- Research
- Open Access

# Eigenspace generalized sidelobe canceller combined with SNR dependent coherence factor for plane wave imaging

- Aácio José Zimbico
^{1, 2}Email author, - Diogo Watchel Granado
^{2}, - Fabio Kurt Schneider
^{2}, - Joaquim Miguel Maia
^{2}, - Amauri Amorin Assef
^{2}, - Nivaldo SchieflerJr.
^{2}and - Eduardo Tavares Costa
^{3}

**17**:109

https://doi.org/10.1186/s12938-018-0541-1

© The Author(s) 2018

**Received:**15 February 2018**Accepted:**23 May 2018**Published:**13 August 2018

## Abstract

### Background

The eigenspace generalized sidelobe canceller (EGSC) beamformer combined with a signal-to-noise ratio (SNR) dependent coherence factor (CF) is suggested for coherent plane wave compounding (PW) imaging. Conventional CF based methods such as generalized CF and subarray CF can improve the image quality, however, they are not suitable for low SNR. On the other hand, the EGSC CF based approach can introduce improvements in image quality, however, in PW imaging is susceptible to suffer from degradation due to low SNR which leads to a poor image quality. To overcome this limitation, the SNR dependent CF method is suggested for application in such situations due to its ability to control the SNR levels.

### Methods

The Field II and the Verasonics ultrasound imaging system with a L11-4v array transducer with a contrast resolution phantom were used to capture the plane wave sequences of simulation and experimental data, respectively. The performance evaluation using full width at half maximum (FWHM), contrast (CR and CNR) and the speckle statistics by using the signal to noise ratio (SNR) complemented by the Rayleigh distribution analysis was performed. In order to evaluate the performance of the \(\text {EGSC}_{3}\) (the SNR CF) beamformer, the comparison is done with particular importance to other CF-based approaches such as \(\text {EGSC}_{1}\) (the generalized CF) and, \(\text {EGSC}_{2}\) (the subarray CF) respectively.

### Results

Taking DAS as reference, \(\text {EGSC}_3\) showed 30.3 and 39.5% of improvement for \(\text {CR(dB)}\) and \(\text {CNR}\), respectively, when using experimental data. The proposed method also slightly outperforms the \(\text {EGSC}_1\) and \(\text {EGSC}_2\) methods for \(\text {CR(dB)}\), \(\text {CNR}\), and speckle statistics assessment.

### Conclusion

The \(\text {EGSC}_3\) is, therefore, suitable for CPWC by improving the spatial resolution and contrast while preserving the speckle pattern.

## Keywords

- Minimum variance
- Ultrasound imaging
- Contrast
- Resolution
- Coherence factor
- Speckle

## Background

Medical ultrasonic imaging is a noninvasive and low-cost technology widely used for diagnosis. Several techniques have been recently introduced for medical ultrasonic imaging in addition to the traditional Delay and Sum (DAS) beamforming method. DAS is a non adaptive beamforming method since it applies a fixed weight function (e.g., boxcar and humming) for array data summation. Adaptive weighting is, therefore, expected to result in increased image quality as introduced later. In addition, image reconstruction techniques have been proposed to increase frame rate as in plane wave imaging (PWI) [1, 2]. In PWI the medium is illuminated using all the aperture elements simultaneously to create a plane ultrasound wave. The signal to noise ratio (SNR) is commonly smaller compared to DAS method because PWI uses a single or a few ultrasound plane wavefronts to interrogate the medium in order to create a frame, therefore reducing the ultrasound power per generated frame opposed to the focused firing in DAS. On the other hand, since frame rate is generally limited by the time of flight of the ultrasound signal, PWI can significantly contribute to increase the frame rate of the system [1, 2].

Besides the lower interrogating power, contrary to DAS, in PWI the received signal in all transducer elements is composed by backscattered echoes from many different points from the interrogated medium resulting in lower SNR [1–4] being expected a tradeoff between the frame rate and the image quality in terms of spatial resolution and contrast [1, 2]. The possibility of compounding a set of low-resolution images to get an image with improved spatial resolution and contrast is the basis of spatial compounding (SC) as described by Berson et al. [5] to reduce speckle noise. Additionally, Montaldo [1] proposed a SC method called coherent plane wave compounding (CPWC) for further improvements.

As mentioned earlier, data-dependent weights can be created to also improve image quality in terms of spatial resolution and contrast by maintaining the main lobe while reducing the side lobes [6–9]. One of the most representative adaptive method is the minimum variance (MV) beamformer proposed by Capon [10]. The aim of the MV beamformer is minimizing the output power of a beamformer while keeping the signal undistorted [7, 8] and was previously evaluated into ultrasound imaging [6–8]. In MV-based methods, the covariance matrix (CM) estimate is one of the key steps which determine the performance of the algorithm [7]. Additionally, the adaptive weights are obtained at the cost of additional computation which comes from the need of inverting the CM [6, 7].

