Phase-Based Level Set Segmentation of Ultrasound Images

Baran T, O’Connell A, Parker K. Enhancing breast ultrasound segmentation through fine-tuning and optimization techniques: Sharp attention UNet

Segmentation of breast ultrasound images is a crucial and challenging task in computer-aided diagnosis systems. Accurately segmenting masses in benign and malignant cases and identifying regions with no mass is a primary objective in breast ultrasound image segmentation. Deep learning (DL) has emerged as a powerful tool in medical image segmentation, revolutionizing how medical professionals analyze and interpret complex imaging data. The UNet architecture is a highly regarded and widely used DL model in medical image segmentation. Its distinctive architectural design and exceptional performance have made it popular among researchers. With the increase in data and model complexity, optimization and fine-tuning models play a vital and more challenging role than before. This paper presents a comparative study evaluating the effect of image preprocessing and different optimization techniques and the importance of fine-tuning different UNet segmentation models for breast ultrasound images. Optimization and fine-tuning techniques have been applied to enhance the performance of UNet, Sharp UNet, and Attention UNet. Building upon this progress, we designed a novel approach by combining Sharp UNet and Attention UNet, known as Sharp Attention UNet. Our analysis yielded the following quantitative evaluation metrics for the Sharp Attention UNet: the Dice coefficient, specificity, sensitivity, and F1 score values obtained were 0.93, 0.99, 0.94, and 0.94, respectively. In addition, McNemar's statistical test was applied to assess significant differences between the approaches. Across a number of measures, our proposed model outperformed all other models, resulting in improved breast lesion segmentation.

The Big Bang of Deep Learning in Ultrasound-Guided Surgery: A Review

Ultrasound (US) imaging is a paramount modality in many image-guided surgeries and percutaneous interventions, thanks to its high portability, temporal resolution, and cost-efficiency. However, due to its imaging principles, the US is often noisy and difficult to interpret. Appropriate image processing can greatly enhance the applicability of the imaging modality in clinical practice. Compared with the classic iterative optimization and machine learning (ML) approach, deep learning (DL) algorithms have shown great performance in terms of accuracy and efficiency for US processing. In this work, we conduct a comprehensive review on deep-learning algorithms in the applications of US-guided interventions, summarize the current trends, and suggest future directions on the topic.

A computer-aided method based on geometrical texture features for a precocious detection of fetal Hydrocephalus in ultrasound images

BACKGROUD

Hydrocephalus is the most common anomaly of the fetal head characterized by an excessive accumulation of fluid in the brain processing. The diagnostic process of fetal heads using traditional evaluation techniques are generally time consuming and error prone. Usually, fetal head size is computed using an ultrasound (US) image around 20-22 weeks, which is the gestational age (GA). Biometrical measurements are extracted and compared with ground truth charts to identify normal or abnormal growth.

METHODS

In this paper, an attempt has been made to enhance the Hydrocephalus characterization process by extracting other geometrical and textural features to design an efficient recognition system. The superiority of this work consists of the reduced time processing and the complexity of standard automatic approaches for routine examination. This proposed method requires practical insidiousness of the precocious discovery of fetuses' malformation to alert the experts about the existence of abnormal outcome. The first task is devoted to a proposed pre-processing model using a standard filtering and a segmentation scheme using a modified Hough transform (MHT) to detect the region of interest. Indeed, the obtained clinical parameters are presented to the principal component analysis (PCA) model in order to obtain a reduced number of measures which are employed in the classification stage.

RESULTS

Thanks to the combination of geometrical and statistical features, the classification process provided an important ability and an interesting performance achieving more than 96% of accuracy to detect pathological subjects in premature ages.

CONCLUSIONS

The experimental results illustrate the success and the accuracy of the proposed classification method for a factual diagnostic of fetal head malformation.

Kidney Segmentation in 3-D Ultrasound Images Using a Fast Phase-Based Approach

Renal ultrasound (US) imaging is the primary imaging modality for the assessment of the kidney's condition and is essential for diagnosis, treatment and surgical intervention planning, and follow-up. In this regard, kidney delineation in 3-D US images represents a relevant and challenging task in clinical practice. In this article, a novel framework is proposed to accurately segment the kidney in 3-D US images. The proposed framework can be divided into two stages: 1) initialization of the segmentation method and 2) kidney segmentation. Within the initialization stage, a phase-based feature detection method is used to detect edge points at kidney boundaries, from which the segmentation is automatically initialized. In the segmentation stage, the B-spline explicit active surface framework is adapted to obtain the final kidney contour. Here, a novel hybrid energy functional that combines localized region- and edge-based terms is used during segmentation. For the edge term, a fast-signed phase-based detection approach is applied. The proposed framework was validated in two distinct data sets: 1) 15 3-D challenging poor-quality US images used for experimental development, parameters assessment, and evaluation and 2) 42 3-D US images (both healthy and pathologic kidneys) used to unbiasedly assess its accuracy. Overall, the proposed method achieved a Dice overlap around 81% and an average point-to-surface error of ~2.8 mm. These results demonstrate the potential of the proposed method for clinical usage.

Harnessing Machine Intelligence in Automatic Echocardiogram Analysis: Current Status, Limitations, and Future Directions

Echocardiography (echo) is a critical tool in diagnosing various cardiovascular diseases. Despite its diagnostic and prognostic value, interpretation and analysis of echo images are still widely performed manually by echocardiographers. A plethora of algorithms has been proposed to analyze medical ultrasound data using signal processing and machine learning techniques. These algorithms provided opportunities for developing automated echo analysis and interpretation systems. The automated approach can significantly assist in decreasing the variability and burden associated with manual image measurements. In this paper, we review the state-of-the-art automatic methods for analyzing echocardiography data. Particularly, we comprehensively and systematically review existing methods of four major tasks: echo quality assessment, view classification, boundary segmentation, and disease diagnosis. Our review covers three echo imaging modes, which are B-mode, M-mode, and Doppler. We also discuss the challenges and limitations of current methods and outline the most pressing directions for future research. In summary, this review presents the current status of automatic echo analysis and discusses the challenges that need to be addressed to obtain robust systems suitable for efficient use in clinical settings or point-of-care testing.

