A bimodal BI-RADS-guided GoogLeNet-based CAD system for solid breast masses discrimination using transfer learning.

Numerous solid breast masses require sophisticated analysis to establish a differential diagnosis. Consequently, complementary modalities such as ultrasound imaging are frequently required to evaluate mammographically further detected masses. Radiologists mentally integrate complementary information from images acquired of the same patient to make a more conclusive and effective diagnosis. However, it has always been a challenging task. This paper details a novel bimodal GoogLeNet-based CAD system that addresses the challenges associated with combining information from mammographic and sonographic images for solid breast mass classification. Each modality is initially trained using two distinct monomodal models in the proposed framework. Then, using the high-level feature maps extracted from both modalities, a bimodal model is trained. In order to fully exploit the BI-RADS descriptors, different image content representations of each mass are obtained and used as input images. In addition, using an ImageNet pre-trained GoogLeNet model, two publicly available databases, and our collected dataset, a two-step transfer learning strategy has been proposed. Our bimodal model achieves the best recognition results in terms of sensitivity, specificity, F1-score, Matthews Correlation Coefficient, area under the receiver operating characteristic curve, and accuracy metrics of 90.91%, 89.87%, 90.32%, 80.78%, 95.82%, and 90.38%, respectively. The promising results indicate that the proposed CAD system can facilitate bimodal suspicious mass analysis and thus contribute significantly to improving breast cancer diagnostic performance.

A novel adaptive apodization to improve the resolution of phased subarray imaging in medical ultrasound.

PURPOSE: Phased subarray imaging (PSA) was previously proposed to extend the receive aperture length. Using overlapped subarrays as transmitters in PSA leads to decrement of sidelobe levels of the overall beam compared to full phased array imaging (PHA). This paper proposes an adaptive compounding of subarray images in PSA to improve both the resolution and contrast compared with PHA. METHOD: Adaptive apodization (ADAP) is defined proportional to the beamformed responses of subarrays such that the overall energy after compounding is minimized. RESULTS: The simulation and experimental results validate the performance of applying ADAP in PSA. The full width at half maximum (FWHM) at a depth of 30 mm in the proposed PSA is about 0.2 mm, compared to a FWHM of 0.6 mm with PHA imaging. Measuring the contrast ratio index shows that the ADAP method also improves the contrast in PSA imaging at least 25% compared to PHA imaging. CONCLUSION: Applying the proposed ADAP, besides conventional compounding in PSA imaging, leads to improvement of both the resolution and contrast compared to PHA imaging.

Linear-array photoacoustic imaging using minimum variance-based delay multiply and sum adaptive beamforming algorithm.

In photoacoustic imaging, delay-and-sum (DAS) beamformer is a common beamforming algorithm having a simple implementation. However, it results in a poor resolution and high sidelobes. To address these challenges, a new algorithm namely delay-multiply-and-sum (DMAS) was introduced having lower sidelobes compared to DAS. To improve the resolution of DMAS, a beamformer is introduced using minimum variance (MV) adaptive beamforming combined with DMAS, so-called minimum variance-based DMAS (MVB-DMAS). It is shown that expanding the DMAS equation results in multiple terms representing a DAS algebra. It is proposed to use the MV adaptive beamformer instead of the existing DAS. MVB-DMAS is evaluated numerically and experimentally. In particular, at the depth of 45 mm MVB-DMAS results in about 31, 18, and 8 dB sidelobes reduction compared to DAS, MV, and DMAS, respectively. The quantitative results of the simulations show that MVB-DMAS leads to improvement in full-width-half-maximum about 96%, 94%, and 45% and signal-to-noise ratio about 89%, 15%, and 35% compared to DAS, DMAS, MV, respectively. In particular, at the depth of 33 mm of the experimental images, MVB-DMAS results in about 20 dB sidelobes reduction in comparison with other beamformers.

Double-Stage Delay Multiply and Sum Beamforming Algorithm Applied to Ultrasound Medical Imaging.

In ultrasound (US) imaging, delay and sum (DAS) is the most common beamformer, but it leads to low-quality images. Delay multiply and sum (DMAS) was introduced to address this problem. However, the reconstructed images using DMAS still suffer from the level of side lobes and low noise suppression. Here, a novel beamforming algorithm is introduced based on expansion of the DMAS formula. We found that there is a DAS algebra inside the expansion, and we proposed use of the DMAS instead of the DAS algebra. The introduced method, namely double-stage DMAS (DS-DMAS), is evaluated numerically and experimentally. The quantitative results indicate that DS-DMAS results in an approximately 25% lower level of side lobes compared with DMAS. Moreover, the introduced method leads to 23%, 22% and 43% improvement in signal-to-noise ratio, full width at half-maximum and contrast ratio, respectively, compared with the DMAS beamformer.