In order to obtain a robust CM estimation, Sasso and Cohen-Bacrie [11] proposed the spatial smoothing method, Synnevag et al. [7] introduced the diagonal loading (DL) technique and, Asl and Mohammadzadeh [12] introduced forward–backforward method. Likewise, Mehdizadeh [13] proposed the Eigenspace-based MV (EMV) beamformer to improve the image quality. The EMV consists of projecting the MV weights onto the signal subspace where a considerable part of the noise is removed from the received data. The signal subspace is constructed from the eigen decomposition of the CM. This procedure consists of sorting the eigenvalues of CM such that the eigenvectors associated to the larger eigenvalues determine the basis for the signal subspace [13, 14].

Another robust representation of MV beamformer is the generalized sidelobe canceller (GSC) beamformer proposed by Applebaum and Chapman [15] and later popularized by Jim and Griffiths [16] in the area of array signal processing. The GSC has been implemented in ultrasound imaging and compared to DAS and MV beamformers [17–20]. Similarly to EMV compared to MV, projecting the GSC weight onto the signal subspace lead to the eigenspace-based GSC (EGSC) beamformer implemented by Aliabadi et al. [19] resulting in improved image quality at increased computational cost when compared to GSC beamformer.

A coherence factor (CF) may also be combined into adaptive beamformers such as MV and was previously used for aberration correction and sidelobe suppression in ultrasound imaging [21].

Li and Li [22] have proposed the implementation of generalized CF (GCF) approach in MV beamformer to improve imaging resolution by protecting the main lobe. To improve the GCF performance, Zhau et al. [23] have suggested the subarray CF (SCF) approach. The SCF has optimized the performance of the EMV beamformer [23] at the cost of increased computational complexity. As in GCF, the SCF results from a combination of the incoherent sum and the coherent sum of signal energy [24, 25]. To address the limitations of GCF and SCF, Wang et al. proposed the SNR dependent CF (SNR-CF). In the SNR-CF method, a weighting scaled factor determined by the sigmoid function is combined with the coherent and incoherent summations of signal energy in order to control local levels of SNR [23]. Due to the SNR-CF features, the effects of signal cancellation at low SNR were minimized while the effects of self-cancellation were controlled at high SNR, which may introduce improvements in spatial resolution and contrast [24].

Considering SNR-CF based method is suitable for applications in signals with low SNR and recognizing that PWI typically presents low SNR [23], the SNR-CF method could aggregate some benefits [24, 25] into PWI-based methods. In this work, we suggest a combination between the EGSC with SNR dependent CF (\(\text {CF}_3\)) method to form the (\(\text {EGSC}_3\)) beamformer. For comparison, the GCF (\(\text {CF}_1\)) is also combined with EGSC to obtain the (\(\text {EGSC}_1\)) beamformer while the SCF (\(\text {CF}_2\)) combines with EGSC to form (\(\text {EGSC}_2\)) beamformer. The tests were performed using simulated and phantom data.

## Methods

The coherent plane wave compounding principle is firstly presented followed by evolved beamforming methods including different adaptive techniques such as MV, EMV, GSC, EGSC, \(\text {EGSC}_1\), \(\text {EGSC}_2\) and \(\text {EGSC}_3\). These techniques are the basis for the proposed SNRCF-based EGSC method and are later used for beamforming performance assessment.

### Beamforming methods

#### Coherent plane wave compounding

*u*is the angular apodization, \(h_{ir}\) is the impulse response of element

*r*to plane wave

*i*. For each time sample, \(\tau _i=(di+dr)/c\) represents the total travel time for

*i*th wave where \(d_i=zcos\alpha _i+xsin\alpha _i\), is the distance from wave emission to point (x, z),

*dr*is the distance from point (x, z) to the receive

*r*th array element and, c is the sound speed set to be 1540 m/s [26].

*K*represents the total amount of time samples recorded in each array element,

*r*for the

*i*th wave emission. In CPWC, the beamformer output Eq. (4) is obtained by averaging the data in Eq. (3) as follows:

#### Minimum variance

While the former is performed taking into account \(F=2T+1\), where *T* is the number of temporal samples used in covariance matrix (CM) estimation, the latter considers the \(G=M-L+1\) overlapping subarrays \(\text {X}_{\text {l}}(k)\) whose length is limited to (\(L\le M/2\)) [7].

*R*, can be subject to diagonal loading (DL) method in order to get a robust estimate [7]. The formulation of CM estimate in Eq. (7) combines the previous aspects as follows [7]:

#### Eigenspace minimum variance

#### The generalized side lobe canceller

### The eigenspace GSC

#### Generalized coherence factor (GCF) and subarray-based coherent factor (SCF)

#### The SNR dependent CF

#### Proposed SNR CF-based EGSC

In this work, the core of the implementation of the CF-based methods is the EGSC beamformer. More emphasis is given to the combination with the SNR CF (\(\text {CF}_3\)) since it appears to offer better performance over low SNR data [23, 24]. Moreover, the EGSC ideally provides the better performance than the DAS, MV, EMV and GSC [27] beamformers, so that the CF-based methods are suggested to be combined with the EGSC beamformer to obtain the EGSC-CF based beamformers. In addition, the (SNR) coherence factor (CF) method (\(\text {EGSC}_3\)) is then compared with the following categories of CF based methods: the GCF (\(\text {CF}_1\)) and the SCF (\(\text {CF}_2\)) [23, 24], respectively.

