Automatic Deep Learning Semantic Segmentation of Ultrasound Thyroid Cineclips Using Recurrent Fully Convolutional Networks

TransEBUS: The interpretation of endobronchial ultrasound image using hybrid transformer for differentiating malignant and benign mediastinal lesions

The purpose of this study is to establish a deep learning automatic assistance diagnosis system for benign and malignant classification of mediastinal lesions in endobronchial ultrasound (EBUS) images. EBUS images are in the form of video and contain multiple imaging modes. Different imaging modes and different frames can reflect the different characteristics of lesions. Compared with previous studies, the proposed model can efficiently extract and integrate the spatiotemporal relationships between different modes and does not require manual selection of representative frames. In recent years, Vision Transformer has received much attention in the field of computer vision. Combined with convolutional neural networks, hybrid transformers can also perform well on small datasets. This study designed a novel deep learning architecture based on hybrid transformer called TransEBUS. By adding learnable parameters in the temporal dimension, TransEBUS was able to extract spatiotemporal features from insufficient data. In addition, we designed a two-stream module to integrate information from three different imaging modes of EBUS. Furthermore, we applied contrastive learning when training TransEBUS, enabling it to learn discriminative representation of benign and malignant mediastinal lesions. The results show that TransEBUS achieved a diagnostic accuracy of 82% and an area under the curve of 0.8812 in the test dataset, outperforming other methods. It also shows that several models can improve performance by incorporating two-stream module. Our proposed system has shown its potential to help physicians distinguishing benign and malignant mediastinal lesions, thereby ensuring the accuracy of EBUS examination.

PlaqueNet: deep learning enabled coronary artery plaque segmentation from coronary computed tomography angiography

Cardiovascular disease, primarily caused by atherosclerotic plaque formation, is a significant health concern. The early detection of these plaques is crucial for targeted therapies and reducing the risk of cardiovascular diseases. This study presents PlaqueNet, a solution for segmenting coronary artery plaques from coronary computed tomography angiography (CCTA) images. For feature extraction, the advanced residual net module was utilized, which integrates a deepwise residual optimization module into network branches, enhances feature extraction capabilities, avoiding information loss, and addresses gradient issues during training. To improve segmentation accuracy, a depthwise atrous spatial pyramid pooling based on bicubic efficient channel attention (DASPP-BICECA) module is introduced. The BICECA component amplifies the local feature sensitivity, whereas the DASPP component expands the network's information-gathering scope, resulting in elevated segmentation accuracy. Additionally, BINet, a module for joint network loss evaluation, is proposed. It optimizes the segmentation model without affecting the segmentation results. When combined with the DASPP-BICECA module, BINet enhances overall efficiency. The CCTA segmentation algorithm proposed in this study outperformed the other three comparative algorithms, achieving an intersection over Union of 87.37%, Dice of 93.26%, accuracy of 93.12%, mean intersection over Union of 93.68%, mean Dice of 96.63%, and mean pixel accuracy value of 96.55%.

Three-dimensional visualization of thyroid ultrasound images based on multi-scale features fusion and hierarchical attention

BACKGROUND

Ultrasound three-dimensional visualization, a cutting-edge technology in medical imaging, enhances diagnostic accuracy by providing a more comprehensive and readable portrayal of anatomical structures compared to traditional two-dimensional ultrasound. Crucial to this visualization is the segmentation of multiple targets. However, challenges like noise interference, inaccurate boundaries, and difficulties in segmenting small structures exist in the multi-target segmentation of ultrasound images. This study, using neck ultrasound images, concentrates on researching multi-target segmentation methods for the thyroid and surrounding tissues.

METHOD

We improved the Unet++ to propose PA-Unet++ to enhance the multi-target segmentation accuracy of the thyroid and its surrounding tissues by addressing ultrasound noise interference. This involves integrating multi-scale feature information using a pyramid pooling module to facilitate segmentation of structures of various sizes. Additionally, an attention gate mechanism is applied to each decoding layer to progressively highlight target tissues and suppress the impact of background pixels.

