Jinbum Kang

A New Feature-Enhanced Speckle Reduction Method Based on Multiscale Analysis for Ultrasound B-Mode Imaging

Goal: Effective speckle reduction in ultrasound B-mode imaging is important for enhancing the image quality and improving the accuracy in image analysis and interpretation. In this paper, a new feature-enhanced speckle reduction (FESR) method based on multiscale analysis and feature enhancement filtering is proposed for ultrasound B-mode imaging. In FESR, clinical features (e.g., boundaries and borders of lesions) are selectively emphasized by edge, coherence, and contrast enhancement filtering from fine to coarse scales while simultaneously suppressing speckle development via robust diffusion filtering. In the simulation study, the proposed FESR method showed statistically significant improvements in edge preservation, mean structure similarity, speckle signal-to-noise ratio, and contrast-to-noise ratio (CNR) compared with other speckle reduction methods, e.g., oriented speckle reducing anisotropic diffusion (OSRAD), nonlinear multiscale wavelet diffusion (NMWD), the Laplacian pyramid-based nonlinear diffusion and shock filter (LPNDSF), and the Bayesian nonlocal means filter (OBNLM). Similarly, the FESR method outperformed the OSRAD, NMWD, LPNDSF, and OBNLM methods in terms of CNR, i.e., 10.70 ± 0.06 versus 9.00 ± 0.06, 9.78 ± 0.06, 8.67 ± 0.04, and 9.22 ± 0.06 in the phantom study, respectively. Reconstructed B-mode images that were developed using the five speckle reduction methods were reviewed by three radiologists for evaluation based on each radiologists diagnostic preferences. All three radiologists showed a significant preference for the abdominal liver images obtained using the FESR methods in terms of conspicuity, margin sharpness, artificiality, and contrast, p<;0.0001. For the kidney and thyroid images, the FESR method showed similar improvement over other methods. However, the FESR method did not show statistically significant improvement compared with the OBNLM method in margin sharpness for the kidney and thyroid images. These results demonstrate that the proposed FESR method can improve the image quality of ultrasound B-mode imaging by enhancing the visualization of lesion features while effectively suppressing speckle noise.

Automatic dynamic range adjustment for ultrasound B-mode imaging.

In medical ultrasound imaging, dynamic range (DR) is defined as the difference between the maximum and minimum values of the displayed signal to display and it is one of the most essential parameters that determine its image quality. Typically, DR is given with a fixed value and adjusted manually by operators, which leads to low clinical productivity and high user dependency. Furthermore, in 3D ultrasound imaging, DR values are unable to be adjusted during 3D data acquisition. A histogram matching method, which equalizes the histogram of an input image based on that from a reference image, can be applied to determine the DR value. However, it could be lead to an over contrasted image. In this paper, a new Automatic Dynamic Range Adjustment (ADRA) method is presented that adaptively adjusts the DR value by manipulating input images similar to a reference image. The proposed ADRA method uses the distance ratio between the log average and each extreme value of a reference image. To evaluate the performance of the ADRA method, the similarity between the reference and input images was measured by computing a correlation coefficient (CC). In in vivo experiments, the CC values were increased by applying the ADRA method from 0.6872 to 0.9870 and from 0.9274 to 0.9939 for kidney and liver data, respectively, compared to the fixed DR case. In addition, the proposed ADRA method showed to outperform the histogram matching method with in vivo liver and kidney data. When using 3D abdominal data with 70 frames, while the CC value from the ADRA method is slightly increased (i.e., 0.6%), the proposed method showed improved image quality in the c-plane compared to its fixed counterpart, which suffered from a shadow artifact. These results indicate that the proposed method can enhance image quality in 2D and 3D ultrasound B-mode imaging by improving the similarity between the reference and input images while eliminating unnecessary manual interaction by the user.

Automatic dynamic range optimization for 3D medical ultrasound imaging

In ultrasound B-mode imaging, dynamic range (DR) is typically given with a fixed value, and operators should manually adjust it. However, in 3D ultrasound imaging, it is difficult to adjust DR during 3D data acquisition, leading to low image quality. In this paper, the automatic DR adjustment (ADRA) method was applied to optimize the DR value in 3D ultrasound imaging. In ADRA, the log average of a reference image is computed and the distance ratio from its log average to each extreme value is obtained. The distance ratio is used for adjusting the DR value of an input image to have similar contrast. When the ADRA method is applied to 3D ultrasound imaging, its first frame can be used as a reference image so that consecutive frames during 3D data acquisition can have comparable contrast. To evaluate the performance of the ADRA method, 70 frames of complex baseband in vivo abdominal data were acquired with a commercial ultrasound system by moving a 3.5-MHz convex array probe from the left lobe of the liver to the right kidney of a volunteer. The proposed ADRA method provides more uniform contrast in the c-plane image over the fixed DR method that suffers from dark lines. The consistent results were obtained with the measured correlation coefficient (CC) values. The proposed ADRA method provided the CC value of 0.9669 ± 0.0199 vs. 0.9611 ± 0.0308 and 0.6768 ± 0.3594, compared to the fixed DR and histogram matching methods, respectively. These results indicate that the proposed ADRA method can automatically adjust DR of the consecutive frames during 3D data acquisition to get uniform contrast, leading to the improvement in image quality of 3D ultrasound imaging.