#### Implementation procedure

Review of the implementation procedures of the proposed beamformer algorithm

I. | Signals in Eq. (1) are arranged to form the 2-D matrix data representation in Eq. (3). Different firing events correspond to different steering angles thus, the calculated time delay is different. For each emission, the RF data are collected and combined with the calculated time delays to get the CPWC signals |

II. | Estimate CM for the receive aperture. The spatial smoothing, the temporal averaging and the DL approach are performed in order to obtain a robust estimate of CM as in Eq. (19) |

III. | Combine the data CM estimate with the steering vector a to compute the MV beamformer weight as in Eq. (6) |

IV. | Compute the GSC beamformer weight in Eq. (13) by combining the adaptive Eq. (11) and the nonadaptive braces Eq. (12), respectively |

V. | Perform the eigen decomposition of CM in order to construct the signal subspace in Eq. (9) and, compute the EGSC weight using Eq. (15) |

VI. | Compute the weights for the CF based methods \(\text {EGSC}_1\), \(\text {EGSC}_2\) and \(\text {EGSC}_3\) in Eqs. (16)–(19) using Eqs. (15), (16), (15)–(17) and (15)–(19), respectively |

VI. | Using Eq. (8) apply Eqs. (20)–(22) to compute the beamformer output for the proposed methods |

### Evaluation metrics

The evaluation metrics for resolution analysis were FWHM and PSL while CR, CNR, SNR, and the theoretical Rayleigh distribution (TRD), were used for image contrast and speckle assessment as follow.

#### Full-width half maximum (FWHM) and peak side lobe levels (PSL)

The FWHM [29] and the PSL [27] values are computed for evaluation of spatial resolution and interference. The former defines the beam width of the main lobe at − 6 dB while the latter defines the level of the first side lobe level for evaluating the interference and noise suppression abilities of a beamformer. Normally, the evaluation of FWHM and PSL is complemented by plots of normalized profiles and, improvements are presented by the narrower beam width and lower PSL [29].

#### Contrast ratio (CR), contrast to noise ratio (CNR) and, signal to noise ratio

For anechoic cyst and hyperechoic target to background evaluation, the contrast ratio (CR) and contrast to noise ratio (CNR) was assessed using Eqs. (24) and (25) [23]. In this context, an increase of CR and CNR indicates improvements in image quality.

Since the speckle pattern of DAS beamformer follows closely the RD, the TRD with \(\text {SNR}=1.91\) [7, 29] was included for illustration in Fig. 3 where the TRD together with the histogram of echo brightness (amplitudes) for simulated data are presented.

One can examine if the speckle pattern of the beamformed data follows the RD by performing the hypothesis test (K-test 5% significance). The test is decided when the computed p-value is compared with the maximal extreme of the significance interval [23].

### Simulations and experiments

In order to test the applicability of the proposed method in CPWC imaging, we used the simulated and phantom dataset. The simulated data were acquired using simulated phantoms in Field II program for individual point scatters and the circular anechoic cysts [30]. A 128-element linear array transducer model L11-4v with center frequency of 6.25 MHz and fractional bandwidth of 60% was simulated. A two-cycle sine wave excitation pulse at the central frequency and sampling frequency of 40 MHz were used.

A set of point targets were simulated in a homogeneous medium. The point target simulation phantom with fifty scattering points was created. The points were laterally distributed (x = 0 mm, x = ±5 mm and x = ±10 mm) from an axial depth of (z = 20 mm) to (z = 45 mm) spaced 5 mm apart from each other.

Two circular anechoic cysts of 5 mm radius were simulated in a speckle pattern. The anechoic cysts with a diameter of 10 mm centered at lateral positions (\(x =+10\)) and (\(x= -10\) mm) at the axial position of (z = 42.5 mm) were placed in a simulated homogenous media with 80,000 scattering points.

For experiments, the contrast detail resolution ultrasound phantom ATS model 532A containing several circular anechoic cysts and hyperechoic targets with different values of contrast was used. In accordance with Fig. 4, we decided recording data positioned at 8 mm diameter along the phantom.

Data corresponding to a pair of closely set anechoic cysts (label A) with − 12 and − 9 dB and another pair of hyperechoic cysts (label B) with + 9 and + 12 dB (see Fig. 4) were acquired. The background (speckle) and the regions inside the cysts have been used as the reference to compare the performance of the different beamformers.

A 128 element linear array transducer working at 6.25 MHz central frequency and pitch of 0.3 mm was used. For data processing, the sampling frequency was set to be 25 MHz. Verasonics imaging system using an L11-4v (Verasonics Ltd, Kirkland-WA, USA) transducer was used to acquire data with a sampling frequency of 25 MHz and transducer center frequency of 6.25 MHz.