RESULTS

Video data obtained from 2D ultrasound thyroid serial scans served as the dataset for this paper.4600 images containing 23,000 annotated regions were divided into training and test sets at a ratio of 9:1, the results showed that: compared with the results of U-net++, the Dice of our model increased from 78.78% to 81.88% (+ 3.10%), the mIOU increased from 73.44% to 80.35% (+ 6.91%), and the PA index increased from 92.95% to 94.79% (+ 1.84%).

CONCLUSIONS

Accurate segmentation is fundamental for various clinical applications, including disease diagnosis, treatment planning, and monitoring. This study will have a positive impact on the improvement of 3D visualization capabilities and clinical decision-making and research in the context of ultrasound image.

[Standardized acquisition and documentation of cine loops on conventional thyroid ultrasound]

Ultrasound is the basic imaging method for the assessment of the thyroid gland. Due to the high prevalence of structural disease, the examination procedure is used very frequently in Germany, in many cases in the context of follow-up. The assessment of thyroid pathologies and their dynamics is subjected to relevant inter- and intraobserver variability. Findings that were not identified during live ultrasound cannot be assessed retrospectively. Applying an SOP for the acquisition and documentation of standardized video sequences of ultrasound images (so-called cine loops), allows for a secondary retrospective evaluation of the thyroid gland, taking into account previously acquired images analogous to other cross-sectional imaging methods such as CT or MRI. The cine loops can be acquired by non-physician personnel, stored to the local PACS and used for educational and research purposes.

[Review on ultrasonographic diagnosis of thyroid diseases based on deep learning]

In recent years, the incidence of thyroid diseases has increased significantly and ultrasound examination is the first choice for the diagnosis of thyroid diseases. At the same time, the level of medical image analysis based on deep learning has been rapidly improved. Ultrasonic image analysis has made a series of milestone breakthroughs, and deep learning algorithms have shown strong performance in the field of medical image segmentation and classification. This article first elaborates on the application of deep learning algorithms in thyroid ultrasound image segmentation, feature extraction, and classification differentiation. Secondly, it summarizes the algorithms for deep learning processing multimodal ultrasound images. Finally, it points out the problems in thyroid ultrasound image diagnosis at the current stage and looks forward to future development directions. This study can promote the application of deep learning in clinical ultrasound image diagnosis of thyroid, and provide reference for doctors to diagnose thyroid disease.

Objective assessment of segmentation models for thyroid ultrasound images

Ultrasound features related to thyroid lesions structure, shape, volume, and margins are considered to determine cancer risk. Automatic segmentation of the thyroid lesion would allow the sonographic features to be estimated. On the basis of clinical ultrasonography B-mode scans, a multi-output CNN-based semantic segmentation is used to separate thyroid nodules' cystic & solid components. Semantic segmentation is an automatic technique that labels the ultrasound (US) pixels with an appropriate class or pixel category, i.e., belongs to a lesion or background. In the present study, encoder-decoder-based semantic segmentation models i.e. SegNet using VGG16, UNet, and Hybrid-UNet implemented for segmentation of thyroid US images. For this work, 820 thyroid US images are collected from the DDTI and ultrasoundcases.info (USC) datasets. These segmentation models were trained using a transfer learning approach with original and despeckled thyroid US images. The performance of segmentation models is evaluated by analyzing the overlap region between the true contour lesion marked by the radiologist and the lesion retrieved by the segmentation model. The mean intersection of union (mIoU), mean dice coefficient (mDC) metrics, TPR, TNR, FPR, and FNR metrics are used to measure performance. Based on the exhaustive experiments and performance evaluation parameters it is observed that the proposed Hybrid-UNet segmentation model segments thyroid nodules and cystic components effectively.