Double-Stage Delay Multiply and Sum Beamforming Algorithm: Application to Linear-Array Photoacoustic Imaging.

Photoacoustic imaging (PAI) is an emerging medical imaging modality capable of providing high spatial resolution of Ultrasound (US) imaging and high contrast of optical imaging. Delay-and-Sum (DAS) is the most common beamforming algorithm in PAI. However, using DAS beamformer leads to low resolution images and considerable contribution of off-axis signals. A new paradigm namely delay-multiply-and-sum (DMAS), which was originally used as a reconstruction algorithm in confocal microwave imaging, was introduced to overcome the challenges in DAS. DMAS was used in PAI systems and it was shown that this algorithm results in resolution improvement and sidelobe degrading. However, DMAS is still sensitive to high levels of noise, and resolution improvement is not satisfying. Here, we propose a novel algorithm based on DAS algebra inside DMAS formula expansion, double stage DMAS (DS-DMAS), which improves the image resolution and levels of sidelobe, and is much less sensitive to high level of noise compared to DMAS. The performance of DS-DMAS algorithm is evaluated numerically and experimentally. The resulted images are evaluated qualitatively and quantitatively using established quality metrics including signal-to-noise ratio (SNR), full-width-half-maximum (FWHM) and contrast ratio (CR). It is shown that DS-DMAS outperforms DAS and DMAS at the expense of higher computational load. DS-DMAS reduces the lateral valley for about 15 dB and improves the SNR and FWHM better than 13% and 30%, respectively. Moreover, the levels of sidelobe are reduced for about 10 dB in comparison with those in DMAS.

Improving beamforming performance by phased synthetic aperture imaging in medical ultrasound.

PURPOSE: Ultrasound image quality is related to the receive beamformer's ability. Delay and sum (DAS), a conventional beamformer, is combined with the coherence factor (CF) technique to suppress side lobe levels. The purpose of this study is to improve these beamformer's abilities. METHODS: It has been shown that extension of the receive aperture can improve the receive beamformer's ability in radar studies. This paper shows that the focusing quality of CF and CF+DAS in medical ultrasound can be increased by extension of the receive aperture's length in phased synthetic aperture (PSA) imaging. RESULTS: The 3-dB width of the main lobe in the receive beam related to CF focusing decreased to 0.55 mm using the proposed PSA compared to the conventional phased array (PHA) imaging, whose FWHM is about 0.9 mm. The clutter-to-total-energy ratio (CTR) represented by R20 dB showed an improvement of 50 and 33% for CF and CF+DAS beamformers, respectively, with PSA as compared to PHA. In addition, simulation results validated the effectiveness of PSA versus PHA. CONCLUSION: In applications where there are no important limitations on the SNR, PSA imaging is recommended as it increases the ability of the receive beamformer for better focusing.

Synthetic Aperture Ultrasound Fourier Beamformation Using Virtual Sources.

An efficient Fourier beamformation algorithm is presented for multistatic synthetic aperture ultrasound imaging using virtual sources. The concept is based on the frequency domain wavenumber algorithm from radar and sonar and is extended to a multielement transmit/receive configuration using virtual sources. Window functions are used to extract the azimuth processing bandwidths and weight the data to reduce side lobes in the final image. Field II simulated data and SARUS (Synthetic Aperture Real-time Ultrasound System) measured data are used to evaluate the results in terms of point spread function, resolution, contrast, signal-to-noise ratio, and processing time. Lateral resolutions of 0.53 and 0.66 mm are obtained for Fourier Beamformation Using Virtual Sources (FBV) and delay and sum (DAS) on point target simulated data. Corresponding axial resolutions are 0.21 mm for FBV and 0.20 mm for DAS. The results are also consistent over different depths evaluated using a simulated phantom containing several point targets at different depths. FBV shows a better lateral resolution at all depths, and the axial and cystic resolutions of -6, -12, and -20 dB are almost the same for FBV and DAS. To evaluate the cyst phantom metrics, three different criteria of power ratio, contrast ratio, and contrast-to-noise ratio have been used. Results show that the algorithms have a different performance in the cyst center and near the boundary. FBV has a better performance near the boundary; however, DAS is better in the more central area of the cyst. Measured data from phantoms are also used for evaluation. The results confirm applicability of FBV in ultrasound, and 20 times less processing time is attained in comparison with DAS. Evaluating the results over a wide variety of parameters and having almost the same results for simulated and measured data demonstrates the ability of FBV in preserving the quality of image as DAS, while providing a more efficient algorithm with 20 times less computations.