For both simulation and the experimental studies, all the array elements were used to emit the plane waves and record the RF data. Furthermore, either in emission or the reception, the acquisition system was not apodized.

Additionally, for simulation and the experiments, the plane waves were emitted with different steering angles for multiple scanning. All the transmitting steering angles were set to range from \(-5^{\circ }\) to \(5^{\circ }\) (21 angles with an interval of \(0.5^{\circ }\)) and, the f-number of 1.75 was used.

Important parameters for simulated and experimental for acquisition and data processing

Imaging parameters | Simulation | Experimental |
---|---|---|

Central, sampling frequencies (MHz) | 6.25, 25 | 6.25, 40 |

Transducer model | L11-4v | L11-4v |

Number of elements (NE) | 128 | 128 |

NE for emission, reception | 128, 128 | 128, 128 |

Subarray length (L) | 64 | 64 |

Sound speed (m/s) | 1540 | 1540 |

Fractional bandwidth (%) | 60 | 60 |

Pitch (mm) | 0.3 | 0.3 |

F number | 1.75 | 1.75 |

Range, angles, gap | \(-5^{\circ }\) to \(5^{\circ }\), 21, \(0.5^{\circ }\) | \(-5^{\circ }\) to \(5^{\circ }\), 21, \(0.5^{\circ }\) |

Data acquiring | From field II | CRP (ATS) |

Diagonal loading | \(\Delta =0.01\) | \(\Delta =0.01\) |

Subspace construction | \(\alpha _{\text {th}}=0.05\) | \(\alpha _{\text {th}}=0.05\) |

SNR-CF weight param | \(\alpha _\text {snr}=\pi\), \(\beta _\text {snr}=0.5\) | \(\alpha _\text {snr}=\pi\), \(\beta _\text {snr}=0.5\) |

## Results and discussion

The simulated and phantom dataset was used to evaluate the performance of the proposed beamformers.

### Simulation: point targets

Fifteen point targets were simulated in order to test the performance of different beamformers as shown in Fig. 5 with a dynamic range of 60 dB. The point targets are respectively located at depths of 20 mm, 25 mm, 30 mm, 35 mm and 40 mm. The lateral profiles for FWHM and PSL assessment are shown in Fig. 6 using the point target highlighted in green box at (x, z) = (0, 30) mm in Fig. 5a as reference.

Figure 6 presents the improvements introduced by adaptive techniques over the traditional DAS method. Figure 7 presents the 0 to − 6 dB region in order to better the visualize the lateral resolution improvement.

From displayed images, it can be seen that DAS has a wide main lobe and higher side lobes compared to the MV beamformer. The EMV provides a slightly narrower main lobe compared to the MV beamformer.

Additionally, the EMV presents the lower sidelobe levels compared to MV beamformer meaning that the EMV outperforms the MV beamformer in terms of lateral resolution by providing a lower FWHM and reduced PSL, respectively.

The GSC technique presented slightly narrower main lobe and lower side lobe level compared to EMV beamformer. The EGSC presented a narrower main lobe while lowering down the side lobe energy compared to GSC beamformer, which justifies the well-defined scatters on the displayed images.

Moreover, the CF-based methods came with remarkable improvements in image quality so that by combining the \(\text {CF}_1\) with \(\text {EGSC}\) resulted in the \(\text {EGSC}_1\) beamformer which outperforms the \(\text {EGSC}\) beamformer in terms of FWHM and PSL, respectively. Analogously, by combining the \(\text {CF}_2\) with \(\text {EGSC}\) we formulated the \(\text {EGSC}_2\) beamformer which appears to outperform \(\text {EGSC}_1\). Finally, on combining the \(\text {CF}_3\) with \(\text {EGSC}\) we formulated the \(\text {EGSC}_3\) which in turn, outperforms \(\text {EGSC}_2\), respectively.

Full width at half maximum and peak side lobe level results for simulated point targets for different beamformers

Beamformer | FWHM (mm) | PSL (dB) |
---|---|---|

DAS | 1.73 | − 23.4 |

MV | 1.18 | − 24.1 |

EMV | 0.92 | − 25.1 |

GSC | 0.89 | − 28.8 |

EGSC | 0.78 | − 29.9 |

\(\text {EGSC}_1\) | 0.76 | − 33.8 |

\(\text {EGSC}_2\) | 0.69 | − 51.4 |

\(\text {EGSC}_3\) | 0.67 | − 53.2 |

### Simulation: circular anechoic cysts

The simulated speckle pattern containing two circular anechoic cysts closely spaced was used for contrast and speckle statistics evaluation. For CR, CNR and SNR calculation, the background region marked in a white box and the interior of the cysts marked with green as shown in Fig. 8a were used for calculating the mean intensity in the background and inside the cyst, respectively. The responses for different techniques are displayed in Fig. 8.