A devised thyroid segmentation with multi-stage modification based on Super-pixel U-Net under insufficient data

OBJECTIVES

The application of deep learning to medical image segmentation has received considerable attention. Nevertheless, when segmenting thyroid ultrasound images, it is difficult to achieve good segmentation results using deep learning methods because of the large number of nonthyroidal regions and insufficient training data.

METHODS

In this study, a Super-pixel U-Net, designed by adding a supplementary path to U-Net, was devised to boost the segmentation results of thyroids. The improved network can introduce more information into the network, boosting auxiliary segmentation results. A multi-stage modification is introduced in this method, which includes boundary segmentation, boundary repair, and auxiliary segmentation. To reduce the negative effects of non-thyroid regions in the segmentation, U-Net was utilized to obtain rough boundary outputs. Subsequently, another U-Net is trained to improve and repair the coverage of the boundary outputs. Super-pixel U-Net was applied in the third stage to assist in the segmentation of the thyroid more precisely. Finally, multidimensional indicators were used to compare the segmentation results of the proposed method with those of other comparison experiments.

DISCUSSION

The proposed method achieved an F1 Score of 0.9161 and an IoU of 0.9279. Furthermore, the proposed method also exhibits better performance in terms of shape similarity, with an average convexity of 0.9395. an average ratio of 0.9109, an average compactness of 0.8976, an average eccentricity of 0.9448, and an average rectangularity of 0.9289. The average area estimation indicator was 0.8857.

CONCLUSION

The proposed method exhibited superior performance, proving the improvements of the multi-stage modification and Super-pixel U-Net.

Assessment of encoder-decoder-based segmentation models for thyroid ultrasound images

Encoder-decoder-based semantic segmentation models classify image pixels into the corresponding class, such as the ROI (region of interest) or background. In the present study, simple / dilated convolution / series / directed acyclic graph (DAG)-based encoder-decoder semantic segmentation models have been implemented, i.e., SegNet (VGG16), SegNet (VGG19), U-Net, mobileNetv2, ResNet18, ResNet50, Xception and Inception networks for the segment TTUS(Thyroid Tumor Ultrasound) images. Transfer learning has been used to train these segmentation networks using original and despeckled TTUS images. The performance of the networks has been calculated using mIoU and mDC metrics. Based on the exhaustive experiments, it has been observed that ResNet50-based segmentation model obtained the best results objectively with values 0.87 for mIoU, 0.94 for mDC, and also according to radiologist opinion on shape, margin, and echogenicity characteristics of segmented lesions. It is noted that the segmentation model, namely ResNet50, provides better segmentation based on objective and subjective assessment. It may be used in the healthcare system to identify thyroid nodules accurately in real time.

Deep convolutional neural networks for multiple histologic types of ovarian tumors classification in ultrasound images

Objective

This study aimed to evaluate and validate the performance of deep convolutional neural networks when discriminating different histologic types of ovarian tumor in ultrasound (US) images.

Material and methods

Our retrospective study took 1142 US images from 328 patients from January 2019 to June 2021. Two tasks were proposed based on US images. Task 1 was to classify benign and high-grade serous carcinoma in original ovarian tumor US images, in which benign ovarian tumor was divided into six classes: mature cystic teratoma, endometriotic cyst, serous cystadenoma, granulosa-theca cell tumor, mucinous cystadenoma and simple cyst. The US images in task 2 were segmented. Deep convolutional neural networks (DCNN) were applied to classify different types of ovarian tumors in detail. We used transfer learning on six pre-trained DCNNs: VGG16, GoogleNet, ResNet34, ResNext50, DensNet121 and DensNet201. Several metrics were adopted to assess the model performance: accuracy, sensitivity, specificity, FI-score and the area under the receiver operating characteristic curve (AUC).

Results

The DCNN performed better in labeled US images than in original US images. The best predictive performance came from the ResNext50 model. The model had an overall accuracy of 0.952 for in directly classifying the seven histologic types of ovarian tumors. It achieved a sensitivity of 90% and a specificity of 99.2% for high-grade serous carcinoma, and a sensitivity of over 90% and a specificity of over 95% in most benign pathological categories.

