 166 endoscopic thyroidectomy videos were labeled. Data from 20 additional full-length videos were used as an independent external cohort. The YOLO V3, Faster R-CNN, and Cascade algorithms were used for deep learning, and the optimal algorithm was selected for independent external cohort analysis. Finally, the identification rate, initial recognition time, and tracking periods of PTAIR (Artificial Intelligence model for Parathyroid gland Recognition), junior surgeons, and senior surgeons were compared.

RESULTS

The Faster R-CNN algorithm showed the best balance after optimizing the hyperparameters of each algorithm and was updated as PTAIR. The precision, recall rate, and F1 score of the PTAIR were 88.7%, 92.3%, and 90.5%, respectively. In the independent external cohort, the parathyroid identification rates of PTAIR, senior surgeons, and junior surgeons were 96.9%, 87.5%, and 71.9%, respectively. In addition, PTAIR recognized parathyroid glands 3.83 s ahead of the senior surgeons (p = 0.008), with a tracking period 62.82 s longer than the senior surgeons (p = 0.006).

CONCLUSIONS

PTAIR can achieve earlier identification and full-time tracing under a particular training strategy. The identification rate of PTAIR is higher than that of junior surgeons and similar to that of senior surgeons. Such systems may have utility in improving surgical outcomes and also in accelerating the education of junior surgeons.

LEVEL OF EVIDENCE

3 Laryngoscope, 2022.

A machine-learning algorithm for distinguishing malignant from benign indeterminate thyroid nodules using ultrasound radiomic features

Background: Ultrasound (US)-guided fine needle aspiration (FNA) cytology is the gold standard for the evaluation of thyroid nodules. However, up to 30% of FNA results are indeterminate, requiring further testing. In this study, we present a machine-learning analysis of indeterminate thyroid nodules on ultrasound with the aim to improve cancer diagnosis. Methods: Ultrasound images were collected from two institutions and labeled according to their FNA (F) and surgical pathology (S) diagnoses [malignant (M), benign (B), and indeterminate (I)]. Subgroup breakdown (FS) included: 90 BB, 83 IB, 70 MM, and 59 IM thyroid nodules. Margins of thyroid nodules were manually annotated, and computerized radiomic texture analysis was conducted within tumor contours. Initial investigation was conducted using five-fold cross-validation paradigm with a two-class Bayesian artificial neural networks classifier, including stepwise feature selection. Testing was conducted on an independent set and compared with a commercial molecular testing platform. Performance was evaluated using receiver operating characteristic analysis in the task of distinguishing between malignant and benign nodules. Results: About 1052 ultrasound images from 302 thyroid nodules were used for radiomic feature extraction and analysis. On the training/validation set comprising 263 nodules, five-fold cross-validation yielded area under curves (AUCs) of 0.75 [Standard Error (SE) = 0.04; P < 0.001 ] and 0.67 (SE = 0.05; P = 0.0012 ) for the classification tasks of MM versus BB, and IM versus IB, respectively. On an independent test set of 19 IM/IB cases, the algorithm for distinguishing indeterminate nodules yielded an AUC value of 0.88 (SE = 0.09; P < 0.001 ), which was higher than the AUC of a commercially available molecular testing platform (AUC = 0.81, SE = 0.11; P < 0.005 ). Conclusion: Machine learning of computer-extracted texture features on gray-scale ultrasound images showed promising results classifying indeterminate thyroid nodules according to their surgical pathology.

Radiomic Detection of Malignancy within Thyroid Nodules Using Ultrasonography-A Systematic Review and Meta-Analysis

Background: Despite investigation, 95% of thyroid nodules are ultimately benign. Radiomics is a field that uses radiological features to inform individualized patient care. We aimed to evaluate the diagnostic utility of radiomics in classifying undetermined thyroid nodules into benign and malignant using ultrasonography (US). Methods: A diagnostic test accuracy systematic review and meta-analysis was performed in accordance with PRISMA guidelines. Sensitivity, specificity, and area under curve (AUC) delineating benign and malignant lesions were recorded. Results: Seventy-five studies including 26,373 patients and 46,175 thyroid nodules met inclusion criteria. Males accounted for 24.6% of patients, while 75.4% of patients were female. Radiomics provided a pooled sensitivity of 0.87 (95% CI: 0.86−0.87) and a pooled specificity of 0.84 (95% CI: 0.84−0.85) for characterizing benign and malignant lesions. Using convolutional neural network (CNN) methods, pooled sensitivity was 0.85 (95% CI: 0.84−0.86) and pooled specificity was 0.82 (95% CI: 0.82−0.83); significantly lower than studies using non-CNN: sensitivity 0.90 (95% CI: 0.89−0.90) and specificity 0.88 (95% CI: 0.87−0.89) (p < 0.05). The diagnostic ability of radiologists and radiomics were comparable for both sensitivity (OR 0.98) and specificity (OR 0.95). Conclusions: Radiomic analysis using US provides a reproducible, reliable evaluation of undetermined thyroid nodules when compared to current best practice.

Machine intelligence in non-invasive endocrine cancer diagnostics

Artificial intelligence (AI) has illuminated a clear path towards an evolving health-care system replete with enhanced precision and computing capabilities. Medical imaging analysis can be strengthened by machine learning as the multidimensional data generated by imaging naturally lends itself to hierarchical classification. In this Review, we describe the role of machine intelligence in image-based endocrine cancer diagnostics. We first provide a brief overview of AI and consider its intuitive incorporation into the clinical workflow. We then discuss how AI can be applied for the characterization of adrenal, pancreatic, pituitary and thyroid masses in order to support clinicians in their diagnostic interpretations. This Review also puts forth a number of key evaluation criteria for machine learning in medicine that physicians can use in their appraisals of these algorithms. We identify mitigation strategies to address ongoing challenges around data availability and model interpretability in the context of endocrine cancer diagnosis. Finally, we delve into frontiers in systems integration for AI, discussing automated pipelines and evolving computing platforms that leverage distributed, decentralized and quantum techniques.