Surgical navigation system for laparoscopic lateral pelvic lymph node dissection in rectal cancer surgery using laparoscopic-vision-tracked ultrasonic imaging

BACKGROUND

Laparoscopic lateral pelvic lymph node dissection (LPLND) in rectal cancer surgery requires considerable skill because the pelvic arteries, which need to be located to guide the dissection, are covered by other tissues and cannot be observed on laparoscopic views. Therefore, surgeons need to localize the pelvic arteries accurately before dissection, to prevent injury to these arteries.

METHODS

This report proposes a surgical navigation system to facilitate artery localization in laparoscopic LPLND by combining ultrasonic imaging and laparoscopy. Specifically, free-hand laparoscopic ultrasound (LUS) is employed to capture the arteries intraoperatively in this approach, and a laparoscopic vision-based tracking system is utilized to track the LUS probe. To extract the artery contours from the two-dimensional ultrasound image sequences efficiently, an artery extraction framework based on local phase-based snakes was developed. After reconstructing the three-dimensional intraoperative artery model from ultrasound images, a high-resolution artery model segmented from preoperative computed tomography (CT) images was rigidly registered to the intraoperative artery model and overlaid onto the laparoscopic view to guide laparoscopic LPLND.

RESULTS

Experiments were conducted to evaluate the performance of the vision-based tracking system, and the average reconstruction error of the proposed tracking system was found to be 2.4 mm. Then, the proposed navigation system was quantitatively evaluated on an artery phantom. The reconstruction time and average navigation error were 8 min and 2.3 mm, respectively. A navigation system was also successfully constructed to localize the pelvic arteries in laparoscopic and open surgeries of a swine. This demonstrated the feasibility of the proposed system in vivo. The construction times in the laparoscopic and open surgeries were 14 and 12 min, respectively.

CONCLUSIONS

The experimental results showed that the proposed navigation system can guide laparoscopic LPLND and requires a significantly shorter setting time than the state-of-the-art navigation systems do.

3D-MRI Brain Tumor Detection Model Using Modified Version of Level Set Segmentation Based on Dragonfly Algorithm

Accurate brain tumor segmentation from 3D Magnetic Resonance Imaging (3D-MRI) is an important method for obtaining information required for diagnosis and disease therapy planning. Variation in the brain tumor’s size, structure, and form is one of the main challenges in tumor segmentation, and selecting the initial contour plays a significant role in reducing the segmentation error and the number of iterations in the level set method. To overcome this issue, this paper suggests a two-step dragonfly algorithm (DA) clustering technique to extract initial contour points accurately. The brain is extracted from the head in the preprocessing step, then tumor edges are extracted using the two-step DA, and these extracted edges are used as an initial contour for the MRI sequence. Lastly, the tumor region is extracted from all volume slices using a level set segmentation method. The results of applying the proposed technique on 3D-MRI images from the multimodal brain tumor segmentation challenge (BRATS) 2017 dataset show that the proposed method for brain tumor segmentation is comparable to the state-of-the-art methods.

K-Net: Integrate Left Ventricle Segmentation and Direct Quantification of Paired Echo Sequence

The integration of segmentation and direct quantification on the left ventricle (LV) from the paired apical views(i.e., apical 4-chamber and 2-chamber together) echo sequence clinically achieves the comprehensive cardiac assessment: multiview segmentation for anatomical morphology, and multidimensional quantification for contractile function. Direct quantification of LV, i.e., to automatically quantify multiple LV indices directly from the image via task-aware feature representation and regression, avoids accumulative error from the inter-step target. This integration sequentially makes a stereoscopical reflection of cardiac activity jointly from the paired orthogonal cross views sequences, overcoming limited observation with a single plane. We propose a K-shaped Unified Network (K-Net), the first end-to-end framework to simultaneously segment LV from apical 4-chamber and 2-chamber views, and directly quantify LV from major- and minor-axis dimensions (1D), area (2D), and volume (3D), in sequence. It works via four components: 1) the K-Net architecture with the Attention Junction enables heterogeneous tasks learning of segmentation task of pixel-wise classification, and direct quantification task of image-wise regression, by interactively introducing the information from segmentation to jointly promote spatial attention map to guide quantification focusing on LV-related region, and transferring quantification feedback to make global constraint on segmentation; 2) the Bi-ResLSTMs distributed in K-Net layer-by-layer hierarchically extract spatial-temporal information in echo sequence, with bidirectional recurrent and short-cut connection to model spatial-temporal information among all frames; 3) the Information Valve tailing the Bi-ResLSTMs selectively exchanges information among multiple views, by stimulating complementary information and suppressing redundant information to make the efficient cross-flow for each view; 4) the Evolution Loss comprehensively guides sequential data learning, with static constraint for frame values, and dynamic constraint for inter-frame value changes. The experiments show that our K-Net gains high performance with a Dice coefficient up to 91.44% and a mean absolute error of the major-axis dimension down to 2.74mm, which reveal its clinical potential.

Automating Spine Curvature Measurement in Volumetric Ultrasound via Adaptive Phase Features

Ultrasound volume projection imaging (VPI) has been recently suggested. This novel imaging method allows a non-radiation assessment of spine deformity with free-hand 3-D ultrasound imaging. This paper presents a fully automatic method to evaluate the spine curve in VPI images corresponding to different projection depth of the volumetric ultrasound, thus making it possible to analyze 3-D spine deformity. The new automatic method is based on prior knowledge about the geometric arrangement of the spinous processes. The frequency bandwidth of log-Gabor filters is adaptively adjusted to calculate the oriented phase congruency, facilitating the segmentation of the spinous column profile. And the spine curvature angle is finally calculated according to the inflection points of the curve over the segmented spinous column profile. The performance of the automatic method is evaluated on spine VPI images among patients with different scoliotic angles. The curvature angles obtained using the proposed method have a high linear correlation with those by the manual method (r = 0.90, p < 0.001) and X-ray Cobb's method (r = 0.87, p < 0.001). The feasibility of 3-D spine deformity assessment is also demonstrated using VPI images corresponding to various projection depth. The results suggest that this method can substantially improve the recognition of the spinous column profile, especially facilitating the applications of 3-D spine deformity assessment.

