Detection of Thyroid Nodules with Ultrasound Images Based on Deep Learning

Author(s):  
Xia Yu ◽  
Hongjie Wang ◽  
Liyong Ma

Background: Thyroid nodules are a common clinical entity with high incidence. Ultrasound is often employed to detect and evaluate thyroid nodules. The development of an efficient automated method to detect thyroid nodules using ultrasound has the potential to reduce both physician workload and operator-dependence. Objective: To study the method of automatic detection of thyroid nodules based on deep learning using ultrasound, and to obtain the detection method with higher accuracy and better performance. Methods: A total of 1200 ultrasound images of thyroid nodules and 800 ultrasound thyroid images without nodule are collected. An improved faster R-CNN based detection method of thyroid nodule is proposed. Instead of using VGG16 as the backbone, ResNet is employed as the backbone for faster R-CNN. SVM, CNN and Faster-RCNN methods are used for thyroid nodule detection test. Precision, sensitivity, specificity and F1-score indicators are used to evaluate the detection performance of different methods. Results: The method based on deep learning is superior to that based on SVM. Faster R-CNN method and the improved method are better than CNN method. Compared with VGG16 as the backbone, RestNet101 backbone based faster R-CNN method achieves better thyroid detection effect. From the accuracy index, the proposed method is 0.084, 0.032 and 0.019 higher than SVM, CNN and faster R-CNN, respectively. Similar results can be seen in precision, sensitivity, specificity and F1-Score indicators. Conclusion: The proposed method of deep learning achieves the best performance values with the highest true positive and true negative detection compared to other methods and performs best in the detection of thyroid nodules.

2020 ◽  
pp. 1-16
Author(s):  
Ling Zhang ◽  
Yan Zhuang ◽  
Zhan Hua ◽  
Lin Han ◽  
Cheng Li ◽  
...  

BACKGROUND: Thyroid ultrasonography is widely used to diagnose thyroid nodules in clinics. Automatic localization of nodules can promote the development of intelligent thyroid diagnosis and reduce workload of radiologists. However, besides the ultrasound image has low contrast and high noise, the thyroid nodules are diverse in shape and vary greatly in size. Thus, thyroid nodule detection in ultrasound images is still a challenging task. OBJECTIVE: This study proposes an automatic detection algorithm to locate nodules in B ultrasound images and Doppler ultrasound images. This method can be used to screen thyroid nodules and provide a basis for subsequent automatic segmentation and intelligent diagnosis. METHODS: We develop and optimize an improved YOLOV3 model for detecting thyroid nodules in ultrasound images with B-mode and Doppler mode. Improvements include (1) using the high-resolution network (HRNet) as the basic network for gradually extracting high-level semantic features to reduce the missed detection and misdetection, (2) optimizing the loss function for single target detection like nodules, and (3) obtaining the anchor boxes by clustering the candidate frames of real nodules in the dataset. RESULTS: The experimental results of applying to 8000 clinical ultrasound images show that the new method developed and tested in this study can effectively detect thyroid nodules. The method achieves 94.53% mean precision and 95.00% mean recall. CONCLUTIONS: The study demonstrates a new automated method that enables to achieve high detection accuracy and effectively locate thyroid nodules in various ultrasound images without any user interaction, which indicates its potential clinical application value for the thyroid nodule screening.


Ultrasound scanning is most excellent significant diagnosis techniques utilized for thyroid nodules identification. A thyroid nodule is unnecessary cells that can develop in your base of neck which can be normal or cancerous. Many Computer added diagnosis systems (CAD) have been developed as a second opinion for radiologist. The thyroid nodules classification using machine learning and deep learning approach is latest trend which is using to improve accuracy for differentiation of thyroid nodules from benign and malignant type. In this paper we review the most recent work on CAD system which uses different feature extraction technique and classifier used for thyroid nodules classification with deep learning approach. This paper we illustrate the result obtained by these studies and highlight the limitation of each proposed methods. Moreover we summarize convolution neural network (CNN) architecture for classification of thyroid nodule. This literature review is meant at researcher but it also useful for radiologist who is interesting in CAD tool in ultrasound imaging for second opinion.