Conclusion

DCNN is a promising technique for classifying different histologic types of ovarian tumors in US images, and provide valuable computer-aided information.

Segmentation of thyroid glands and nodules in ultrasound images using the improved U-Net architecture

BACKGROUND

Identifying thyroid nodules' boundaries is crucial for making an accurate clinical assessment. However, manual segmentation is time-consuming. This paper utilized U-Net and its improved methods to automatically segment thyroid nodules and glands.

METHODS

The 5822 ultrasound images used in the experiment came from two centers, 4658 images were used as the training dataset, and 1164 images were used as the independent mixed test dataset finally. Based on U-Net, deformable-pyramid split-attention residual U-Net (DSRU-Net) by introducing ResNeSt block, atrous spatial pyramid pooling, and deformable convolution v3 was proposed. This method combined context information and extracts features of interest better, and had advantages in segmenting nodules and glands of different shapes and sizes.

RESULTS

DSRU-Net obtained 85.8% mean Intersection over Union, 92.5% mean dice coefficient and 94.1% nodule dice coefficient, which were increased by 1.8%, 1.3% and 1.9% compared with U-Net.

CONCLUSIONS

Our method is more capable of identifying and segmenting glands and nodules than the original method, as shown by the results of correlational studies.

Hybrid transformer UNet for thyroid segmentation from ultrasound scans

Deep learning based medical image segmentation methods have been widely used for thyroid gland segmentation from ultrasound images, which is of great importance for the diagnosis of thyroid disease since it can provide various valuable sonography features. However, existing thyroid gland segmentation models suffer from: (1) low-level features that are significant in depicting thyroid boundaries are gradually lost during the feature encoding process, (2) contextual features reflecting the changes of difference between thyroid and other anatomies in the ultrasound diagnosis process are either omitted by 2D convolutions or weakly represented by 3D convolutions due to high redundancy. In this work, we propose a novel hybrid transformer UNet (H-TUNet) to segment thyroid glands in ultrasound sequences, which consists of two parts: (1) a 2D Transformer UNet is proposed by utilizing a designed multi-scale cross-attention transformer (MSCAT) module on every skipped connection of the UNet, so that the low-level features from different encoding layers are integrated and refined according to the high-level features in the decoding scheme, leading to better representation of differences between anatomies in one ultrasound frame; (2) a 3D Transformer UNet is proposed by applying a 3D self-attention transformer (SAT) module to the very bottom layer of 3D UNet, so that the contextual features representing visual differences between regions and consistencies within regions could be strengthened from successive frames in the video. The learning process of the H-TUNet is formulated as a unified end-to-end network, so the intra-frame feature extraction and inter-frame feature aggregation can be learned and optimized jointly. The proposed method was evaluated on Thyroid Segmentation in Ultrasonography Dataset (TSUD) and TG3k Dataset. Experimental results have demonstrated that our method outperformed other state-of-the-art methods with respect to the certain benchmarks for thyroid gland segmentation.

Quantifying Valve Regurgitation Using 3-D Doppler Ultrasound Images and Deep Learning

Accurate quantification of cardiac valve regurgitation jets is fundamental for guiding treatment. Cardiac ultrasound is the preferred diagnostic tool, but current methods for measuring the regurgitant volume (RVol) are limited by low accuracy and high interobserver variability. Following recent research, quantitative estimators of orifice size and RVol based on high frame rate 3-D ultrasound have been proposed, but measurement accuracy is limited by the wide point spread function (PSF) relative to the orifice size. The aim of this article was to investigate the use of deep learning to estimate both the orifice size and the RVol. A simulation model was developed to simulate the power-Doppler images of blood flow through orifices with different geometries. A convolutional neural network (CNN) was trained on 30 000 image pairs. The network was used to reconstruct orifices from power-Doppler data, which facilitated estimators for regurgitant orifice areas and flow volumes. We demonstrate that the network improves orifice shape reconstruction, as well as the accuracy of orifice area and flow volume estimation, compared with a previous approach based on thresholding of the power-Doppler signal (THD), and compared with spatially invariant deconvolution (DC). Our approach reduces the area estimation error on simulations: (THD: 13.2 ± 9.9 mm2, DC: 12.8 ± 15.8 mm2, and ours: 3.5 ± 3.2 mm2). In a phantom experiment, our approach reduces both area estimation error (THD: 10.4 ± 8.4 mm2, DC: 10.98 ± 8.17, and ours: 9.9 ± 6.0 mm2) and flow rate estimation error (THD: 20.3 ± 9.9 ml/s, DC: 18.14 ± 13.01 ml/s, and ours: 7.1 ± 10.6 ml/s). We also demonstrate in vivo feasibility for six patients with aortic insufficiency, compared with standard echocardiography and magnetic resonance references.

