Search

Abdominal aortic aneurysms (AAAs) are rupture-prone dilatations of the aorta. In current clinical practice, the maximal diameter of AAAs is monitored with 2D ultrasound to estimate their rupture risk. Recent studies have shown that 3-dimensional and mechanical AAA parameters might be better predictors for aneurysm growth and rupture than the diameter. These parameters can be obtained with time-resolved 3D ultrasound (3D+t US), which requires robust and automatic segmentation of AAAs from 3D+t US. This study proposes and validates a deep learning (DL) approach for automatic segmentation of AAAs. 500 AAA patients were included for follow-up 3D+t US imaging, resulting in 2495 3D+t US images. Segmentation masks for model training were obtained using a conventional automatic segmentation algorithm ('nonDL'). Four different DL models were trained and validated by (1) comparison to CT and (2) reader scoring. Performance of the nonDL and different DL segmentation strategies were evaluated by comparing Hausdorff distance, Dice scores, accuracy, sensitivity, and specificity with a sign test. All DL models had higher median Dice scores, accuracy, and sensitivity (all p < 0.003) compared to nonDL segmentation. The full image-resolution model without data augmentation showed the highest median Dice score and sensitivity (p < 0.001). Applying the DL model on an independent test group produced fewer poor segmentation scores of 1 to 2 on a five-point scale (8% for DL, 18% for nonDL). This demonstrates that a robust and automatic segmentation algorithm for segmenting abdominal aortic aneurysms from 3D+t US images was developed, showing improved performance compared to conventional segmentation.

In multiaperture ultrasound, several ultrasound probes with different insonification angles are combined to increase the field of view and angular coverage of image structures. A full reconstruction incorporating all possible combinations of transmitting and receiving probes has been shown to improve resolution, contrast, and angular coverage beyond what can be achieved by the registration of single images from different probes. A major challenge in multiaperture imaging is the correct determination of relative probe locations. A registration based on the content of images from different probes is challenging due to the decorrelation of image structures and speckle with increasing angle between the probes. We propose a probe localization method for plane-wave ultrasound that uses solely the receive dataset of a nontransmitting probe. The localization is performed by signal tracking in the Radon domain. To demonstrate that the method does not rely on common structures in the individual images, we show that a satisfying localization can be performed in pure speckle for angles, where the speckle patterns have completely decorrelated. The method shows potential for real-time probe localization in free-hand multiprobe ultrasound imaging or for flexible and wearable multiarray combination of multiple capacitive micromachined (CMUT)-based systems in the future.

In the last decade, many advances have been made in high frame rate 3-D ultrasound imaging, including more flexible acquisition systems, transmit (TX) sequences, and transducer arrays. Compounding multiangle transmits of diverging waves has shown to be fast and effective for 2-D matrix arrays, where heterogeneity between transmits is key in optimizing the image quality. However, the anisotropy in contrast and resolution remains a drawback that cannot be overcome with a single transducer. In this study, a bistatic imaging aperture is demonstrated that consists of two synchronized matrix ( 32×32 ) arrays, allowing for fast interleaved transmits with a simultaneous receive (RX). First, for a single array, the aperture efficiency for high volume rate imaging was evaluated between sparse random arrays and fully multiplexed arrays. Second, the performance of the bistatic acquisition scheme was analyzed for various positions on a wire phantom and was showcased in a dynamic setup mimicking the human abdomen and aorta. Sparse array volume images were equal in resolution and lower in contrast compared to fully multiplexed arrays but can efficiently minimize decorrelation during motion for multiaperture imaging. The dual-array imaging aperture improved the spatial resolution in the direction of the second transducer, reducing the average volumetric speckle size with 72% and the axial-lateral eccentricity with 8%. In the aorta phantom, the angular coverage increased by a factor of 3 in the axial-lateral plane, raising the wall-lumen contrast with 16% compared to single-array images, despite accumulation of thermal noise in the lumen.

Lightweight segmentation models are becoming more popular for fast diagnosis on small and low cost medical imaging devices. This study focuses on the segmentation of the left ventricle (LV) in cardiac ultrasound (US) images. A new lightweight model [LV network (LVNet)] is proposed for segmentation, which gives the benefits of requiring fewer parameters but with improved segmentation performance in terms of Dice score (DS). The proposed model is compared with state-of-the-art methods, such as UNet, MiniNetV2, and fully convolutional dense dilated network (FCdDN). The model proposed comes with a post-processing pipeline that further enhances the segmentation results. In general, the training is done directly using the segmentation mask as the output and the US image as the input of the model. A new strategy for segmentation is also introduced in addition to the direct training method used. Compared with the UNet model, an improvement in DS performance as high as 5% for segmentation with papillary (WP) muscles was found, while showcasing an improvement of 18.5% when the papillary muscles are excluded. The model proposed requires only 5% of the memory required by a UNet model. LVNet achieves a better trade-off between the number of parameters and its segmentation performance as compared with other conventional models. The developed codes are available at https://github.com/navchetanawasthi/Left&#95;Ventricle&#95;Segmentation.