In Fig. 8, it is shown that DAS, MV and GSC have a poor contrast due to higher side lobes level while the EMV and the EGSC can provide a better contrast. The EGSC with all versions of coherent factors has a relatively higher contrast and also exhibits clearly the cyst edge.

Quantitative results show that all the adaptive beamformers outperformed the DAS beamformer are presented in Table 4. Taking DAS as reference, the \(\text {CR(dB)/CNR}\) presented the following improvements 1.5/0.08, 2.8/0.11, 3.0/0.16, 4.2/0.93, 6.1/1.07, 7.6/1.10 and, 9.1/1.12, for the MV, EMV GSC, EGSC, \(\text {EGSC}_1\), \(\text {EGSC}_2\), \(\text {EGSC}_3\), respectively. In terms of percentage (%) improvement, the values are 5.9/5.36, 10.4/7.23, 11.1/10.19, 14.8/39.74, 20.2/43.14, 23.9/43.82, 27.4/44.26, respectively. As compared to \(\text {EGSC}_1\) and \(\text {EGSC}_2\), values of 1.5/0.03 and 3.0/0.05 were found representing in percentage (%) 4.7/1.19 and 9.0/1.97, respectively.

Additionally, we have measured the speckle statistics for the different beamformers as presented in Fig. 9 complemented by different value of SNR presented in Table 4. For this purpose, the normalized pdf of the speckle region (see Fig. 8a white box) was computed for different beamformers.

Since the speckle pattern of DAS beamformer follows closely the TRD [7, 29], was included for illustration. All the beamformers were subject to the hypothesis test separately and the results showed that they followed the RD [7, 23, 29].

Contrast results for simulated circular anechoic cysts for different beamformers

Beamformer | IIC (dB) | IOC (dB) | CR (dB) | CNR | SNR |
---|---|---|---|---|---|

DAS | − 38.2 | − 14.1 | 24.1 | 1.41 | 1.65 |

MV | − 40.1 | − 14.5 | 25.6 | 1.49 | 1.74 |

EMV | − 42.4 | − 15.5 | 26.9 | 1.52 | 1.64 |

GSC | − 46.6 | − 19.5 | 27.1 | 1.57 | 1.56 |

EGSC | -48.1 | − 19.8 | 28.3 | 2.34 | 1.48 |

\(\text {EGSC}_{1}\) | − 51.4 | − 21.2 | 30.2 | 2.48 | 1.49 |

\(\text {EGSC}_{2}\) | − 52.2 | − 20.5 | 31.7 | 2.51 | 1.56 |

\(\text {EGSC}_{3}\) | − 53.7 | − 20.5 | 33.2 | 2.53 | 1.61 |

### Experiments: phantom circular anechoic cyst

The cysts illustrated in Fig. 4 are ordered such that the rightmost cyst has − 9 dB contrast whereas, the cyst on the left has − 12 dB contrast, respectively. The images for the anechoic cyst using the various techniques are shown in Fig. 10.

The DAS, MV, and GSC exhibited a poor contrast due to higher side lobes level while the EMV and the EGSC can provide a better contrast than DAS, MV, and GSC, respectively, but the visibility of the edge of the cyst is still difficult. However, the EGSC with all versions of CF outperforms in terms of contrast and exhibits clearly the improvements of the cyst definition. Taking DAS as reference, the \(\text {CR(dB)/CNR}\) presented the following improvements 1.5/0.18, 2.8/0.22, 4.0/0.31, 5.2/0.62, 7.1/0.84, 9.6/0.87 and, 10.1/0.90, for the MV, EMV GSC, EGSC, \(\text {EGSC}_1\), \(\text {EGSC}_2\), \(\text {EGSC}_3\), respectively. In terms of percentage (%) improvement, the values are 5.9/10.1, 10.4/12.0, 14.2/16.1, 17.7/27.8, 22.8/34.3, 28.5/35.1 and, 29.5/35.9, respectively. As compared to \(\text {EGSC}_1\) and \(\text {EGSC}_2\), values of 2.5/0.03 and 3.0/0.06 were found representing in percentage (%) 7.4/1.2 and 8.8/2.4, respectively.

Additionally, we have measured the speckle statistics for the different beamformers as presented in Fig. 11 complemented by different value of SNR presented in Table 5. For this purpose, the normalized pdf of the speckle region (see Fig. 11a white box) was computed for different beamformers. Similarly to the simulated data, all the beamformers passed the hypothesis tests.