2020 ◽  
Vol 185 ◽  
pp. 03021
Author(s):  
Meng Zhou ◽  
Rui Wang ◽  
Peng Fu ◽  
Yang Bai ◽  
Ligang Cui

As the most common malignancy in the endocrine system, thyroid cancer is usually diagnosed by discriminating the malignant nodules from the benign ones using ultrasonography, whose interpretation results primarily depends on the subjectivity judgement of the radiologists. In this study, we propose a novel cascade deep learning model to achieve automatic objective diagnose during ultrasound examination for assisting radiologists in recognizing benign and malignant thyroid nodules. First, the simplified U-net is employed to segment the region of interesting (ROI) of the thyroid nodules in each frame of the ultrasound image automatically. Then, to alleviate the limitation that medical training data are relatively small in size, the improved Conditional Variational Auto-Encoder (CVAE) learning the probability distribution of ROI images is trained to generate new images for data augmentation. Finally, ResNet50 is trained with both original and generated ROI images. As consequence, the deep learning model formed by the trained U-net and trained Resnet-50 cascade can achieve malignant thyroid nodule recognition with the accuracy of 87.4%, the sensitivity of 92%, and the specificity of 86.8%.


Circulation ◽  
2020 ◽  
Vol 142 (Suppl_3) ◽  
Author(s):  
Juhwan Lee ◽  
Yazan Gharaibeh ◽  
Vladislav N Zimin ◽  
Luis A Dallan ◽  
Hiram G Bezerra ◽  
...  

Introduction: Major calcifications are of great concern when performing percutaneous coronary intervention as they hinder stent deployment. Calcifications can lead to under-expansion and strut malapposition, with increased risk of thrombosis and in-stent restenosis. Therefore, accurate identification, visualization, and quantification of calcifications are important. Objective: In this study, we developed a 2-step deep learning approach to enable segmentation of major calcifications in a typical 500+ frame intravascular optical coherence tomography (IVOCT) images. Methods: The dataset consisted of a total of 12,551 IVOCT frames across 68 patients with 68 pullbacks. We applied a series of pre-processing steps including guidewire/shadow removal, lumen detection, pixel shifting, and Gaussian filtering. To detect the major calcifications in step 1, we implemented the 3D convolutional neural network consisting of 5 convolutional, 5 max-pooling, and 2 fully-connected layers. In step-2, SegNet deep learning model was used to segment calcified plaques. In both steps, classification errors were reduced using conditional random field. Results: Step-1 reliably identified major calcifications (sensitivity/specificity: 97.7%/87.7%). Semantic segmentation of calcifications following step-2 was typically visually quite good (Fig. 1) with (sensitivity/specificity: 86.2%/96.7%). Our method was superior to a single step approach and showed excellent reproducibility on repetitive IVOCT pullbacks, with very small differences of clinically relevant attributes (maximum angle, maximum thickness, and length) and the exact same IVOCT calcium scores for assessment of stent deployment. Conclusions: We developed the fully-automated method for identifying calcifications in IVOCT images based on a 2-step deep learning approach. Extensive analyses indicate that our method is very informative for both live-time treatment planning and research purposes.


2021 ◽  
Author(s):  
Zijian Zhao ◽  
Congmin Yang ◽  
Qian Wang ◽  
Huawei Zhang ◽  
Linlin Shi ◽  
...  

2020 ◽  
Vol 28 (6) ◽  
pp. 1123-1139
Author(s):  
Liqun Zhang ◽  
Ke Chen ◽  
Lin Han ◽  
Yan Zhuang ◽  
Zhan Hua ◽  
...  