Bimodal Automated Carotid Ultrasound Segmentation Using Geometrically Constrained Deep Neural Networks

For asymptomatic patients suffering from carotid stenosis, the assessment of plaque morphology is an important clinical task which allows monitoring of the risk of plaque rupture and future incidents of stroke. Ultrasound Imaging provides a safe and non-invasive modality for this, and the segmentation of media-adventitia boundaries and lumen-intima boundaries of the Carotid artery form an essential part in this monitoring process. In this paper, we propose a novel Deep Neural Network as a fully automated segmentation tool, and its application in delineating both the media-adventitia boundary and the lumen-intima boundary. We develop a new geometrically constrained objective function as part of the Network's Stochastic Gradient Descent optimisation, thus tuning it to the problem at hand. Furthermore, we also apply a bimodal fusion of amplitude and phase congruency data proposed by us in previous work, as an input to the network, as the latter provides an intensity-invariant data source to the network. We finally report the segmentation performance of the network on transverse sections of the carotid. Tests are carried out on an augmented dataset of 81,000 images, and the results are compared to other studies by reporting the DICE coefficient of similarity, modified Hausdorff Distance, sensitivity and specificity. Our proposed modification is shown to yield improved results on the standard network over this larger dataset, with the advantage of it being fully automated. We conclude that Deep Neural Networks provide a reliable trained manner in which carotid ultrasound images may be automatically segmented, using amplitude data and intensity invariant phase congruency maps as a data source.

Phase asymmetry ultrasound despeckling with fractional anisotropic diffusion and total variation

We propose an ultrasound speckle filtering method for not only preserving various edge features but also filtering tissue-dependent complex speckle noises in ultrasound images. The key idea is to detect these various edges using a phase congruence-based edge significance measure called phase asymmetry (PAS), which is invariant to the intensity amplitude of edges and takes 0 in non-edge smooth regions and 1 at the idea step edge, while also taking intermediate values at slowly varying ramp edges. By leveraging the PAS metric in designing weighting coefficients to maintain a balance between fractional-order anisotropic diffusion and total variation (TV) filters in TV cost function, we propose a new fractional TV framework to not only achieve the best despeckling performance with ramp edge preservation but also reduce the staircase effect produced by integral-order filters. Then, we exploit the PAS metric in designing a new fractional-order diffusion coefficient to properly preserve low-contrast edges in diffusion filtering. Finally, different from fixed fractional-order diffusion filters, an adaptive fractional order is introduced based on the PAS metric to enhance various weak edges in the spatially transitional areas between objects. The proposed fractional TV model is minimized using the gradient descent method to obtain the final denoised image. The experimental results and real application of ultrasound breast image segmentation show that the proposed method outperforms other state-of-the-art ultrasound despeckling filters for both speckle reduction and feature preservation in terms of visual evaluation and quantitative indices. The best scores on feature similarity indices have achieved 0.867, 0.844 and 0.834 under three different levels of noise, while the best breast ultrasound segmentation accuracy in terms of the mean and median dice similarity coefficient are 96.25% and 96.15%, respectively.

Medical breast ultrasound image segmentation by machine learning

Breast cancer is the most commonly diagnosed cancer, which alone accounts for 30% all new cancer diagnoses for women, posing a threat to women's health. Segmentation of breast ultrasound images into functional tissues can aid tumor localization, breast density measurement, and assessment of treatment response, which is important to the clinical diagnosis of breast cancer. However, manually segmenting the ultrasound images, which is skill and experience dependent, would lead to a subjective diagnosis; in addition, it is time-consuming for radiologists to review hundreds of clinical images. Therefore, automatic segmentation of breast ultrasound images into functional tissues has received attention in recent years, amidst the more numerous studies of detection and segmentation of masses. In this paper, we propose to use convolutional neural networks (CNNs) for segmenting breast ultrasound images into four major tissues: skin, fibroglandular tissue, mass, and fatty tissue, on three-dimensional (3D) breast ultrasound images. Quantitative metrics for evaluation of segmentation results including Accuracy, Precision, Recall, and F1_measure, all reached over 80%, which indicates that the method proposed has the capacity to distinguish functional tissues in breast ultrasound images. Another metric called the Jaccard similarity index (JSI) yields an 85.1% value, outperforming our previous study using the watershed algorithm with 74.54% JSI value. Thus, our proposed method might have the potential to provide the segmentations necessary to assist the clinical diagnosis of breast cancer and improve imaging in other modes in medical ultrasound.

Fast Segmentation of the Left Atrial Appendage in 3-D Transesophageal Echocardiographic Images

Left atrial appendage (LAA) has been generally described as "our most lethal attachment," being considered the major source of thromboembolism in patients with nonvalvular atrial fibrillation. Currently, LAA occlusion can be offered as a treatment for these patients, obstructing the LAA through a percutaneously delivered device. Nevertheless, correct device sizing is not straightforward, requiring manual analysis of peri-procedural images. This approach is suboptimal, time demanding, and highly variable between experts, which can result in lengthy procedures and excess manipulations. In this paper, a semiautomatic LAA segmentation technique for 3-D transesophageal echocardiography (TEE) images is presented. Specifically, the proposed technique relies on a novel segmentation pipeline where a curvilinear blind-ended model is optimized through a double stage strategy: 1) fast contour evolution using global terms and 2) contour refinement based on regional energies. To reduce its computational cost, and thus make it more attractive to real interventions, the B-spline explicit active surface framework was used. This novel method was evaluated in a clinical database of 20 patients. Manual analysis performed by two observers was used as ground truth. The 3-D segmentation results corroborated the accuracy, robustness to the variation of the parameters, and computationally attractiveness of the proposed method, taking approximately 14 s to segment the LAA with an average accuracy of ~0.9 mm. Moreover, a performance comparable to the interobserver variability was found. Finally, the advantages of the segmented model were evaluated, while semiautomatically extracting the clinical measurements for device selection, showing a similar accuracy but with a higher reproducibility when compared to the current practice. Overall, the proposed segmentation method shows potential for an improved planning of LAA occlusion, demonstrating its added value for normal clinical practice.