High PRF ultrafast sliding compound doppler imaging: fully qualitative and quantitative analysis of blood flow

Ultrafast compound Doppler imaging based on plane-wave excitation (UCDI) can be used to evaluate cardiovascular diseases using high frame rates. In particular, it provides a fully quantifiable flow analysis over a large region of interest with high spatio-temporal resolution. However, the pulse-repetition frequency (PRF) in the UCDI method is limited for high-velocity flow imaging since it has a tradeoff between the number of plane-wave angles (N) and acquisition time. In this paper, we present high PRF ultrafast sliding compound Doppler imaging method (HUSDI) to improve quantitative flow analysis. With the HUSDI method, full scanline images (i.e. each tilted plane wave data) in a Doppler frame buffer are consecutively summed using a sliding window to create high-quality ensemble data so that there is no reduction in frame rate and flow sensitivity. In addition, by updating a new compounding set with a certain time difference (i.e. sliding window step size or L), the HUSDI method allows various Doppler PRFs with the same acquisition data to enable a fully qualitative, retrospective flow assessment. To evaluate the performance of the proposed HUSDI method, simulation, in vitro and in vivo studies were conducted under diverse flow circumstances. In the simulation and in vitro studies, the HUSDI method showed improved hemodynamic representations without reducing either temporal resolution or sensitivity compared to the UCDI method. For the quantitative analysis, the root mean squared velocity error (RMSVE) was measured using 9 angles (-12° to 12°) with L of 1-9, and the results were found to be comparable to those of the UCDI method (L  =  N  =  9), i.e.  ⩽0.24 cm s-1, for all L values. For the in vivo study, the flow data acquired from a full cardiac cycle of the femoral vessels of a healthy volunteer were analyzed using a PW spectrogram, and arterial and venous flows were successfully assessed with high Doppler PRF (e.g. 5 kHz at L  =  4). These results indicate that the proposed HUSDI method can improve flow visualization and quantification with a higher frame rate, PRF and flow sensitivity in cardiovascular imaging.

A wide field-of-view microvascular imaging using diverging transmit beams in a curved-array transducer

Ultrafast compound Doppler imaging enables high-sensitivity microvascular imaging using the large amount of spatial and temporal samples within a short acquisition time. In addition, the longer ensemble length from the ultrafast transmit strategy facilitates efficient clutter rejection filtering. However, the ultrafast microvascular imaging using plane-wave excitation, which is widely used in linear array transducer, may not be suitable for abdominal applications such as hepatic or kidney vessels due to limited field-of-view (FOV) and imaging depth. In this paper, we presents a wide FOV microvascular imaging method using diverging transmit beams in a curved array transducer.

A new high definition voxel-based multi-planar reformatting method in 3-D ultrasound imaging

Multi-planar reformatting (MPR) in 3-D ultrasound imaging provides the viewer with planar cross-sectional images extracted from the 3-D data. To produce the MPR imaging from the acquired 3-D volume data, the 3-D scan conversion (SC), which transforms the acquired data in the 3-D polar coordinate system to the 3-D Cartesian coordinate system, is performed for display. Direct 3-D SC and separable 3-D SC method are most widely used approaches that they are followed by interpolation (e.g., trilinear or bilinear) using neighboring pixels. However, the direct 3-D SC and separable 3-D SC methods still suffer from blurring artifact; it may cause the deterioration of image quality. In this paper, we presents a new high definition MPR method using voxel based beamforming for reducing blurring artifacts.

A new multi-planar reconstruction method using voxel based beamforming for 3D ultrasound imaging

For multi-planar reconstruction in 3D ultrasound imaging, direct and separable 3D scan conversion (SC) have been used for transforming the ultrasound data acquired in the 3D polar coordinate system to the 3D Cartesian coordinate system. These 3D SC methods can visualize an arbitrary plane for 3D ultrasound volume data. However, they suffer from blurring and blocking artifacts due to resampling during SC. In this paper, a new multi-planar reconstruction method based on voxel based beamforming (VBF) is proposed for reducing blurring and blocking artifacts. In VBF, unlike direct and separable 3D SC, each voxel on an arbitrary imaging plane is directly reconstructed by applying the focusing delay to radio-frequency (RF) data so that the blurring and blocking artifacts can be removed. From the phantom study, the proposed VBF method showed the higher contrast and less blurring compared to the separable and direct 3D SC methods. This result is consistent with the measured information entropy contrast (IEC) values, i.e., 98.9 vs. 42.0 vs. 47.9, respectively. In addition, the 3D SC methods and VBF method were implemented on a high-end GPU by using CUDA programming. The execution times for the VBF and direct 3D SC methods are 1656.1ms, 1633.3ms and 1631.4ms, which are I/O bounded. These results indicate that the proposed VBF method can improve image quality of 3D ultrasound B-mode imaging by removing blurring and blocking artifacts associated with 3D scan conversion and show the feasibility of pseudo-real-time operation.