BACKGROUND: Calcification is an important criterion for classification between benign and malignant thyroid nodules. Deep learning provides an important means for automatic calcification recognition, but it is tedious to annotate pixel-level labels for calcifications with various morphologies. OBJECTIVE: This study aims to improve accuracy of calcification recognition and prediction of its location, as well as to reduce the number of pixel-level labels in model training. METHODS: We proposed a collaborative supervision network based on attention gating (CS-AGnet), which was composed of two branches: a segmentation network and a classification network. The reorganized two-stage collaborative semi-supervised model was trained under the supervision of all image-level labels and few pixel-level labels. RESULTS: The results show that although our semi-supervised network used only 30% (289 cases) of pixel-level labels for training, the accuracy of calcification recognition reaches 92.1%, which is very close to 92.9% of deep supervision with 100% (966 cases) pixel-level labels. The CS-AGnet enables to focus the model’s attention on calcification objects. Thus, it achieves higher accuracy than other deep learning methods. CONCLUSIONS: Our collaborative semi-supervised model has a preferable performance in calcification recognition, and it reduces the number of manual annotations of pixel-level labels. Moreover, it may be of great reference for the object recognition of medical dataset with few labels.


2021 ◽  
Vol 8 (4) ◽  
pp. 1264
Author(s):  
Snigdha Kamini ◽  
Jainendra K. Arora ◽  
Sunil Kumar Jain

Background: Thyroid nodules are a common endocrine disease whose prevalence in India is approximately 12.2%. Although most patients with suspected nodules have benign conditions, the overestimation of malignancy leads to the performance of unnecessary procedures. No clinical, radiological and cytological parameters has singularly shown significant impact on clinical practice and post-operative histopathological examination remains the gold standard in the diagnosis of malignancy.Methods: 55 patients with thyroid nodules were evaluated and the Clinical assessment findings were recorded by McGill thyroid nodule score, ultrasonography findings using TIRADS and FNAC findings by the Bethesda system. The triple test was then used to classify them and these results were compared with the HPE of the post-operative specimen.Results: The sensitivity and specificity of TIRADS, FNAC were higher as compared to clinical score; clinical score had lowest sensitivity of 72.73%. The sensitivity, specificity, PPV, NPV and accuracy of triple test was 100%. Triple test had higher sensitivity, specificity and accuracy in differentiating thyroid nodules as compared to any of the three parameters used individually.Conclusions: Triple test has higher accuracy, sensitivity and specificity in determining the nature of thyroid nodule than each of the parameters used individually and it is especially useful in follicular lesions. On the basis of the results of this study, we conclude that the triple test can reliably be used to differentiate benign and malignant nodules preoperatively.


2021 ◽  
Author(s):  
Johnson Thomas ◽  
Tracy Haertling

AbstractBackgroundCurrent classification systems for thyroid nodules are very subjective. Artificial intelligence (AI) algorithms have been used to decrease subjectivity in medical image interpretation. 1 out of 2 women over the age of 50 may have a thyroid nodule and at present the only way to exclude malignancy is through invasive procedures. Hence, there exists a need for noninvasive objective classification of thyroid nodules. Some cancers have benign appearance on ultrasonogram. Hence, we decided to create an image similarity algorithm rather than image classification algorithm.MethodsUltrasound images of thyroid nodules from patients who underwent either biopsy or thyroid surgery from February of 2012 through February of 2017 in our institution were used to create AI models. Nodules were excluded if there was no definitive diagnosis of benignity or malignancy. 482 nodules met the inclusion criteria and all available images from these nodules were used to create the AI models. Later, these AI models were used to test 103 thyroid nodules which underwent biopsy or surgery from March of 2017 through July of 2018.ResultsNegative predictive value of the image similarity model was 93.2%. Sensitivity, specificity, positive predictive value and accuracy of the model was 87.8%, 78.5%, 65.9% and 81.5% respectively.ConclusionWhen compared to published results of ACR TIRADS and ATA classification system, our image similarity model had comparable negative predictive value with better sensitivity specificity and positive predictive value. By using image similarity AI models, we can eliminate subjectivity and decrease the number of unnecessary biopsies. Using image similarity AI model, we were able to create an explainable AI model which increases physician’s confidence in the predictions.


Sign in / Sign up

Export Citation Format

Share Document