Image-guided laparoscopic pelvic lymph node dissection using stereo visual tracking free-hand laparoscopic ultrasound

Laparoscopic pelvic lymph node dissection is a delicate operation because pelvic arteries, which should be located first to guide the dissection, are often concealed by tissues and cannot be identified in the endoscopic view. Consequently, arteries can be damaged if they are not located accurately. To improve dissection safety and efficiency, we have developed an image-guided navigation system to provide pelvic artery position information by registering a 3D artery model extracted from CT images to a 3D model reconstructed from free-hand laparoscopic ultrasound images. The ultrasound probe is tracked using a proposed stereo vision-based tracking strategy that can simplify the system and reduce setup time. The artery is segmented from 2D ultrasound images using a local phase-based snakes framework. The accuracy of the proposed navigation system was estimated in a phantom experiment (the TRE error was 1.58 ± 0.70 mm), and the feasibility of the proposed navigation system was confirmed in an animal experiment.

Left ventricular MRI wall motion assessment by monogenic signal amplitude image computation

BACKGROUND

Cardiac Magnetic Resonance Imaging (MRI) is the commonly used technique for the assessment of left ventricular (LV) function. Apart manually or semi-automatically contouring LV boundaries for quantification of By visual interpretation of cine images, assessment of regional wall motion is performed by visual interpretation of cine images, thus relying on an experience-dependent and subjective modality.

OBJECTIVE

The aim of this work is to describe a novel algorithm based on the computation of the monogenic amplitude image to be utilized in conjunction with conventional cine-MRI visualization to assess LV motion abnormalities and to validate it against gold standard expert visual interpretation.

METHODS

The proposed method uses a recent image processing tool called "monogenic signal" to decompose the MR images into features, which are relevant for motion estimation. Wall motion abnormalities are quantified locally by measuring the temporal variations of the monogenic signal amplitude. The new method was validated by two non-expert radiologists using a wall motion scoring without and with the computed image, and compared against the expert interpretation. The proposed approach was tested on a population of 40 patients, including 8 subjects with normal ventricular function and 32 pathological cases (20 with myocardial infarction, 9 with myocarditis, and 3 with dilated cardiomyopathy).

RESULTS

The results show that, for both radiologists, sensitivity, specificity and accuracy of cine-MRI alone were similar and around 59%, 77%, and 71%, respectively. Adding the proposed amplitude image while visualizing the cine MRI images significantly increased both sensitivity, specificity and accuracy up to 75%, 89%, and 84%, respectively.

CONCLUSION

Accuracy of wall motion interpretation adding amplitude image to conventional visualization was proven feasible and superior to standard image interpretation on the considered population, in inexperienced observers. Adding the amplitude images as a diagnostic tool in clinical routine is likely to improve the detection of myocardial segments presenting a cardiac dysfunction.

Three-Dimensional Registration of Freehand-Tracked Ultrasound to CT Images of the Talocrural Joint

A rigid surface⁻volume registration scheme is presented in this study to register computed tomography (CT) and free-hand tracked ultrasound (US) images of the talocrural joint. Prior to registration, bone surfaces expected to be visible in US are extracted from the CT volume and bone contours in 2D US data are enhanced based on monogenic signal representation of 2D US images. A 3D monogenic signal data is reconstructed from the 2D data using the position of the US probe recorded with an optical tracking system. When registering the surface extracted from the CT scan to the monogenic signal feature volume, six transformation parameters are estimated so as to optimize the sum of monogenic signal features over the transformed surface. The robustness of the registration algorithm was tested on a dataset collected from 12 cadaveric ankles. The proposed method was used in a clinical case study to investigate the potential of US imaging for pre-operative planning of arthroscopic access to talar (osteo)chondral defects (OCDs). The results suggest that registrations with a registration error of 2 mm and less is achievable, and US has the potential to be used in assessment of an OCD' arthroscopic accessibility, given the fact that 51% of the talar surface could be visualized.

Empirical Mode Decomposition and Monogenic Signal-Based Approach for Quantification of Myocardial Infarction From MR Images

Quantification of myocardial infarction on late Gadolinium enhancement cardiovascular magnetic resonance (LGE-CMR) images into heterogeneous infarct periphery (or gray zone) and infarct core plays an important role in cardiac diagnosis, especially in identifying patients at high risk of cardiovascular mortality. However, quantification task is challenging due to noise corrupted in cardiac MR images, the contrast variation, and limited resolution of images. In this study, we propose a novel approach for automatic myocardial infarction quantification, termed DEMPOT, which consists of three key parts: Decomposition of image into intrinsic modes, monogenic phase performing on combined dominant modes, and multilevel Otsu thresholding on the phase. In particular, inspired by the Hilbert-Huang transform, we perform the multidimensional ensemble empirical mode decomposition and 2-D generalization of the Hilbert transform known as the Riesz transform on the MR image to obtain the monogenic phase that is robust to noise and contrast variation. Then, a two-stage algorithm using multilevel Otsu thresholding is accomplished on the monogenic phase to automatically quantify the myocardium into healthy, gray zone, and infarct core regions. Experiments on LGE-CMR images with myocardial infarction from 82 patients show the superior performance of the proposed approach in terms of reproducibility, robustness, and effectiveness.

Automatic Myotendinous Junction Tracking in Ultrasound Images with Phase-Based Segmentation

Displacement of the myotendinous junction (MTJ) obtained by ultrasound imaging is crucial to quantify the interactive length changes of muscles and tendons for understanding the mechanics and pathological conditions of the muscle-tendon unit during motion. However, the lack of a reliable automatic measurement method restricts its application in human motion analysis. This paper presents an automated measurement of MTJ displacement using prior knowledge on tendinous tissues and MTJ, precluding the influence of nontendinous components on the estimation of MTJ displacement. It is based on the perception of tendinous features from musculoskeletal ultrasound images using Radon transform and thresholding methods, with information about the symmetric measures obtained from phase congruency. The displacement of MTJ is achieved by tracking manually marked points on tendinous tissues with the Lucas-Kanade optical flow algorithm applied over the segmented MTJ region. The performance of this method was evaluated on ultrasound images of the gastrocnemius obtained from 10 healthy subjects (26.0 ± 2.9 years of age). Waveform similarity between the manual and automatic measurements was assessed by calculating the overall similarity with the coefficient of multiple correlation (CMC). In vivo experiments demonstrated that MTJ tracking with the proposed method (CMC = 0.97 ± 0.02) was more consistent with the manual measurements than existing optical flow tracking methods (CMC = 0.79 ± 0.11). This study demonstrated that the proposed method was robust to the interference of nontendinous components, resulting in a more reliable measurement of MTJ displacement, which may facilitate further research and applications related to the architectural change of muscles and tendons.

