Deep convolutional neural network for classification of thyroid nodules on ultrasound: comparison of the diagnostic performance with that of radiologists

From Bench-to-Bedside: How Artificial Intelligence is Changing Thyroid Nodule Diagnostics, a Systematic Review

CONTEXT

Use of artificial intelligence (AI) to predict clinical outcomes in thyroid nodule diagnostics has grown exponentially over the past decade. The greatest challenge is in understanding the best model to apply to one's own patient population, and how to operationalize such a model in practice.

EVIDENCE ACQUISITION

A literature search of PubMed and IEEE Xplore was conducted for English-language publications between January 1, 2015 and January 1, 2023, studying diagnostic tests on suspected thyroid nodules that used AI. We excluded articles without prospective or external validation, nonprimary literature, duplicates, focused on nonnodular thyroid conditions, not using AI, and those incidentally using AI in support of an experimental diagnostic outside standard clinical practice. Quality was graded by Oxford level of evidence.

EVIDENCE SYNTHESIS

A total of 61 studies were identified; all performed external validation, 16 studies were prospective, and 33 compared a model to physician prediction of ground truth. Statistical validation was reported in 50 papers. A diagnostic pipeline was abstracted, yielding 5 high-level outcomes: (1) nodule localization, (2) ultrasound (US) risk score, (3) molecular status, (4) malignancy, and (5) long-term prognosis. Seven prospective studies validated a single commercial AI; strengths included automating nodule feature assessment from US and assisting the physician in predicting malignancy risk, while weaknesses included automated margin prediction and interobserver variability.

CONCLUSION

Models predominantly used US images to predict malignancy. Of 4 Food and Drug Administration-approved products, only S-Detect was extensively validated. Implementing an AI model locally requires data sanitization and revalidation to ensure appropriate clinical performance.

A multi-task model for reliable classification of thyroid nodules in ultrasound images

Thyroid nodules are common, and patients with potential malignant lesions are usually diagnosed using ultrasound imaging to determine further treatment options. This study aims to propose a computer-aided diagnosis method for benign and malignant classification of thyroid nodules in ultrasound images. We propose a novel multi-task framework that combines the advantages of dense connectivity, Squeeze-and-Excitation (SE) connectivity, and Atrous Spatial Pyramid Pooling (ASPP) layer to enhance feature extraction. The Dense connectivity is used to optimize feature reuse, the SE connectivity to optimize feature weights, the ASPP layer to fuse feature information, and a multi-task learning framework to adjust the attention of the network. We evaluate our model using a 10-fold cross-validation approach based on our established Thyroid dataset. We assess the performance of our method using six average metrics: accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and AUC, which are 93.49, 95.54, 91.52, 91.63, 95.47, and 96.84%, respectively. Our proposed method outperforms other classification networks in all metrics, achieving optimal performance. We propose a multi-task model, DSMA-Net, for distinguishing thyroid nodules in ultrasound images. This method can further enhance the diagnostic ability of doctors for suspected cancer patients and holds promise for clinical applications.

Learnability of Thyroid Nodule Assessment on Ultrasonography: Using a Big Data Set

OBJECTIVE

The aims of the work described here were to evaluate the learnability of thyroid nodule assessment on ultrasonography (US) using a big data set of US images and to evaluate the diagnostic utilities of artificial intelligence computer-aided diagnosis (AI-CAD) used by readers with varying experience to differentiate benign and malignant thyroid nodules.

METHODS

Six college freshmen independently studied the "learning set" composed of images of 13,560 thyroid nodules, and their diagnostic performance was evaluated after their daily learning sessions using the "test set" composed of images of 282 thyroid nodules. The diagnostic performance of two residents and an experienced radiologist was evaluated using the same "test set." After an initial diagnosis, all readers once again evaluated the "test set" with the assistance of AI-CAD.

RESULTS

Diagnostic performance of almost all students increased after the learning program. Although the mean areas under the receiver operating characteristic curves (AUROCs) of residents and the experienced radiologist were significantly higher than those of students, the AUROCs of five of the six students did not differ significantly compared with that of the one resident. With the assistance of AI-CAD, sensitivity significantly increased in three students, specificity in one student, accuracy in four students and AUROC in four students. Diagnostic performance of the two residents and the experienced radiologist was better with the assistance of AI-CAD.