Phantom and in vivo evaluation of sound speed estimation methods: Preliminary results

For medical ultrasound imaging, the dynamic receive beamforming is important for improving image quality, i.e., spatial and contrast resolution. In current dynamic receive beamforming, a constant sound speed (e.g., 1540m/s) is assumed. However, the sound speed dispersed in soft tissues leads to defocusing and degradation of image quality. Various methods have been proposed to estimate the proper sound speed with received data, but these methods have not been verified their performance in clinical cases (e.g., breast tissue). In this paper, the five different sound speed estimation methods (i.e., coherent factor (CF), minimum average phase variance (MAPV), minimum average sum-of-absolute difference (MASAD), focus quality spectra (FQS), and modified nonlinear anisotropic difference (MNAD)) are evaluated with the tissue mimicking phantom and the in vivo breast data under the same condition. The pre-beamformed radio-frequency data (RF) for the tissue mimicking phantom and in vivo breast data are acquired using a 7.5-MHz linear array transducer with the SonixTouch research platform connected to the SonixDAQ parallel data acquisition system. In the phantom study, the five methods show considerable performance in estimating the optimal sound speed (i.e., 1450 ± 25 m/s). The CF and FQS methods also show the low errors in the in vivo breast study, but the MAPV, MASAD and MNAD methods have difficulty in estimating the optimal sound speed (1530 m/s) i.e., 25.0 ± 12.9 and 20.0 ± 8.2 vs. 72.5 ± 45.0, 72.5 ± 41.9, 52.5 ± 28.7, respectively. These results indicated that the CF and FQS methods can robustly estimate the optimal sound speed in the homogenous phantom and heterogeneous soft tissues (e.g., breast).

A new feature-enhanced speckle reduction method based on multiscale analysis and synthesis for ultrasound B-mode imaging

Effective speckle reduction in ultrasound B-mode imaging is important for improving image quality and the accuracy in image analysis. While multiscale analysis-based speckle reduction methods such as Laplacian pyramid nonlinear diffusion (LPND) and nonlinear multiscale wavelet diffusion (NMWD) showed enhanced speckle reduction, they suffer from excessive blurring and artificial appearance. In this paper, a new feature-enhanced speckle reduction (FESR) method based on multiscale analysis and feature enhancement filtering is presented for ultrasound B-mode imaging. To separate true clinical features (e.g., boundaries of lesions) from noise, the subband images from a Laplacian pyramid model are firstly generated. Then, a robust anisotropic diffusion process is applied to suppress the identified noise and the extracted features are selectively emphasized by suitable edge, coherence and contrast enhancement filtering from fine to coarse scales. The performance of the proposed FESR method was compared with the LPND and NMWD methods by measuring speckles signal-to-noise ratio (SSNR) and contrast-to-noise ratio (CNR). With the FESR method, the mean SSNR value and the mean CNR value are significantly higher compared to the LPND and NMWD methods, i.e., 8.06±0.74 vs. 5.69±0.48, 7.14±0.93 and 6.89 ± 0.68 vs. 5.08 ± 0.33, 6.01 ± 0.53, respectively. These preliminary results demonstrates that the proposed FESR method can improve the image quality of ultrasound B-mode imaging by enhancing the visualization of borders and boundaries of lesions while effectively suppressing speckle.

A real-time realization of an automatic dynamic range adjustment method on a smart mobile device for point-of-care ultrasound imaging

Dynamic range (DR) is one of the most essential parameters that determine the image quality and feature representation in ultrasound B-mode imaging. Thus, the automatic DR adjustment (ADRA) method was previously proposed for reducing user dependency and improving image quality. In this paper, for a smart phone based point-of-care ultrasound imaging system, the real-time realization of the ADRA method on a smart phone using a mobile GPU is presented. The ADRA method was implemented on the NEXUS 4 running on the Android 4.3 by utilizing the several graphics pipelines supported from the OpenGL ES platform. For evaluating the real-time ADRA method, 200 frames of in vivo abdominal data were acquired by a 3.5-MHz convex array transducer from with a commercial ultrasound scanner equipped with a research package (Accuvix V10, Samsung Medison, Seoul, Korea). The GPU-based ADRA method showed the enhanced contrast-to-noise ratio (CNR) compared to the conventional method with a fixed DR value, i.e., 3.89±0.71 vs. 3.35±0.88, respectively. In addition, it showed a higher frame rate (i.e., >60 frames/sec) compared to the CPU-based ADRA method while providing the comparable CNR value. These results indicate that the ADRA method can be successfully supported in the smartphone while improving the image quality with less user dependency.