Contrast results for phantom circular anechoic cysts for different beamformers

Beamformer | IIC (dB) | IOC (dB) | CR (dB) | CNR | SNR |
---|---|---|---|---|---|

DAS | − 45.2 | − 21.1 | 24.1 | 1.61 | 1.63 |

MV | − 47.1 | − 22.6 | 25.6 | 1.79 | 1.65 |

EMV | − 48.4 | − 22.5 | 26.9 | 1.83 | 1.61 |

GSC | − 49.6 | − 22.5 | 28.1 | 1.92 | 1.58 |

EGSC | − 50.1 | − 21.8 | 29.3 | 2.23 | 1.56 |

\(\text {EGSC}_{1}\) | − 50.4 | − 21.2 | 31.2 | 2.45 | 1.58 |

\(\text {EGSC}_{2}\) | − 53.2 | − 23.5 | 33.7 | 2.48 | 1.61 |

\(\text {EGSC}_{3}\) | − 55.7 | − 25.5 | 34.2 | 2.51 | 1.74 |

### Experiments: phantom circular hyperechoic targets

The beamformed images reveal that the DAS, MV and GSC shows a poor contrast due to the higher side lobes level while the EMV and the EGSC provide an improved contrast compared to MV and the GSC.

Contrast results for experimental circular hyperchoic cysts for different beamformers

Beamformer | IIC (dB) | IOC (dB) | CR (dB) | CNR | SNR |
---|---|---|---|---|---|

DAS | − 43.7 | − 20.5 | 23.2 | 1.53 | 1.72 |

MV | − 45.2 | − 20.7 | 25.5 | 1.68 | 1.74 |

EMV | − 48.4 | − 22.2 | 27.2 | 1.74 | 1.68 |

GSC | − 49.4 | − 22.1 | 27.3 | 1.85 | 1.62 |

EGSC | − 50.9 | − 22.4 | 29.5 | 2.12 | 1.67 |

\(\text {EGSC}_1\) | − 52.1 | − 22.7 | 30.4 | 2.26 | 1.67 |

\(\text {EGSC}_2\) | − 53.8 | − 23.6 | 32.2 | 2.37 | 1.69 |

\(\text {EGSC}_3\) | − 55.2 | − 23.9 | 33.3 | 2.53 | 1.71 |

Taking DAS as reference, the \(\text {CR(dB)/CNR}\) presented the following improvements 2.1/0.15, 4.0/0.21, 4.1/0.32, 6.3/0.59, 7.2/0.73, 9.0/0.84 and, 10.1/1.0, for the MV, EMV GSC, EGSC, \(\text {EGSC}_1\), \(\text {EGSC}_2\), \(\text {EGSC}_3\), respectively. In terms of percentage (%) improvement, the values are 8.3/8.9, 14.7/12.1, 15.0/17.3, 21.4/27.8, 23.7/32.3, 27.9/35.4 and, 30.3/39.5, respectively. As compared to \(\text {EGSC}_1\) and \(\text {EGSC}_2\), values of 2.9/0.27 and 1.1/0.16 were found representing in percentage (%) 8.7/10.7 and 3.3/6.3, respectively.

In GSC method, a slightly dark image can be seen in Fig. 11d which justifies the reduced SNR value, however, the \(\text {EGSC}_1\), \(\text {EGSC}_2\) and \(\text {EGSC}_3\), respectively presented an increased SNR and hence, their corresponding plots of speckle pattern in Fig. 13 exhibited that their intensity levels in dB increased. This effect was in agreement with the SNR values presented in Table 5 in the sense that, among other beamformers, they were relatively close compared to DAS beamformer.

The average of magnitude (AM) of speckle for simulated and real data

Beamformer | Simulated AC anechoic cysts AM (dB) | Real data AC anechoic cysts AM (dB) | Real data HC hypoechoic targets AM (dB) |
---|---|---|---|

DAS | − 10.1 | − 11.1 | − 10.9 |

MV | − 20.0 | − 21.3 | − 18.3 |

EMV | − 16.2 | − 20.0 | − 18.1 |

GSC | − 17.3 | − 19.0 | − 19.8 |

EGSC | − 17.2 | − 16.6 | − 17.0 |

\(\text {EGSC}_1\) | − 14.8 | − 15.2 | − 16.4 |

\(\text {EGSC}_2\) | − 15.7 | − 14.4 | − 15.8 |

\(\text {EGSC}_3\) | − 13.8 | − 13.6 | − 14.8 |

### Computational complexity analysis

The main purpose of CPWC is to reduce the computational complexity (CC) aiming to increase the frame rate of the imaging system hence, the analysis of the CC is an important task. The indicative CC required to perform the DAS BF goes under *O*(*M*) where M is the array length set to be 128 and hence, it will need *O*(*M*) floating operations.

In adaptive methods the CC increases because the inversion and eigen decomposition operations of the CM. The CM inversion for MV require \(2/3 \times \text {L}^3\) floating operations when applying Gauss-Jordan eliminations [4, 31] while the Hessian matrix inversion in Eq. (11) for GSC [17] needs \(\text {L}^3\) floating operations. Note that the HM is more demanding than the CM. However, EMV and EGSC undergo the eigen decomposition of CM which in accordance with the Golub–Reinsch algorithm [32], need \(21 \times \text {L}^3\) floating operations. Hence, the EMV will need \(2/3 \times \text {L}^3+21 \times \text {L}^3\) while the EGSC \(\text {L}^3+21 \times \text {L}^3=22 \times \text {L}^3\) [4, 17, 31].