Automatic Measurement of Spine Curvature on 3-D Ultrasound Volume Projection Image With Phase Features

This paper presents an automated measurement of spine curvature by using prior knowledge on vertebral anatomical structures in ultrasound volume projection imaging (VPI). This method can be used in scoliosis assessment with free-hand 3-D ultrasound imaging. It is based on the extraction of bony features from VPI images using a newly proposed two-fold thresholding strategy, with information of the symmetric and asymmetric measures obtained from phase congruency. The spinous column profile is detected from the segmented bony regions, and it is further used to extract a curve representing spine profile. The spine curvature is then automatically calculated according to the inflection points along the curve. The algorithm was evaluated on volunteers with the different severity of scoliosis. The results obtained using the newly developed method had a good linear correlation with those by the manual method (r ≥ 0.90, p <; 0.001) and X-ray Cobb's method (r = 0.83, p <; 0.001). The bigger variations observed in the manual measurement also implied that the automatic method is more reliable. The proposed method can be a promising approach for facilitating the applications of 3-D ultrasound imaging in the diagnosis, treatment, and screening of scoliosis.

FUIQA: Fetal Ultrasound Image Quality Assessment With Deep Convolutional Networks

The quality of ultrasound (US) images for the obstetric examination is crucial for accurate biometric measurement. However, manual quality control is a labor intensive process and often impractical in a clinical setting. To improve the efficiency of examination and alleviate the measurement error caused by improper US scanning operation and slice selection, a computerized fetal US image quality assessment (FUIQA) scheme is proposed to assist the implementation of US image quality control in the clinical obstetric examination. The proposed FUIQA is realized with two deep convolutional neural network models, which are denoted as L-CNN and C-CNN, respectively. The L-CNN aims to find the region of interest (ROI) of the fetal abdominal region in the US image. Based on the ROI found by the L-CNN, the C-CNN evaluates the image quality by assessing the goodness of depiction for the key structures of stomach bubble and umbilical vein. To further boost the performance of the L-CNN, we augment the input sources of the neural network with the local phase features along with the original US data. It will be shown that the heterogeneous input sources will help to improve the performance of the L-CNN. The performance of the proposed FUIQA is compared with the subjective image quality evaluation results from three medical doctors. With comprehensive experiments, it will be illustrated that the computerized assessment with our FUIQA scheme can be comparable to the subjective ratings from medical doctors.

Weighted Level Set Evolution Based on Local Edge Features for Medical Image Segmentation

Level set methods have been widely used to implement active contours for image segmentation applications due to their good boundary detection accuracy. In the context of medical image segmentation, weak edges and inhomogeneities remain important issues that may hinder the accuracy of any segmentation method based on active contours implemented using level set methods. This paper proposes a method based on active contours implemented using level set methods for segmentation of such medical images. The proposed method uses a level set evolution that is based on the minimization of an objective energy functional whose energy terms are weighted according to their relative importance in detecting boundaries. This relative importance is computed based on local edge features collected from the adjacent region located inside and outside of the evolving contour. The local edge features employed are the edge intensity and the degree of alignment between the image's gradient vector flow field and the evolving contour's normal. We evaluate the proposed method for segmentation of various regions in real MRI and CT slices, X-ray images, and ultra sound images. Evaluation results confirm the advantage of weighting energy forces using local edge features to reduce leakage. These results also show that the proposed method leads to more accurate boundary detection results than state-of-the-art edge-based level set segmentation methods, particularly around weak edges.

A novel approach to segmentation and measurement of medical image using level set methods

The study proposes a novel approach for segmentation and visualization plus value-added surface area and volume measurements for brain medical image analysis. The proposed method contains edge detection and Bayesian based level set segmentation, surface and volume rendering, and surface area and volume measurements for 3D objects of interest (i.e., brain tumor, brain tissue, or whole brain). Two extensions based on edge detection and Bayesian level set are first used to segment 3D objects. Ray casting and a modified marching cubes algorithm are then adopted to facilitate volume and surface visualization of medical-image dataset. To provide physicians with more useful information for diagnosis, the surface area and volume of an examined 3D object are calculated by the techniques of linear algebra and surface integration. Experiment results are finally reported in terms of 3D object extraction, surface and volume rendering, and surface area and volume measurements for medical image analysis.

Semi-automatic outlining of levator hiatus

OBJECTIVE

To create a semi-automated outlining tool for the levator hiatus, to reduce interobserver variability and and speed up analysis.

METHODS

The proposed automated hiatus segmentation (AHS) algorithm takes a C-plane image, in the plane of minimal hiatal dimensions, and manually defined vertical hiatal limits as input. The AHS then creates an initial outline by fitting predefined templates on an intensity-invariant edge map, which is further refined using the B-spline explicit active surfaces framework. The AHS was tested using 91 representative C-plane images. Reference hiatal outlines were obtained manually and compared with the AHS outlines by three independent observers. The mean absolute distance (MAD), Hausdorff distance and Dice and Jaccard coefficients were used to quantify segmentation accuracy. Each of these metrics was calculated both for computer-observer differences (COD) and for interobserver differences. The Williams index was used to test the null hypothesis that the automated method would agree with the operators at least as well as the operators agreed with each other. Agreement between the two methods was assessed using the intraclass correlation coefficient (ICC) and Bland-Altman plots.

