Liver Ultrasound Image Segmentation Using Region-Difference Filters.

SEG-LUS: A novel ultrasound segmentation method for liver and its accessory structures based on muti-head self-attention

Motion tracking aims to accurately localize the moving lesion during radiotherapy to ensure the accuracy of radiation delivery. Ultrasound (US) imaging is a promising imaging modality to guide radiation therapy in real time. This study proposed a deep learning-based motion tracking method to track the moving lesion in US images. To reduce the searching region, a box regression-based method is adopted to predefine a region of interest (ROI). Within the ROI, the feature pyramid network (FPN) that uses a top-down architecture with lateral connection was adopted to extract image features, and the region proposal network (RPN) that learns the attention mechanism of the annotated anatomical landmarks was then used to yield a number of proposals. The training of the networks was supervised by three training objectives, including a bounding box regression loss, a proposal classification loss and a classification loss. In addition, we employed long-short-term-memory (LSTM) to capture the temporal features from the US image sequence. The weights from transform learning were used as the initial values of our network. Two-dimensional liver US images from 24 patients and the corresponding annotated anatomical landmarks were used to train our proposed method. In the testing experiments on 11 patients, our method achieves a mean tracking error of 0.58 mm with a standard derivation of 0.44 mm in a temporal resolution of 69 frames per second. Our proposed method provides an effective and clinically feasible solution to monitor the lesion motion in radiation therapy.

Cross-modal Transfer Learning Based on an Improved CycleGAN Model for Accurate Kidney Segmentation in Ultrasound Images

Segmentation of ultrasound liver images presents a unique challenge because these images contain strong speckle noise and attenuated artifacts. Most ultrasound image segmentation techniques focus on region growing or active contours. These are semi-automatic segmenting systems, in which seed points or initial contours have to be manually identified. In this paper, we propose a fully automatic segmentation system for ultrasound liver images. We apply the Peak-and-valley method to pixels scanned along the Hubert curve, and propose a "windows adaptive threshold" procedure to further reduce noise from the images. After Otsu's segmentation algorithm is applied to the images, a core area algorithm is employed to detect liver objects with the help of a feature knowledge base. We compared our method with other techniques and the manual segmentation method. The results indicate the accuracy of our system and our automatically segmented images contain less noise than the other methods.

With the success of deep learning-based methods applied in medical image analysis, convolutional neural networks (CNNs) have been investigated for classifying liver disease from ultrasound (US) data. However, the scarcity of available large-scale labeled US data has hindered the success of CNNs for classifying liver disease from US data. In this work, we propose a novel generative adversarial network (GAN) architecture for realistic diseased and healthy liver US image synthesis. We adopt the concept of stacking to synthesize realistic liver US data. Quantitative and qualitative evaluation is performed on 550 in-vivo B-mode liver US images collected from 55 subjects. We also show that the synthesized images, together with real in vivo data, can be used to significantly improve the performance of traditional CNN architectures for Nonalcoholic fatty liver disease (NAFLD) classification.

Ultrasound (US) is the preferred modality for fatty liver disease diagnosis due to its noninvasive, real-time, and cost-effective imaging capabilities. However, traditional B-mode US is qualitative, and therefore, the assessment is very subjective. Computer-aided diagnostic tools can improve the specificity and sensitivity of US and help clinicians to perform uniform diagnoses. In this work, we propose a novel deep learning model for nonalcoholic fatty liver disease classification from US data. We design a multi-feature guided multi-scale residual convolutional neural network (CNN) architecture to capture features of different receptive fields. B-mode US images are combined with their corresponding local phase filtered images and radial symmetry transformed images as multi-feature inputs for the network. Various fusion strategies are studied to improve prediction accuracy. We evaluate the designed network architectures on B-mode in vivo liver US images collected from 55 subjects. We also provide quantitative results by comparing our proposed multi-feature CNN architecture against traditional CNN designs and machine learning methods. Quantitative results show an average classification accuracy above 90% over tenfold cross-validation. Our proposed method achieves a 97.8% area under the ROC curve (AUC) for the patient-specific leave-one-out cross-validation (LOOCV) evaluation. Comprehensive validation results further demonstrate that our proposed approaches achieve significant improvements compared to training mono-feature CNN architectures ([Formula: see text]). Feature combination is valuable for the traditional classification methods, and the use of multi-scale CNN can improve liver classification accuracy. Based on the promising performance, the proposed method has the potential in practical applications to help radiologists diagnose nonalcoholic fatty liver disease.

