An efficient deep convolutional neural network model for visual localization and automatic diagnosis of thyroid nodules on ultrasound images

Abstract

Background:The aim of this study was to construct a deep convolutional neural network (CNN) model for localization and diagnosis of thyroid nodules on ultrasound and evaluate its diagnostic performance. Methods:We developed and trained a deep CNN model called the Brief Efficient Thyroid Network (BETNET) using 16,401 ultrasound images. According to the parameters of the model, we developed a computer-aided diagnosis (CAD) system to localize and differentiate thyroid nodules. The validation dataset (1,000 images) was used to compare the diagnostic performance of the model using three state-of-the-art algorithms. We used an internal test set (300 images) to evaluate the BETNET model by comparing it with diagnoses from five radiologists with varying degrees of experience in thyroid nodule diagnosis. Lastly, we demonstrated the general applicability of our artificial intelligence (AI) system for diagnosing thyroid cancer in an external test set (1,032 images).Results: The BETNET model accurately detected thyroid nodules in visualization experiments. The model demonstrated higher values for area under the receiver operating characteristic (AUC-ROC) curve [0.983, 95% confidence interval (CI): 0.973-0.990], sensitivity (99.19%), accuracy (98.30%), and Youden index (0.9663) than the three state-of-the-art algorithms (P<0.05). In the internal test dataset, the diagnostic accuracy of the BETNET model was 91.33%, which was markedly higher than the accuracy of one experienced (85.67%) and two less experienced radiologists (77.67% and 69.33%). The area under the ROC curve of the BETNET model (0.951) was similar to that of the two highly skilled radiologists (0.940 and 0.953) and significantly higher than that of one experienced and two less experienced radiologists (P<0.01).The kappa coefficient of the BETNET model and the pathology results showed good agreement (0.769). In addition, the BETNET model achieved an excellent diagnostic performance (AUC =0.970, 95% CI: 0.958-0.980) when applied to ultrasound images from another independent hospital. Conclusions:We developed a deep learning model which could accurately locate and automatically diagnose thyroid nodules on ultrasound images. The BETNET model exhibited better diagnostic performance than three state-of-the-art algorithms, which in turn performed similarly in diagnosis as the

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

“…Finally, the all included studies used the deep learning VGGNet model. The 11 sets of data from eight papers used the deep learning VGG-16 models (7,14,(19)(20)(21)(22)(23)25), and 6 sets of data from four papers used the deep learning VGG-19 models Moreover, the performance of the DL model is closely connected with the number of training data, and the DL model performs better when the data of the training sample are sufficiently large (36). Based on an analysis of 11 included studies, 2 sets of data from three papers did not give an explicit number of training sets, 14 sets of data from eight papers did give the number of training sets, but the amount of pre-training varied across studies and the amount of learning varied; thus, it is difficult to know the overfitting results of the model.…”

“…From then on, deep learning entered an era of rapid development and played a pivotal role in the medical field, especially in medical image recognition. Some studies used the deep learning convolutional neural network to extract ultrasound features to identify and diagnose benign and malignant thyroid nodules, and some of the studies with diagnostic performance could be comparable to or better than the advanced physicians, which could reduce unnecessary punctures and overtreatment, and help grassroots and inexperienced physicians improve diagnostic efficiency and accuracy (5)(6)(7). In addition, Lee et al (8) explored the use of deep learning convolutional neural networks in predicting the presence of lymph node metastasis in thyroid cancer on ultrasound, and their results indicated good predictive diagnostic accuracy (accuracy of 83.0%).…”

“…Characteristics of the included studies. radiologists(5,7,22). Qin et al(20) extracted both ultrasound image features and ultrasound elastic image features.…”

Rapid and automatic assessment of early gestational age using computer vision and biometric measurements based on ultrasound video

“…This study also provided 159 video datasets and we have made them publicly available (https://drive.google.com/ drive/folders/1cP25UNROveiafvumT9vInQ2OQj3xdug?usp=sharing) so that more researchers can participate in improving and clinically verifying this method. In the future, we hope to expand and improve this computing framework through deep learning, artificial intelligence, and image segmentation (27)(28)(29).…”

Deep learning based on ultrasound images assists breast lesion diagnosis in China: a multicenter diagnostic study

“…To acquire a decent performance, we chose the visual geometry group (VGG) deep CNN [33] since we observed that it does not include a global average pooling layer after the last feature extraction layer but directly flattens the feature maps and connects them with three fully connected layers sequentially, which may lead to better diagnostic performance as all extracted significant features are weighted. Indeed, VGG has demonstrated convincing classification accuracy in various medical image analysis tasks [17,[34][35][36][37], including breast mass classification [32,38,39]. The VGG-19 architecture was adopted as the backbone for breast cancer diagnosis modeling in this study.…”