CONCLUSION

A self-learning method using a big data set of US images has potential as an ancillary tool alongside traditional training methods. With the assistance of AI-CAD, the diagnostic performance of readers with varying experience in thyroid imaging could be further improved.

Diagnosis of Salivary Gland Tumors Using Transfer Learning with Fine-Tuning and Gradual Unfreezing

Ultrasound is the primary tool for evaluating salivary gland tumors (SGTs); however, tumor diagnosis currently relies on subjective features. This study aimed to establish an objective ultrasound diagnostic method using deep learning. We collected 446 benign and 223 malignant SGT ultrasound images in the training/validation set and 119 benign and 44 malignant SGT ultrasound images in the testing set. We trained convolutional neural network (CNN) models from scratch and employed transfer learning (TL) with fine-tuning and gradual unfreezing to classify malignant and benign SGTs. The diagnostic performances of these models were compared. By utilizing the pretrained ResNet50V2 with fine-tuning and gradual unfreezing, we achieved a 5-fold average validation accuracy of 0.920. The diagnostic performance on the testing set demonstrated an accuracy of 89.0%, a sensitivity of 81.8%, a specificity of 91.6%, a positive predictive value of 78.3%, and a negative predictive value of 93.2%. This performance surpasses that of other models in our study. The corresponding Grad-CAM visualizations were also presented to provide explanations for the diagnosis. This study presents an effective and objective ultrasound method for distinguishing between malignant and benign SGTs, which could assist in preoperative evaluation.

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.

Deep learning for the diagnosis of suspicious thyroid nodules based on multimodal ultrasound images

OBJECTIVES

This study aimed to differentially diagnose thyroid nodules (TNs) of Thyroid Imaging Reporting and Data System (TI-RADS) 3-5 categories using a deep learning (DL) model based on multimodal ultrasound (US) images and explore its auxiliary role for radiologists with varying degrees of experience.

METHODS

Preoperative multimodal US images of 1,138 TNs of TI-RADS 3-5 categories were randomly divided into a training set (n = 728), a validation set (n = 182), and a test set (n = 228) in a 4:1:1.25 ratio. Grayscale US (GSU), color Doppler flow imaging (CDFI), strain elastography (SE), and region of interest mask (Mask) images were acquired in both transverse and longitudinal sections, all of which were confirmed by pathology. In this study, fivefold cross-validation was used to evaluate the performance of the proposed DL model. The diagnostic performance of the mature DL model and radiologists in the test set was compared, and whether DL could assist radiologists in improving diagnostic performance was verified. Specificity, sensitivity, accuracy, positive predictive value, negative predictive value, and area under the receiver operating characteristics curves (AUC) were obtained.

RESULTS

The AUCs of DL in the differentiation of TNs were 0.858 based on (GSU + SE), 0.909 based on (GSU + CDFI), 0.906 based on (GSU + CDFI + SE), and 0.881 based (GSU + Mask), which were superior to that of 0.825-based single GSU (p = 0.014, p< 0.001, p< 0.001, and p = 0.002, respectively). The highest AUC of 0.928 was achieved by DL based on (G + C + E + M)US, the highest specificity of 89.5% was achieved by (G + C + E)US, and the highest accuracy of 86.2% and sensitivity of 86.9% were achieved by DL based on (G + C + M)US. With DL assistance, the AUC of junior radiologists increased from 0.720 to 0.796 (p< 0.001), which was slightly higher than that of senior radiologists without DL assistance (0.796 vs. 0.794, p > 0.05). Senior radiologists with DL assistance exhibited higher accuracy and comparable AUC than that of DL based on GSU (83.4% vs. 78.9%, p = 0.041; 0.822 vs. 0.825, p = 0.512). However, the AUC of DL based on multimodal US images was significantly higher than that based on visual diagnosis by radiologists (p< 0.05).

CONCLUSION

The DL models based on multimodal US images showed exceptional performance in the differential diagnosis of suspicious TNs, effectively increased the diagnostic efficacy of TN evaluations by junior radiologists, and provided an objective assessment for the clinical and surgical management phases that follow.