The aim of this study was to test the hypothesis that the Computer Aided UltraSound (CAUS) method developed by the authors [1-4] for the estimation of UltraSound Tissue Characteristics (UTC) parameters on transcutaneous (Transc) ultrasound (US) images can predict the liver fat content with similar accuracy and precision as with intraoperative (Intraop) US. A large animal study in post partum dairy cows (N=151) was performed to test these hypotheses. Five Transc B-Mode US liver image were acquired before surgery. During abomasal displacement surgery five Intraop US B-Mode liver images and a liver biopsy was taken. In liver tissue samples, triacylglycerol (TAG) content was measured by biochemical analysis. Firstly the equipment preset, which was kept fixed during whole study time, was carefully calibrated[5]. For the echo level calibration a TMP was used, and all UTC parameters were expressed relatively to those of the phantom. Prior to UTC parameters estimation several pre-processing steps were performed: Back-Scan Conversion (BSC); Look Up Table (LUT) correction; superficial tissue layers (Fat layer) attenuation correction and Automatic Gain Correction (AGC) were performed. Also several postprocessing steps were incorporated like: Automatic segmentation and residual attenuation correction were performed. Stepwise multiple linear regression analysis on a training set (N=76) was performed. In all cases the Residual Attenuation coefficient (ResAtt, R=0.81) was the only selected parameter. The results were tested on the residual cows (test set N=75) to predict the TAG content in the liver. Receiver Operating Characteristics (ROC) analysis then was applied to estimate the Area Under the Curve (AUC) and the sensitivity and specificity of the CAUS method. Equivalent high predictive values for AUC (95%), sensitivity(87%) and specificity (83%) for Intraop and Transc applications were found. Consequently, it can be concluded, applied Fat layer attenuation correction to Transc US images was performed adequately.

This paper presents a wavelet-based hybrid threshold method according to the soft- and hard-threshold functions proposed by Donoho. The wavelet-based hybrid threshold method may help doctors to know more details on the liver disease through denoising the ultrasound image of the liver. First of all, an analytical expression for the hybrid threshold function is discussed. The wavelet-based hybrid threshold method is then investigated for ultrasound image of the liver. Finally, we test the influence of this parameter on the proposed method with the real ultrasound image corrupted by speckle noise with different variances. Moreover, we compare the proposed method under the varying parameters with the soft-threshold function and the hard-threshold function. Three metrics, which are correlation coefficient, edge preservation index and structural similarity index, are measured to quantify the denoised results of ultrasound liver image. Experimental results demonstrate the potential of the proposed method for ultrasound liver image denosing.

Ultrasound imaging is an effective approach for diagnosing breast cancer, but it is highly operator-dependent. Recent advances in computer-aided diagnosis have suggested that it can assist physicians in diagnosis. Definition of the region of interest before computer analysis is still needed. Since manual outlining of the tumor contour is tedious and time-consuming for a physician, developing an automatic segmentation method is important for clinical application. The present paper represents a novel method to segment breast ultrasound images. It utilizes a combination of region-based active contour and neutrosophic theory to overcome the natural properties of ultrasound images including speckle noise and tissue-related textures. First, due to inherent speckle noise and low contrast of these images, we have utilized a non-local means filter and fuzzy logic method for denoising and image enhancement, respectively. This paper presents an improved weighted region-scalable active contour to segment breast ultrasound images using a new feature derived from neutrosophic theory. This method has been applied to 36 breast ultrasound images. It generates true-positive and false-positive results, and similarity of 95%, 6%, and 90%, respectively. The purposed method indicates clear advantages over other conventional methods of active contour segmentation, i.e., region-scalable fitting energy and weighted region-scalable fitting energy.