Convolutional Neural Network for Predicting Thyroid Cancer Based on Ultrasound Elastography Image of Perinodular Region

We aimed to develop deep learning models based on perinodular regions' shear-wave elastography (SWE) images and ultrasound (US) images of thyroid nodules (TNs) and determine their performances in predicting thyroid cancer. A total of 1747 American College of Radiology Thyroid Imaging Reporting & Data System 4 (TR4) thyroid nodules (TNs) in 1582 patients were included in this retrospective study. US images, SWE images, and 2 quantitative SWE parameters (maximum elasticity of TNs; 5-point average maximum elasticity of TNs) were obtained. Based on US and SWE images of TNs and perinodular tissue, respectively, 7 single-image convolutional neural networks (CNN) models [US, internal SWE, 0.5 mm SWE, 1.0 mm SWE, 1.5 mm SWE, 2.0 mm SWE of perinodular tissue, and whole SWE region of interest (ROI) image] and another 6 fusional-image CNN models (US + internal SWE, US + 0.5 mm SWE, US + 1.0 mm SWE, US + 1.5 mm SWE, US + 2.0 mm SWE, US + ROI SWE) were established using RestNet18. All of the CNN models and quantitative SWE parameters were built on a training cohort (1247 TNs) and evaluated on a validation cohort (500 TNs). In predicting thyroid cancer, US + 2.0 mm SWE image CNN model obtained the highest area under the curve in 10 mm < TNs ≤ 20 mm (0.95 for training; 0.92 for validation) and TNs > 20 mm (0.95 for training; 0.92 for validation), while US + 1.0 mm SWE image CNN model obtained the highest area under the curve in TNs ≤ 10 mm (0.95 for training; 0.92 for validation). The CNN models based on the fusion of SWE segmentation images and US images improve the radiological diagnostic accuracy of thyroid cancer.

Tracked 3D ultrasound and deep neural network-based thyroid segmentation reduce interobserver variability in thyroid volumetry

Thyroid volumetry is crucial in the diagnosis, treatment, and monitoring of thyroid diseases. However, conventional thyroid volumetry with 2D ultrasound is highly operator-dependent. This study compares 2D and tracked 3D ultrasound with an automatic thyroid segmentation based on a deep neural network regarding inter- and intraobserver variability, time, and accuracy. Volume reference was MRI. 28 healthy volunteers (24—50 a) were scanned with 2D and 3D ultrasound (and by MRI) by three physicians (MD 1, 2, 3) with different experience levels (6, 4, and 1 a). In the 2D scans, the thyroid lobe volumes were calculated with the ellipsoid formula. A convolutional deep neural network (CNN) automatically segmented the 3D thyroid lobes. 26, 6, and 6 random lobe scans were used for training, validation, and testing, respectively. On MRI (T1 VIBE sequence) the thyroid was manually segmented by an experienced MD. MRI thyroid volumes ranged from 2.8 to 16.7ml (mean 7.4, SD 3.05). The CNN was trained to obtain an average Dice score of 0.94. The interobserver variability comparing two MDs showed mean differences for 2D and 3D respectively of 0.58 to 0.52ml (MD1 vs. 2), −1.33 to −0.17ml (MD1 vs. 3) and −1.89 to −0.70ml (MD2 vs. 3). Paired samples t-tests showed significant differences for 2D (p = .140, p = .002 and p = .002) and none for 3D (p = .176, p = .722 and p = .057). Intraobsever variability was similar for 2D and 3D ultrasound. Comparison of ultrasound volumes and MRI volumes showed a significant difference for the 2D volumetry of all MDs (p = .002, p = .009, p <.001), and no significant difference for 3D ultrasound (p = .292, p = .686, p = 0.091). Acquisition time was significantly shorter for 3D ultrasound. Tracked 3D ultrasound combined with a CNN segmentation significantly reduces interobserver variability in thyroid volumetry and increases the accuracy of the measurements with shorter acquisition times.