RESULTS

The AHS contours matched well with the manual ones (median COD, 2.10 (interquartile range (IQR), 1.54) mm for MAD). The Williams index was greater than or close to 1 for all quality metrics, indicating that the algorithm performed at least as well as did the manual references in terms of interrater variability. The interobserver differences using each of the metrics were significantly lower, and a higher ICC was achieved (0.93), when obtaining outlines using the AHS compared with manually. The Bland-Altman plots showed negligible bias between the two methods. Using the AHS took a median time of 7.07 (IQR, 3.49) s, while manual outlining took 21.31 (IQR, 5.43) s, thus being almost three-fold faster. Using the AHS, in general, the hiatus could be outlined completely using only three points, two for initialization and one for manual adjustment.

CONCLUSIONS

We present a method for tracing the levator hiatal outline with minimal user input. The AHS is fast, robust and reliable and improves interrater agreement. Copyright © 2015 ISUOG. Published by John Wiley & Sons Ltd.

Carotid ultrasound segmentation using radio-frequency derived phase information and gabor filters

Ultrasound image segmentation is a field which has garnered much interest over the years. This is partially due to the complexity of the problem, arising from the lack of contrast between different tissue types which is quite typical of ultrasound images. Recently, segmentation techniques which treat RF signal data have also become popular, particularly with the increasing availability of such data from open-architecture machines. It is believed that RF data provides a rich source of information whose integrity remains intact, as opposed to the loss which occurs through the signal processing chain leading to Brightness Mode Images. Furthermore, phase information contained within RF data has not been studied in much detail, as the nature of the information here appears to be mostly random. In this work however, we show that phase information derived from RF data does elicit structure, characterized by texture patterns. Texture segmentation of this data permits the extraction of rough, but well localized, carotid boundaries. We provide some initial quantitative results, which report the performance of the proposed technique.

Computer-Aided Segmentation of the Mid-Brain in Trans-Cranial Ultrasound Images

This paper presents a novel and rapid method developed for semi-automated segmentation of the mid-brain region in B-mode trans-cranial ultrasound (TCS) images. TCS is a relatively new neuroimaging tool having promising application in early diagnosis of Parkinson's disease. The quality of TCS images is much lower compared with the ultrasound images obtained during scanning of the soft tissues; the structures of interest in TCS are difficult to extract and to evaluate. The combination of an experience-based statistical shape model and intensity-amplitude invariant edge detector was proposed for the extraction of fuzzy boundaries of the mid-brain in TCS images. A statistical shape model was constructed using 90 manual delineations of the mid-brain region made by professional neurosonographer. Local phase-based edge detection strategy was applied for determination of plausible mid-brain boundary points used for statistical shape fitting. The proposed method was tested on other 40 clinical TCS images evaluated by two experts. The obtained averaged results of segmentation revealed that the differences between manual and automated measurements are statistically insignificant (p > 0.05).

Twofold processing for denoising ultrasound medical images

Ultrasound medical (US) imaging non-invasively pictures inside of a human body for disease diagnostics. Speckle noise attacks ultrasound images degrading their visual quality. A twofold processing algorithm is proposed in this work to reduce this multiplicative speckle noise. First fold used block based thresholding, both hard (BHT) and soft (BST), on pixels in wavelet domain with 8, 16, 32 and 64 non-overlapping block sizes. This first fold process is a better denoising method for reducing speckle and also inducing object of interest blurring. The second fold process initiates to restore object boundaries and texture with adaptive wavelet fusion. The degraded object restoration in block thresholded US image is carried through wavelet coefficient fusion of object in original US mage and block thresholded US image. Fusion rules and wavelet decomposition levels are made adaptive for each block using gradient histograms with normalized differential mean (NDF) to introduce highest level of contrast between the denoised pixels and the object pixels in the resultant image. Thus the proposed twofold methods are named as adaptive NDF block fusion with hard and soft thresholding (ANBF-HT and ANBF-ST). The results indicate visual quality improvement to an interesting level with the proposed twofold processing, where the first fold removes noise and second fold restores object properties. Peak signal to noise ratio (PSNR), normalized cross correlation coefficient (NCC), edge strength (ES), image quality Index (IQI) and structural similarity index (SSIM), measure the quantitative quality of the twofold processing technique. Validation of the proposed method is done by comparing with anisotropic diffusion (AD), total variational filtering (TVF) and empirical mode decomposition (EMD) for enhancement of US images. The US images are provided by AMMA hospital radiology labs at Vijayawada, India.

Feature-based fuzzy connectedness segmentation of ultrasound images with an object completion step

Medical ultrasound (US) image segmentation and quantification can be challenging due to signal dropouts, missing boundaries, and presence of speckle, which gives images of similar objects quite different appearance. Typically, purely intensity-based methods do not lead to a good segmentation of the structures of interest. Prior work has shown that local phase and feature asymmetry, derived from the monogenic signal, extract structural information from US images. This paper proposes a new US segmentation approach based on the fuzzy connectedness framework. The approach uses local phase and feature asymmetry to define a novel affinity function, which drives the segmentation algorithm, incorporates a shape-based object completion step, and regularises the result by mean curvature flow. To appreciate the accuracy and robustness of the methodology across clinical data of varying appearance and quality, a novel entropy-based quantitative image quality assessment of the different regions of interest is introduced. The new method is applied to 81 US images of the fetal arm acquired at multiple gestational ages, as a means to define a new automated image-based biomarker of fetal nutrition. Quantitative and qualitative evaluation shows that the segmentation method is comparable to manual delineations and robust across image qualities that are typical of clinical practice.

A robust and automatic method for human parasite egg recognition in microscopic images

With the accelerated movement of population, human parasitoses become an increasingly serious public health's problem. Currently, detections of parasite eggs through microscopic images are still the golden standard for diagnoses. However, this conventional method relies heavily on the experiences of inspectors, thus giving rise to misdiagnoses and missed diagnoses occasionally. And, as the number of clinical specimens increases rapidly, manual identification seems impractical. Hence, a fully automatic method is in desperate need. In this paper, we propose a robust method to segment and recognize the parasite eggs. Their contours are extracted using phase coherence technology, and the support vector machine (SVM) method based on shape and texture features is employed to classification of parasite eggs. Our novel method was comparable to the traditional method. The overall recognition rate was up to 95%, and the overall robustness indexes, including si, fnvf, fvpf, tpvf, were 95.7, 4.9, 3.7, 95.1, respectively, suggesting that our method is effective and the robustness is good, which has great potential to become a diagnostic method in the parasitological clinic.