Ultrasound (US) image segmentation is a difficult task because of its heavy speckle noise, low quality and blurry boundaries. In this paper, a new neural network based method is proposed for ultrasound images segmentation. A modified self organizing map (SOM) network, named finite impulse response SOM (FIR-SOM), is utilized to segment ultrasound images. A two dimensional (2D) discrete wavelet transform (DWT) is used to build the input feature space of the network. Experimental results show that FIR-SOM discovers the pattern of the input image properly and is robust against noise. Segmentation results of breast ultrasound images (BUS) demonstrate that there is a strong correlation between tumor region selected by a physician and the tumor region segmented by our proposed method.

PurposeTo overcome the severe intensity inhomogeneity and blurry boundaries in HIFU (High Intensity Focused Ultrasound) ultrasound images, an accurate and efficient multi-scale and shape constrained localized region-based active contour model (MSLCV), was developed to accurately and efficiently segment the target region in HIFU ultrasound images of uterine fibroids.MethodsWe incorporated a new shape constraint into the localized region-based active contour, which constrained the active contour to obtain the desired, accurate segmentation, avoiding boundary leakage and excessive contraction. Localized region-based active contour modeling is suitable for ultrasound images, but it still cannot acquire satisfactory segmentation for HIFU ultrasound images of uterine fibroids. We improved the localized region-based active contour model by incorporating a shape constraint into region-based level set framework to increase segmentation accuracy. Some improvement measures were proposed to overcome the sensitivity of initialization, and a multi-scale segmentation method was proposed to improve segmentation efficiency. We also designed an adaptive localizing radius size selection function to acquire better segmentation results.ResultsExperimental results demonstrated that the MSLCV model was significantly more accurate and efficient than conventional methods. The MSLCV model has been quantitatively validated via experiments, obtaining an average of 0.94 for the DSC (Dice similarity coefficient) and 25.16 for the MSSD (mean sum of square distance). Moreover, by using the multi-scale segmentation method, the MSLCV model’s average segmentation time was decreased to approximately 1/8 that of the localized region-based active contour model (the LCV model).ConclusionsAn accurate and efficient multi-scale and shape constrained localized region-based active contour model was designed for the semi-automatic segmentation of uterine fibroid ultrasound (UFUS) images in HIFU therapy. Compared with other methods, it provided more accurate and more efficient segmentation results that are very close to those obtained from manual segmentation by a specialist.

Segmenting the lesion region from 2D ultrasound (US) images is an importance step in defining the intra-operative planning of computer-assisted therapy. However, it still remains a challenge owing to two reasons: (1) The signal-to-noise ratio (SNR) of the US images is originally low, and the inhomogeneous distribution of intensity causes local fuzzy boundaries. (2) The deformation of soft tissue causes makes the geometry contour of the segmentation area complex. Because of these factors, the classical segmentation methods do not yield the desired results in US images. To solve these problems, a novel segmentation method based on a dynamics statistical shape model is proposed. In this study, first, a radial time series model was defined for use as a parametric shape model of the lesion region, and a stochastic difference equation (SDE) was used as the contour dynamics generating model; then, the coefficients of the dynamics generating model and the shape prior probability were deduced by utilizing the property of Fokker-Planck equation (FPE) and training the samples. This statistical shape model (SSM) was referred to as the SF-SSM. Furthermore, a radial profile feature model (RPFM) was developed based on the statistical features of the intensity distribution from the inside to outside of the lesion region and based on the radial time series model. The RPFM provided a likelihood probability for the prior shape. Finally, through the segmentation optimization process, the maximum posteriori process was determined. In order to verify the performance of our method, we used real clinical ultrasound images as the training and test sets. We compared the proposed model with three well-known segmentation methods for the same test set. The experimental results showed that the proposed method provided more accurate results than the other methods for different quality US images. Furthermore, the friendly initialization of the proposed method improves the efficiency of interactive segmentation, thus benefitting computer-assisted therapy.