Computational Complexity Reduction of Synthetic-aperture Focus in Ultrasound Imaging Using Frequency-domain Reconstruction.

A new frequency-domain implementation of a synthetic aperture focusing technique is presented in the paper. The concept is based on synthetic aperture radar (SAR) and sonar that is a developed version of the convolution model in the frequency domain. Compared with conventional line-by-line imaging, synthetic aperture imaging has a better resolution and contrast at the cost of more computational load. To overcome this problem, point-by-point reconstruction methods have been replaced by block-processing algorithms in radar and sonar; however, these techniques are relatively unknown in medical imaging. In this paper, we extended one of these methods called wavenumber to medical ultrasound imaging using a simple model of synthetic aperture focus. The model, derived here for monostatic mode, can be generalized to multistatic as well. The method consists of 4 steps: a 2D fast Fourier transform of the data, frequency shift of the data to baseband, interpolation to convert polar coordinates to rectangular ones, and returning the data to the spatial-domain using a 2D inverse Fourier transform. We have also used chirp pulse excitation followed by matched filtering and spotlighting algorithm to compensate the effect of differences in parameters between radar and medical imaging. Computational complexities of the two methods, wavenumber and delay-and-sum (DAS), have been calculated. Field II simulated point data have been used to evaluate the results in terms of resolution and contrast. Evaluations with simulated data show that for typical phantoms, reconstruction by the wavenumber algorithm is almost 20 times faster than classical DAS while retaining the resolution.

Weighted Capon beamformer combined with coded excitation in ultrasound imaging.

BACKGROUND: Capon-based beamformers are well-known methods to improve the SNR and quality of medical ultrasound images. Furthermore, they can improve the resolution of the images unexpectedly more than conventional DAS beamformers. Another method used in radar, sonar, and ultrasound imaging to increase the SNR is coded excitation with linear frequency-modulated signal (chirp). METHOD: In this paper, we illustrate that the combination of coded excitation with a Capon beamformer provides better resolution and higher SNR. However, this combination suffers from high sidelobe levels. We propose a weighted Capon beamformer (WCB) with chirp excitation to suppress the sidelobes and obtain a higher contrast with approximately the same resolution as the standard Capon beamformer (SCB). The weights of the WCB are obtained using element-wise multiplication of the weights of the SCB and a desired window (Hanning) with the same dimension. RESULTS: The results show a 20 dB reduction in the sidelobe levels in simulated point targets and a 6 dB increase in background contrast in simulated cyst phantoms.

Disorganization of Equilibrium Directional Interactions in the Brain Motor Network of Parkinson's disease: New Insight of Resting State Analysis Using Granger Causality and Graphical Approach.

Parkinson's disease (PD) is a progressive neurological disorder characterized by tremor, rigidity, and slowness of movements. Particular changes related to various pathological attacks in PD could result in causal interactions of the brain network from resting state functional magnetic resonance imaging (rs-fMRI) data. In this paper, we aimed to disclose the network structure of the directed influences over the brain using multivariate Granger causality analysis and graph theory in patients with PD as compared with control group. rs-fMRI at rest from 10 PD patients and 10 controls were analyzed. Topological properties of the networks showed that information flow in PD is smaller than that in healthy individuals. We found that there is a balanced local network in healthy control group, including positive pair-wise cross connections between caudate and cerebellum and reciprocal connections between motor cortex and caudate in the left and right hemispheres. The results showed that this local network is disrupted in PD due to disturbance of the interactions in the motor networks. These findings suggested alteration of the functional organization of the brain in the resting state that affects the information transmission from and to other brain regions related to both primary dysfunctions and higher-level cognition impairments in PD. Furthermore, we showed that regions with high degree values could be detected as betweenness centrality nodes. Our results demonstrate that properties of small-world connectivity could also recognize and quantify the characteristics of directed influence brain networks in PD.