For GCF computation in (16) an additional computational burden lead to the computation of the coherent and the incoherent energy [23] which is \(\text {L}^2\). In the formulation of SCF (\(\text {CF}_2\)) Eq. (17) and the SNR-CF (\(\text {CF}_3\)) Eq. (19), the difference of the IS and CS is weighted using the scaling factors of \(1/\text {L}\) and Eq. (18), respectively. In SNR-CF (\(\text {CF}_3\)) formulation, the weight \(\eta (\text {SNR})\) Eq. (18) is more complex compared to \(1/\text {L}\) in SCF (\(\text {CF}_2\)) formulation so that the computation of \(\text {CF}_3\) will be more complex than \(\text {CF}_2\).

Time cost (in seconds) for all beamformers for simulated and real data

Beamformer | Simulated PT | Simulated AC | Real data AC | Real data HC |
---|---|---|---|---|

Size (1620 \(\times\) 128) | Size (1526 \(\times\) 128) | Size (2048 \(\times\) 128) | Size (2048 \(\times\) 128) | |

DAS | 109.1 ± 0.21 | 120.2 ± 0.23 | 133.2 ± 0.13 | 123.6 ± 1.23 |

MV | 621.3 ± 0.31 | 792 ± 0.31 | 675.0 ± 0.22 | 732.6 ± 0.32 |

EMV | 1075.2 ± 0.26 | 7236.1 ± 0.41 | 1260.6 ± 0.91 | 1239.0 ± 0.60 |

GSC | 798.7 ± 0.31 | 864.8 ± 0.53 | 739.2 ± 0.43 | 795.6 ± 0.51 |

EGSC | 1234.8 ± 0.22 | 1332.6 ± 0.23 | 1328.4 ± 0.51 | 1456.2 ± 0.4 |

\(\text {EGSC}_1\) | 1432.2 ± 0.33 | 1584.0 ± 0.14 | 1464.6 ± 0.64 | 1506.0 ± 0.22 |

\(\text {EGSC}_2\) | 1465.8 ± 0.25 | 1638.4 ± 0.32 | 1506.0 ± 0.32 | 1567.8 ± 0.08 |

\(\text {EGSC}_3\) | 1543.2 ± 0.45 | 1752.2 ± 0.33 | 1674.6 ± 0.31 | 1708.2 ± 0.43 |

Among the CF based methods, the processing flow indicated values of time somewhat close one another however, there was an agreement with the indicative CC analysis.

In general, the CF-based methods increase the computational complexity so that the \(\text {EGSC}_3\) method will be more time demanding than \(\text {EGSC}_2\), and \(\text {EGSC}_2\) more time consuming than the \(\text {EGSC}_1\) technique. Therefore, the improvements in image quality introduced by the EGSC-CF beamformers were at the cost of an extra computational load compared to the different adaptive methods. For real-time applications, the most significant computational complexity added (i.e., the matrix inversion computation of the CM or HM) can be executed applying appropriate algorithms and architectures such as recursive updating [14], and graphics processing units [4], respectively.

## Conclusion

The eigenspace generalized sidelobe canceller EGSC beamformer combined with the SNR dependent coherence factor SNRCF method has been proposed for coherent plane wave compounding imaging. The technique, here called \(\text {EGSC}_3\) beamformer was compared with various adaptive beamformers (i.e., MV, EMV, GSC and EGSC) and other coherence factor based adaptive beamformers such as generalized coherence factor (GCF) and subarray coherence factor (SCF) using simulated and experimental data.

Taking DAS as reference, \(\text {EGSC}_3\) showed improvements of 30.3 and 39.5% for \(\text {CR (dB)}\) and \(\text {CNR}\), respectively, using experimental data. As compared to \(\text {EGSC}_1\) and \(\text {EGSC}_2\), improvements of 2.9 and 1.1% for \(\text {CR (dB)}\) and 0.27 and 0.16 in \(\text {CNR}\) were found, respectively. Even thought the image quality improvement compared to other EGSC methods is small, the values of speckle statistics of \(\text {EGSC}_3\) outperforms the \(\text {EGSC}_1\) and \(\text {EGSC}_2\) methods and remained close to DAS contrary to the remaining adaptive beamformers. The \(\text {EGSC}_3\) is, therefore, suitable for CPWC by improving the spatial resolution and contrast while preserving the speckle pattern different from those beamforming methods that do not use coherence factors.

## Declarations

### Authors' contributions

All authors are part of the Lineup for ultrasound image processing and Reconstruction at the Graduate Program in Electrical and Computer Engineering (CPGEI) at the Federal University of Technology-Paraná (UTFPR), Brazil. AJZ was the leader for the proposed method conceiving and conducting the experiments and writing the manuscript with FKS. DWG, FKS, JMM, NSJ , ETC were responsible for reviewing the ultrasonic signal model, application and results evaluation. All authors read and approved the final manuscript.