Spatial and spatio-temporal feature extraction from 4D echocardiography images

BACKGROUND

Ultrasound images are difficult to segment because of their noisy and low contrast nature which makes it challenging to extract the important features. Typical intensity-gradient based approaches are not suitable for these low contrast images while it has been shown that the local phase based technique provides better results than intensity based methods for ultrasound images. The spatial feature extraction methods ignore the continuity in the heart cycle and may also capture spurious features. It is believed that the spurious features (noise) that are not consistent along the frames can be excluded by considering the temporal information.

METHODS

In this paper, we present a local phase based 4D (3D+time) feature asymmetry (FA) measure using the monogenic signal. We have investigated the spatio-temporal feature extraction to explore the effect of adding time information in the feature extraction process.

RESULTS

To evaluate the impact of time dimension, the results of 4D based feature extraction are compared with the results of 3D based feature extraction which shows the favorable 4D feature extraction results when temporal resolution is good. The paper compares the band-pass filters (difference of Gaussian, Cauchy and Gaussian derivative) in terms of their feature extraction performance. Moreover, the feature extraction is further evaluated quantitatively by left ventricle segmentation using the extracted features.

CONCLUSIONS

The results demonstrate that the spatio-temporal feature extraction is promising in frames with good temporal resolution.

A Split-and-Merge-Based Uterine Fibroid Ultrasound Image Segmentation Method in HIFU Therapy

High-intensity focused ultrasound (HIFU) therapy has been used to treat uterine fibroids widely and successfully. Uterine fibroid segmentation plays an important role in positioning the target region for HIFU therapy. Presently, it is completed by physicians manually, reducing the efficiency of therapy. Thus, computer-aided segmentation of uterine fibroids benefits the improvement of therapy efficiency. Recently, most computer-aided ultrasound segmentation methods have been based on the framework of contour evolution, such as snakes and level sets. These methods can achieve good performance, although they need an initial contour that influences segmentation results. It is difficult to obtain the initial contour automatically; thus, the initial contour is always obtained manually in many segmentation methods. A split-and-merge-based uterine fibroid segmentation method, which needs no initial contour to ensure less manual intervention, is proposed in this paper. The method first splits the image into many small homogeneous regions called superpixels. A new feature representation method based on texture histogram is employed to characterize each superpixel. Next, the superpixels are merged according to their similarities, which are measured by integrating their Quadratic-Chi texture histogram distances with their space adjacency. Multi-way Ncut is used as the merging criterion, and an adaptive scheme is incorporated to decrease manual intervention further. The method is implemented using Matlab on a personal computer (PC) platform with Intel Pentium Dual-Core CPU E5700. The method is validated on forty-two ultrasound images acquired from HIFU therapy. The average running time is 9.54 s. Statistical results showed that SI reaches a value as high as 87.58%, and normHD is 5.18% on average. It has been demonstrated that the proposed method is appropriate for segmentation of uterine fibroids in HIFU pre-treatment imaging and planning.

Adaptive Mesh Expansion Model (AMEM) for liver segmentation from CT image

This study proposes a novel adaptive mesh expansion model (AMEM) for liver segmentation from computed tomography images. The virtual deformable simplex model (DSM) is introduced to represent the mesh, in which the motion of each vertex can be easily manipulated. The balloon, edge, and gradient forces are combined with the binary image to construct the external force of the deformable model, which can rapidly drive the DSM to approach the target liver boundaries. Moreover, tangential and normal forces are combined with the gradient image to control the internal force, such that the DSM degree of smoothness can be precisely controlled. The triangular facet of the DSM is adaptively decomposed into smaller triangular components, which can significantly improve the segmentation accuracy of the irregularly sharp corners of the liver. The proposed method is evaluated on the basis of different criteria applied to 10 clinical data sets. Experiments demonstrate that the proposed AMEM algorithm is effective and robust and thus outperforms six other up-to-date algorithms. Moreover, AMEM can achieve a mean overlap error of 6.8% and a mean volume difference of 2.7%, whereas the average symmetric surface distance and the root mean square symmetric surface distance can reach 1.3 mm and 2.7 mm, respectively.

Segmentation of tumor ultrasound image in HIFU therapy based on texture and boundary encoding

It is crucial in high intensity focused ultrasound (HIFU) therapy to detect the tumor precisely with less manual intervention for enhancing the therapy efficiency. Ultrasound image segmentation becomes a difficult task due to signal attenuation, speckle effect and shadows. This paper presents an unsupervised approach based on texture and boundary encoding customized for ultrasound image segmentation in HIFU therapy. The approach oversegments the ultrasound image into some small regions, which are merged by using the principle of minimum description length (MDL) afterwards. Small regions belonging to the same tumor are clustered as they preserve similar texture features. The mergence is completed by obtaining the shortest coding length from encoding textures and boundaries of these regions in the clustering process. The tumor region is finally selected from merged regions by a proposed algorithm without manual interaction. The performance of the method is tested on 50 uterine fibroid ultrasound images from HIFU guiding transducers. The segmentations are compared with manual delineations to verify its feasibility. The quantitative evaluation with HIFU images shows that the mean true positive of the approach is 93.53%, the mean false positive is 4.06%, the mean similarity is 89.92%, the mean norm Hausdorff distance is 3.62% and the mean norm maximum average distance is 0.57%. The experiments validate that the proposed method can achieve favorable segmentation without manual initialization and effectively handle the poor quality of the ultrasound guidance image in HIFU therapy, which indicates that the approach is applicable in HIFU therapy.