A low-complexity adaptive beamformer for ultrasound imaging using structured covariance matrix.

In recent years, adaptive beamforming methods have been successfully applied to medical ultrasound imaging, resulting in simultaneous improvement in imaging resolution and contrast. These improvements have been achieved at the expense of higher computational complexity, with respect to the conventional non-adaptive delay-and-sum (DAS) beamformer, in which computational complexity is proportional to the number of elements, O(M). The computational overhead results from the covariance matrix inversion needed for computation of the adaptive weights, the complexity of which is cubic with the subarray size, O(L(3)). This is a computationally intensive procedure, which makes the implementation of adaptive beamformers less attractive in spite of their advantages. Considering that, in medical ultrasound applications, most of the energy is scattered from angles close to the steering angle, assuming spatial stationarity is a good approximation, allowing us to assume the Toeplitz structure for the estimated covariance matrix. Based on this idea, in this paper, we have applied the Toeplitz structure to the spatially smoothed covariance matrix by averaging the entries along all subdiagonals. Because the inverse of the resulting Toeplitz covariance matrix can be computed in O(L(2)) operations, this technique results in a greatly reduced computational complexity. By using simulated and experimental RF data-point targets as well as cyst phantoms-we show that the proposed low-complexity adaptive beamformer significantly outperforms the DAS and its performance is comparable to that of the minimum variance beamformer, with reduced computational complexity.

Contrast enhancement and robustness improvement of adaptive ultrasound imaging using forward-backward minimum variance beamforming.

In adaptive ultrasound imaging, accurate estimation of the array covariance matrix is of great importance, and biases the performance of the adaptive beamformer. The more accurately the covariance matrix can be estimated, the better the resolution and contrast can be achieved in the ultrasound image. To this end, in this paper, we have used the forward-backward spatial averaging for array covariance matrix estimation, which is then employed in minimum variance (MV) weights calculation. The performance of the proposed forward-backward MV (FBMV) beamformer is tested on simulated data obtained using Field II. Data for two closely located point targets surrounded by speckle pattern are simulated showing the higher amplitude resolution of the FBMV beamformer in comparison to the forward-only (F-only) MV beamformers, without the need for diagonal loading. A circular cyst with a diameter of 6 mm and a phantom containing wire targets and two cysts with different diameters of 8 mm and 6 mm are also simulated. The simulations show that the FBMV beamformer, in contrast to the F-only MV, could estimate the background speckle statistics without the need for temporal smoothing, resulting in higher contrast for the FBMV-resulted image in comparison to the MV images. In addition, the effect of steering vector errors is investigated by applying an error of the sound speed estimate to the ultrasound data. The simulations show that the proposed FBMV beamformer presents a satisfactory robustness against data misalignment resulted from steering vector errors, outperforming the regularized F-only MV beamformer. These improvements are achieved without compromising the good resolution of the MV beamformer and resulted from more accurate estimation of the covariance matrix and consequently, the more accurate setting of the MV weights.

Retinal image analysis using curvelet transform and multistructure elements morphology by reconstruction.

Retinal images can be used in several applications, such as ocular fundus operations as well as human recognition. Also, they play important roles in detection of some diseases in early stages, such as diabetes, which can be performed by comparison of the states of retinal blood vessels. Intrinsic characteristics of retinal images make the blood vessel detection process difficult. Here, we proposed a new algorithm to detect the retinal blood vessels effectively. Due to the high ability of the curvelet transform in representing the edges, modification of curvelet transform coefficients to enhance the retinal image edges better prepares the image for the segmentation part. The directionality feature of the multistructure elements method makes it an effective tool in edge detection. Hence, morphology operators using multistructure elements are applied to the enhanced image in order to find the retinal image ridges. Afterward, morphological operators by reconstruction eliminate the ridges not belonging to the vessel tree while trying to preserve the thin vessels unchanged. In order to increase the efficiency of the morphological operators by reconstruction, they were applied using multistructure elements. A simple thresholding method along with connected components analysis (CCA) indicates the remained ridges belonging to vessels. In order to utilize CCA more efficiently, we locally applied the CCA and length filtering instead of considering the whole image. Experimental results on a known database, DRIVE, and achieving to more than 94% accuracy in about 50 s for blood vessel detection, proved that the blood vessels can be effectively detected by applying our method on the retinal images.