Ultrasound-based deep learning using the VGGNet model for the differentiation of benign and malignant thyroid nodules: A meta-analysis

Objective

The aim of this study was to evaluate the accuracy of deep learning using the convolutional neural network VGGNet model in distinguishing benign and malignant thyroid nodules based on ultrasound images.

Methods

Relevant studies were selected from PubMed, Embase, Cochrane Library, China National Knowledge Infrastructure (CNKI), and Wanfang databases, which used the deep learning-related convolutional neural network VGGNet model to classify benign and malignant thyroid nodules based on ultrasound images. Cytology and pathology were used as gold standards. Furthermore, reported eligibility and risk bias were assessed using the QUADAS-2 tool, and the diagnostic accuracy of deep learning VGGNet was analyzed with pooled sensitivity, pooled specificity, diagnostic odds ratio, and the area under the curve.

Results

A total of 11 studies were included in this meta-analysis. The overall estimates of sensitivity and specificity were 0.87 [95% CI (0.83, 0.91)] and 0.85 [95% CI (0.79, 0.90)], respectively. The diagnostic odds ratio was 38.79 [95% CI (22.49, 66.91)]. The area under the curve was 0.93 [95% CI (0.90, 0.95)]. No obvious publication bias was found.

Conclusion

Deep learning using the convolutional neural network VGGNet model based on ultrasound images performed good diagnostic efficacy in distinguishing benign and malignant thyroid nodules.

Systematic Review Registration

https://www.crd.york.ac.nk/prospero, identifier CRD42022336701.

Machine learning for identifying benign and malignant of thyroid tumors: A retrospective study of 2,423 patients

Thyroid tumors, one of the common tumors in the endocrine system, while the discrimination between benign and malignant thyroid tumors remains insufficient. The aim of this study is to construct a diagnostic model of benign and malignant thyroid tumors, in order to provide an emerging auxiliary diagnostic method for patients with thyroid tumors. The patients were selected from the Chongqing General Hospital (Chongqing, China) from July 2020 to September 2021. And peripheral blood, BRAFV600E gene, and demographic indicators were selected, including sex, age, BRAFV600E gene, lymphocyte count (Lymph#), neutrophil count (Neu#), neutrophil/lymphocyte ratio (NLR), platelet/lymphocyte ratio (PLR), red blood cell distribution width (RDW), platelets count (PLT), red blood cell distribution width-coefficient of variation (RDW-CV), alkaline phosphatase (ALP), and parathyroid hormone (PTH). First, feature selection was executed by univariate analysis combined with least absolute shrinkage and selection operator (LASSO) analysis. Afterward, we used machine learning algorithms to establish three types of models. The first model contains all predictors, the second model contains indicators after feature selection, and the third model contains patient peripheral blood indicators. The four machine learning algorithms include extreme gradient boosting (XGBoost), random forest (RF), light gradient boosting machine (LightGBM), and adaptive boosting (AdaBoost) which were used to build predictive models. A grid search algorithm was used to find the optimal parameters of the machine learning algorithms. A series of indicators, such as the area under the curve (AUC), were intended to determine the model performance. A total of 2,042 patients met the criteria and were enrolled in this study, and 12 variables were included. Sex, age, Lymph#, PLR, RDW, and BRAFV600E were identified as statistically significant indicators by univariate and LASSO analysis. Among the model we constructed, RF, XGBoost, LightGBM and AdaBoost with the AUC of 0.874 (95% CI, 0.841-0.906), 0.868 (95% CI, 0.834-0.901), 0.861 (95% CI, 0.826-0.895), and 0.837 (95% CI, 0.802-0.873) in the first model. With the AUC of 0.853 (95% CI, 0.818-0.888), 0.853 (95% CI, 0.818-0.889), 0.837 (95% CI, 0.800-0.873), and 0.832 (95% CI, 0.797-0.867) in the second model. With the AUC of 0.698 (95% CI, 0.651-0.745), 0.688 (95% CI, 0.639-0.736), 0.693 (95% CI, 0.645-0.741), and 0.666 (95% CI, 0.618-0.714) in the third model. Compared with the existing models, our study proposes a model incorporating novel biomarkers which could be a powerful and promising tool for predicting benign and malignant thyroid tumors.