Knowledge of aneurysm geometry and local mechanical wall parameters using ultrasound (US) can contribute to a better prediction of rupture risk in abdominal aortic aneurysms (AAAs). However, aortic strain imaging using conventional US is limited by the lateral lumen-wall contrast and resolution. In this study, ultrafast multiperspective bistatic (MP BS) imaging is used to improve aortic US, in which two curved array transducers receive simultaneously on each transmit event. The advantage of such bistatic US imaging on both image quality and strain estimations was investigated by comparing it to single-perspective monostatic (SP MS) and MP monostatic (MP MS) imaging, i.e., alternately transmitting and receiving with either transducer. Experimental strain imaging was performed in US simulations and in an experimental study on porcine aortas. Different compounding strategies were tested to retrieve the most useful information from each received US signal. Finally, apart from the conventional sector grid in curved array US imaging, a polar grid with respect to the vessel's local coordinate system is introduced. This new reconstruction method demonstrated improved displacement estimations in aortic US. The US simulations showed increased strain estimation accuracy using MP BS imaging bistatic imaging compared to MP MS imaging, with a decrease in the average relative error between 41% and 84% in vessel wall regions between transducers. In the experimental results, the mean image contrast-to-noise ratio was improved by up to 8 dB in the vessel wall regions between transducers. This resulted in an increased mean elastographic signal-to-noise ratio by about 15 dB in radial strain and 6 dB in circumferential strain.

Accurate 3-D geometries of arteries and veins are important clinical data for diagnosis of arterial disease and intervention planning. Automatic segmentation of vessels in the transverse view suffers from the low lateral resolution and contrast. Convolutional neural networks are a promising tool for automatic segmentation of medical images, outperforming the traditional segmentation methods with high robustness. In this study, we aim to create a general, robust, and accurate method to segment the lumen-wall boundary of healthy central and peripheral vessels in large field-of-view freehand ultrasound (US) datasets. Data were acquired using the freehand US, in combination with a probe tracker. A total of ±36 000 cross-sectional images, acquired in the common, internal, and external carotid artery ( N = 37 ), in the radial, ulnar artery, and cephalic vein ( N = 12 ), and in the femoral artery ( N = 5 ) were included. To create masks (of the lumen) for training data, a conventional automatic segmentation method was used. The neural networks were trained on: 1) data of all vessels and 2) the carotid artery only. The performance was compared and tested using an open-access dataset. The recall, precision, DICE, and intersection over union (IoU) were calculated. Overall, segmentation was successful in the carotid and peripheral arteries. The Multires U-net architecture performs best overall with DICE = 0.93 when trained on the total dataset. Future studies will focus on the inclusion of vascular pathologies.

Near-field (NF) clutter in echocardiography is depicted as a diffuse haze hindering the visualization of the myocardium and the blood-pool, thereby degrading its diagnostic value. Several clutter filters have been developed, which are limited in patients with contraction motion and rhythm anomalies, and in 3-D ultrasound (US). This study introduces a new NF clutter reduction method, which preserves US speckles required for strain imaging. The filter developed detects the NF clutter region in the spatial frequency domain. The filter employs an oriented, multiscale approach, and assumes the NF clutter to be predominantly present in the highest and lowest bandpass images. These bandpass images were filtered, whilst sparing features in the myocardium and NF clutter-free regions. The performance of the filter was assessed in a volunteer study, in ten 3-D apical and parasternal view acquisitions, and in a retrospective clinical study composed of 20 cardiac patients with different indications for echocardiography. The filter reduced NF clutter in all data sets, whilst preserving all or most of the myocardium. Additionally, it demonstrated a consistent enhancement of image quality, with an increase in contrast of 4.3 dB on average, and generated a clearer myocardial boundary distinction. Furthermore, the speckles were preserved according to the quality index based on local variance, the structural similarity index method, and normalized cross correlation values, being 0.82, 0.92, and 0.95 on average, respectively. Global longitudinal strain measurements on NF clutter reduced images were improved or equivalent compared to the original acquisitions, with an average increase in strain signal-to-noise ratio of 34%.

Photoacoustic imaging (PAI) is a promising imaging modality due to its high optical specificity. However, the low signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of in vivo PA images are major challenges that prevent PAI from finding its place in clinics. This paper investigates the merit of spatial compounding of PA images in arterial phantoms and the achievable improvements of SNR, when in vivo conditions are mimicked. The analysis of the compounding technique was performed on a polyvinyl alcohol vessel phantom with black threads embedded in its wall. The in vivo conditions were mimicked by limiting the rotation range in ±30°, adding turbid surrounding medium, and filling the lumen with porcine blood. Finally, the performance of the technique was evaluated in ex vivo human carotid plaque samples. Results showed that spatial compounding elevates the SNR by 5-10 dB and CNR by 1-5 dB, depending on the location of the absorbers. This paper elucidates prospective in vivo PA characterization of carotid plaques by proposing a method to enhance PA image quality.