Speckle reduction by phase-based weighted least squares

Although ultrasonography has been widely used in clinical applications, the doctor suffers great difficulties in diagnosis due to the artifacts of ultrasound images, especially the speckle noise. This paper proposes a novel framework for speckle reduction by using a phase-based weighted least squares optimization. The proposed approach can effectively smooth out speckle noise while preserving the features in the image, e.g., edges with different contrasts. To this end, we first employ a local phase-based measure, which is theoretically intensity-invariant, to extract the edge map from the input image. The edge map is then incorporated into the weighted least squares framework to supervise the optimization during despeckling, so that low contrast edges can be retained while the noise has been greatly removed. Experimental results in synthetic and clinical ultrasound images demonstrate that our approach performs better than state-of-the-art methods.

Curvelet based automatic segmentation of supraspinatus tendon from ultrasound image: a focused assistive diagnostic method

Background

Disorders of rotator cuff tendons results in acute pain limiting the normal range of motion for shoulder. Of all the tendons in rotator cuff, supraspinatus (SSP) tendon is affected first of any pathological changes. Diagnosis of SSP tendon using ultrasound is considered to be operator dependent with its accuracy being related to operator’s level of experience.

Methods

The automatic segmentation of SSP tendon ultrasound image was performed to provide focused and more accurate diagnosis. The image processing techniques were employed for automatic segmentation of SSP tendon. The image processing techniques combines curvelet transform and mathematical concepts of logical and morphological operators along with area filtering. The segmentation assessment was performed using true positives rate, false positives rate and also accuracy of segmentation. The specificity and sensitivity of the algorithm was tested for diagnosis of partial thickness tears (PTTs) and full thickness tears (FTTs). The ultrasound images of SSP tendon were taken from medical center with the help of experienced radiologists. The algorithm was tested on 116 images taken from 51 different patients.

Results

The accuracy of segmentation of SSP tendon was calculated to be 95.61% in accordance with the segmentation performed by radiologists, with true positives rate of 91.37% and false positives rate of 8.62%. The specificity and sensitivity was found to be 93.6%, 94% and 95%, 95.6% for partial thickness tears and full thickness tears respectively. The proposed methodology was successfully tested over a database of more than 116 US images, for which radiologist assessment and validation was performed.

Conclusions

The segmentation of SSP tendon from ultrasound images helps in focused, accurate and more reliable diagnosis which has been verified with the help of two experienced radiologists. The specificity and sensitivity for accurate detection of partial and full thickness tears has been considerably increased after segmentation when compared with existing literature.

Semiautomatic segmentation of aortic valve from sequenced ultrasound image using a novel shape-constraint GCV model

PURPOSE

Effective and accurate segmentation of the aortic valve (AV) from sequenced ultrasound (US) images remains a technical challenge because of intrinsic factors of ultrasound images that impact the quality and the continuous changes of shape and position of segmented objects. In this paper, a novel shape-constraint gradient Chan-Vese (GCV) model is proposed for segmenting the AV from time serial echocardiography.

METHODS

The GCV model is derived by incorporating the energy of the gradient vector flow into a CV model framework, where the gradient vector energy term is introduced by calculating the deviation angle between the inward normal force of the evolution contour and the gradient vector force. The flow force enlarges the capture range and enhances the blurred boundaries of objects. This is achieved by adding a circle-like contour (constructed using the AV structure region as a constraint shape) as an energy item to the GCV model through the shape comparison function. This shape-constrained energy can enhance the image constraint force by effectively connecting separate gaps of the object edge as well as driving the evolution contour to quickly approach the ideal object. Because of the slight movement of the AV in adjacent frames, the initial constraint shape is defined by users, with the other constraint shapes being derived from the segmentation results of adjacent sequence frames after morphological filtering. The AV is segmented from the US images by minimizing the proposed energy function.

RESULTS

To evaluate the performance of the proposed method, five assessment parameters were used to compare it with manual delineations performed by radiologists (gold standards). Three hundred and fifteen images acquired from nine groups were analyzed in the experiment. The area-metric overlap error rate was 6.89% ± 2.88%, the relative area difference rate 3.94% ± 2.63%, the average symmetric contour distance 1.08 ± 0.43 mm, the root mean square symmetric contour distance 1.37 ± 0.52 mm, and the maximum symmetric contour distance was 3.57 ± 1.72 mm.

CONCLUSIONS

Compared with the CV model, as a result of the combination of the gradient vector and neighborhood shape information, this semiautomatic segmentation method significantly improves the accuracy and robustness of AV segmentation, making it feasible for improved segmentation of aortic valves from US images that have fuzzy boundaries.

A multiresolution framework for ultrasound image segmentation by combinative active contours

We propose a novel multiresolution framework for ultrasound image segmentation in this paper. The framework exploits both local intensity and local phase information to tackle the degradations of ultrasound images. First, multiresolution scheme is adopted to build a Gaussian pyramid for each speckled image. Speckle noise is gradually smoothed out at higher levels of the pyramid. Then local intensity-driven active contours are employed to locate the coarse contour of the target from the coarsest image, followed by local phase-based geodesic active contours to further refine the contour in finer images. Compared with traditional gradient-based methods, phase-based methods are more suitable for ultrasound images because they are invariant to variations in image contrast. Experimental results on left ventricle segmentation from echocardiographic images demonstrate the advantages of the proposed model.

Phase-based probabilistic active contour for nerve detection in ultrasound images for regional anesthesia

Ultrasound guided regional anesthesia (UGRA) is steadily growing in popularity, owing to advances in ultrasound imaging technology and the advantages that this technique presents for safety and efficiency. The aim of this work is to assist anaesthetists during the UGRA procedure by automatically detecting the nerve blocks in the ultrasound images. The main disadvantage of ultrasound images is the poor quality of the images, which are also affected by the speckle noise. Moreover, the nerve structure is not salient amid the other tissues, which makes its detection a challenging problem. In this paper we propose a new method to tackle the problem of nerve zone detection in ultrasound images. The method consists in a combination of three approaches: probabilistic, edge phase information and active contours. The gradient vector flow (GVF) is adopted as an edge-based active contour. The phase analysis of the monogenic signal is used to provide reliable edges for the GVF. Then, a learned probabilistic model reduces the false positives and increases the likelihood energy term of the target region. It yields a new external force field that attracts the active contour toward the desired region of interest. The proposed scheme has been applied to sciatic nerve regions. The qualitative and quantitative evaluations show a high accuracy and a significant improvement in performance.