Due to the low contrast, blurry boundaries, and large amount of shadows in breast ultrasound (BUS) images, automatic tumor segmentation remains a challenging task. Deep learning provides a solution to this problem, since it can effectively extract representative features from lesions and the background in BUS images. A novel automatic tumor segmentation method is proposed by combining a dilated fully convolutional network (DFCN) with a phase-based active contour (PBAC) model. The DFCN is an improved fully convolutional neural network with dilated convolution in deeper layers, fewer parameters, and batch normalization techniques; and has a large receptive field that can separate tumors from background. The predictions made by the DFCN are relatively rough due to blurry boundaries and variations in tumor sizes; thus, the PBAC model, which adds both region-based and phase-based energy functions, is applied to further improve segmentation results. The DFCN model is trained and tested in dataset 1 which contains 570BUS images from 89 patients. In dataset 2, a 10-fold support vector machine (SVM) classifier is employed to verify the diagnostic ability using 460 features extracted from the segmentation results of the proposed method. Advantages of the present method were compared with three state-of-the-art networks; the FCN-8s, U-net, and dilated residual network (DRN). Experimental results from 170BUS images show that the proposed method had a Dice Similarity coefficient of 88.97±10.01%, a Hausdorff distance (HD) of 35.54±29.70 pixels, and a mean absolute deviation (MAD) of 7.67±6.67 pixels, which showed the best segmentation performance. In dataset 2, the area under curve (AUC) of the 10-fold SVM classifier was 0.795 which is similar to the classification using the manual segmentation results. The proposed automatic method may be sufficiently accurate, robust, and efficient for medical ultrasound applications.

The preliminary study presented within this paper shows a comparative study of various texture features extracted from liver ultrasonic images by employing Multilayer Perceptron (MLP), a type of artificial neural network, to study the presence of disease conditions. An ultrasound (US) image shows echo-texture patterns, which defines the organ characteristics. Ultrasound images of liver disease conditions such as “fatty liver,” “cirrhosis,” and “hepatomegaly” produce distinctive echo patterns. However, various ultrasound imaging artifacts and speckle noise make these echo-texture patterns difficult to identify and often hard to distinguish visually. Here, based on the extracted features from the ultrasonic images, we employed an artificial neural network for the diagnosis of disease conditions in liver and finding of the best classifier that distinguishes between abnormal and normal conditions of the liver. Comparison of the overall performance of all the feature classifiers concluded that “mixed feature set” is the best feature set. It showed an excellent rate of accuracy for the training data set. The gray level run length matrix (GLRLM) feature shows better results when the network was tested against unknown data.

The visual quality of ultrasound (US) images is crucial for clinical diagnosis and treatment. The main source of image quality degradation is the inherent speckle noise generated during US image acquisition. Current deep learning-based methods cannot preserve the maximum boundary contrast when removing noise and speckle. In this paper, we address the issue by proposing a novel wavelet-based generative adversarial network (GAN) for real-time high-quality US image reconstruction, viz. WGAN-DUS. First, we propose a batch normalization module (BNM) to balance the importance of each sub-band image and fuse sub-band features simultaneously. Then, a wavelet reconstruction module (WRM) integrated with a cascade of wavelet residual channel attention block (WRCAB) is proposed to extract distinctive sub-band features used to reconstruct denoised images. A gradual tuning strategy is proposed to fine-tune our generator for better despeckling performance. We further propose a wavelet-based discriminator and a comprehensive loss function to effectively suppress speckle noise and preserve the image features. Besides, we have designed an algorithm to estimate the noise levels during despeckling of real US images. The performance of our network was then evaluated on natural, synthetic, simulated and clinical US images and compared against various despeckling methods. To verify the feasibility of WGAN-DUS, we further extend our work to uterine fibroid segmentation with the denoised US image of the proposed approach. Experimental result demonstrates that our proposed method is feasible and can be generalized to clinical applications for despeckling of US images in real-time without losing its fine details.