Eigenspace-based minimum variance beamforming applied to medical ultrasound imaging.

Recently, adaptive beamforming methods have been successfully applied to medical ultrasound imaging, resulting in significant improvement in image quality compared with non-adaptive delay-and-sum (DAS) beamformers. Most of the adaptive beamformers presented in the ultrasound imaging literature are based on the minimum variance (MV) beamformer which can significantly improve the imaging resolution, although their success in enhancing the contrast has not yet been satisfactory. It is desirable for the beamformer to improve the resolution and contrast at the same time. To this end, in this paper, we have applied the eigenspace-based MV (EIBMV) beamformer to medical ultrasound imaging and have shown a simultaneous improvement in imaging resolution and contrast. EIBMV beamformer utilizes the eigenstructure of the covariance matrix to enhance the performance of the MV beamformer. The weight vector of the EIBMV is found by projecting the MV weight vector onto a vector subspace constructed from the eigenstructure of the covariance matrix. Using EIBMV weights instead of the MV ones leads to reduced sidelobes and improved contrast, without compromising the high resolution of the MV beamformer. In addition, the proposed EIBMV beamformer presents a satisfactory robustness against data misalignment resulting from steering vector errors, outperforming the regularized MV beamformer.

Quantitative assessment of pressure sore generation and healing through numerical analysis of high-frequency ultrasound images.

Abstract-This article focuses on the development of a method to quantitatively assess the healing process of artificially induced pressure sores using high-frequency (20 MHz) ultrasound images. We induced sores in guinea pigs and monitored predefined regions on days 3, 7, 14, and 21 after sore generation. We extracted relevant parameters regarding the tissue echographic structure and attenuation properties. We examined tissue healing by defining a healing function that used the extracted parameters. We verified the significance of the extracted features by using analysis of variance and multiple comparison tests. The features displayed ascending/descending behavior during wound generation and reverse behavior during healing. We optimized the parameters of our healing function by using a pattern search method. We tested the efficiency of the optimized values by calculating the healing function value on assessment days and then comparing these results with the expected pattern of changes in the tissue conditions after removing the applied pressure. The results of this study suggest that the methodology developed may be a viable tool for quantitative assessment of pressure sores during their early generation as well as during healing stages.

Minimum variance beamforming combined with adaptive coherence weighting applied to medical ultrasound imaging.

Currently, the nonadaptive delay-and-sum (DAS) beamformer is used in medical ultrasound imaging. However, due to its data-independent nature, DAS leads to images with limited resolution and contrast. In this paper, an adaptive minimum variance (MV)-based beamformer that combines the MV and coherence factor (CF) weighting is introduced and adapted to medical ultrasound imaging. MV-adaptive beamformers can improve the image quality in terms of resolution and sidelobes by suppressing off-axis signals, while keeping onaxis ones. In addition, CF weighting can improve contrast and sidelobes by emphasizing the in-phase signals and reducing the out-of-phase ones. Combining MV and CF weighting results in simultaneous improvement of imaging resolution and contrast, outperforming both DAS and MV beamformers. In addition, because of the power of CF in reducing the focusing errors, the proposed method presents satisfactory robustness against sound velocity inhomogeneities, outperforming the regularized MV beamformer. The excellent performance of the proposed beamforming approach is demonstrated by several simulated examples.

A transformation based method to design ultrasound array.

In medical ultrasound imaging, the desired lateral field distribution at each focal distance can be obtained by optimal apodization. On the other hand, the lateral field is a function of focal distance. Hence, finding the optimal apodization is a very arduous process. To overcome this, we have introduced a suboptimal method by which optimal apodization can be calculated in any distance through a nonlinear transformation by the knowledge of the optimal one at a distance. This transformation is established on a fact that the lateral field distribution at focal distance can be expressed as the Fourier transform of a nonlinear function of the aperture weighting, instead of direct expression as the Fourier transform of the above. We have applied this method to map the apodization which obtains the desired beam pattern into the apodization which maintains the same properties on the lateral field distribution. For example, applying this method on a 50-elements lambda/2 spaced linear array with length D has resulted in apodization that is optimal at distances D or D/2 by precision better than 9%. This method is useful especially in optimization problems, having no explicit constraint on the main lobe width, such as minimizing the sidelobe levels or minimizing main lobe width constrained to a predetermined value of sidelobe level. However, as the results show, this technique provides acceptable results in other cases.