TNSNet: Thyroid nodule segmentation in ultrasound imaging using soft shape supervision

BACKGROUND AND OBJECTIVES

Thyroid nodules are a common disorder of the endocrine system. Segmentation of thyroid nodules on ultrasound images is an important step in the evaluation and diagnosis of nodules and an initial step in computer-aided diagnostic systems. The accuracy and consistency of segmentation remain a challenge due to the low contrast, speckle noise, and low resolution of ultrasound images. Therefore, the study of deep learning-based algorithms for thyroid nodule segmentation is important. This study utilizes soft shape supervision to improve the performance of detection and segmentation of boundaries of nodules. Soft shape supervision can emphasize the boundary features and assist the network in segmenting nodules accurately.

METHODS

We propose a dual-path convolution neural network, including region and shape paths, which use DeepLabV3+ as the backbone. Soft shape supervision blocks are inserted between the two paths to implement cross-path attention mechanisms. The blocks enhance the representation of shape features and add them to the region path as auxiliary information. Thus, the network can accurately detect and segment thyroid nodules.

RESULTS

We collect 3786 ultrasound images of thyroid nodules to train and test our network. Compared with the ground truth, the test results achieve an accuracy of 95.81% and a DSC of 85.33. The visualization results also suggest that the network has learned clear and accurate boundaries of the nodules. The evaluation metrics and visualization results demonstrate the superior segmentation performance of the network to other classical deep learning-based networks.

CONCLUSIONS

The proposed dual-path network can accurately realize automatic segmentation of thyroid nodules on ultrasound images. It can also be used as an initial step in computer-aided diagnosis. It shows superior performance to other classical methods and demonstrates the potential for accurate segmentation of nodules in clinical applications.

Comparing deep learning-based automatic segmentation of breast masses to expert interobserver variability in ultrasound imaging

Deep learning is a powerful tool that became practical in 2008, harnessing the power of Graphic Processing Unites, and has developed rapidly in image, video, and natural language processing. There are ongoing developments in the application of deep learning to medical data for a variety of tasks across multiple imaging modalities. The reliability and repeatability of deep learning techniques are of utmost importance if deep learning can be considered a tool for assisting experts, including physicians, radiologists, and sonographers. Owing to the high costs of labeling data, deep learning models are often evaluated against one expert, and it is unknown if any errors fall within a clinically acceptable range. Ultrasound is a commonly used imaging modality for breast cancer screening processes and for visually estimating risk using the Breast Imaging Reporting and Data System score. This process is highly dependent on the skills and experience of the sonographers and radiologists, thereby leading to interobserver variability and interpretation. For these reasons, we propose an interobserver reliability study comparing the performance of a current top-performing deep learning segmentation model against three experts who manually segmented suspicious breast lesions in clinical ultrasound (US) images. We pretrained the model using a US thyroid segmentation dataset with 455 patients and 50,993 images, and trained the model using a US breast segmentation dataset with 733 patients and 29,884 images. We found a mean Fleiss kappa value of 0.78 for the performance of three experts in breast mass segmentation compared to a mean Fleiss kappa value of 0.79 for the performance of experts and the optimized deep learning model.