### Acknowledgements

The authors thank for the financial support of Eduardo Mondlane University and by the Brazilian CAPES, Fundação Araucária and FINEP funding agencies and the Ministry of Health.

### Competing interests

The authors declare that they have no competing interests.

### Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

## Authors’ Affiliations

## References

- Montaldo G, Tanter M, Bercoff J, Benech N, Fink M. Coherent plane-wave compounding for very high frame rate ultrasonography and transient elastography. IEEE Trans Ultrason Ferroelectr Freq Control. 2009;56(3):489–506.View ArticleGoogle Scholar
- Bercoff J, Tanter M, Fink M. Supersonic shear imaging: a new technique for soft tissue elasticity mapping. IEEE Trans Ultrason Ferroelectr Freq Control. 2004;51(4):396–409.View ArticleGoogle Scholar
- Tanter M, Fink M. Ultrafast imaging in biomedical ultrasound. IEEE Trans Ultrason Ferroelectr Freq Control. 2014;61(1):102–19.View ArticleGoogle Scholar
- Zhao J, Wang Y, Zeng X, Yu J, Yiu BY, Alfred C. Plane wave compounding based on a joint transmitting-receiving adaptive beamformer. IEEE Trans Ultrason Ferroelectr Freq Control. 2015;62(8):1440–52.View ArticleGoogle Scholar
- Berson M, Roncin A, Pourcelot L. Compound scanning with an electrically steered beam. Ultrason Imaging. 1981;3(3):303–8.View ArticleGoogle Scholar
- Mann JA, Walker W. A constrained adaptive beamformer for medical ultrasound: initial results. In: IEEE Proceedings of ultrasonics symposium, 2002, vol. 2, p. 1807–10. New Jersey: IEEE; 2002.Google Scholar
- Synnevag J, Austeng A, Holm S. Adaptive beamforming applied to medical ultrasound imaging. IEEE Trans Ultrason Ferroelectr Freq Control. 2007;54(8):1606.View ArticleGoogle Scholar
- Synnevag JF, Austeng A, Holm S. Benefits of minimum-variance beamforming in medical ultrasound imaging. IEEE Trans Ultrason Ferroelectr Freq Control. 2009;56(9):1868–79.View ArticleGoogle Scholar
- Van Trees HL. Optimum array processing: Part IV of detection, estimation, and modulation theory. New York: Wiley; 2002. p. 362–82.View ArticleGoogle Scholar
- Capon J. High-resolution frequency–wavenumber spectrum analysis. Proc IEEE. 1969;57(8):1408–18.View ArticleGoogle Scholar
- Sasso M, Cohen-Bacrie C. Medical ultrasound imaging using the fully adaptive beamformer. In: Proceedings of IEEE International Conference on Acoustics, speech, and signal processing, 2005 (ICASSP’05), vol. 2, p. 489. New Jersey: IEEE; 2005.Google Scholar
- Asl BM, Mahloojifar A. Contrast enhancement and robustness improvement of adaptive ultrasound imaging using forward–backward minimum variance beamforming. IEEE Trans Ultrason Ferroelectr Freq Control. 2011;58(4):858–67.View ArticleGoogle Scholar
- Mehdizadeh S. Adaptive beamformers for ultrasound imaging of acoustically hard tissues. Ph.D. thesis, NTNU-trykk; 2012.Google Scholar
- Mehdizadeh S, Austeng A, Johansen TF, Holm S. Eigenspace based minimum variance beamforming applied to ultrasound imaging of acoustically hard tissues. IEEE Trans Med Imaging. 2012;31(10):1912–21.View ArticleGoogle Scholar
- Applebaum S, Chapman D. Adaptive arrays with main beam constraints. IEEE Trans Antennas Propagation. 1976;24(5):650–62.View ArticleGoogle Scholar
- Griffiths L, Jim C. An alternative approach to linearly constrained adaptive beamforming. IEEE Trans Antennas Propagation. 1982;30(1):27–34.View ArticleGoogle Scholar
- Albulayli M, Rakhmatov D. Hybrid adaptive/nonadaptive beamforming for ultrasound imaging. In: IEEE international conference on acoustics, speech and signal processing (ICASSP), 2013. p. 1061–65. New Jersey: IEEE; 2013.Google Scholar
- Zimbico A, Granado D, Schneider F, Maia J, Assef A, Pipa D, Costa E. Beam domain adaptive beamforming using generalized side lobe canceller with coherence factor for medical ultrasound imaging. In: IEEE international ultrasonics symposium (IUS), 2017. p. 1–4. Washington DC: IEEE; 2017.Google Scholar
- Aliabadi S, Wang Y, Yu J, Zhao J, Guo W, Zhang S. Eigenspace-based beamformer using oblique signal subspace projection for ultrasound plane-wave imaging. Biomed Eng Online. 2016;15(1):127.View ArticleGoogle Scholar
- Li J, Chen X, Wang Y, Chen X, Yu D. Forward–backward generalized sidelobe canceler beamforming applied to medical ultrasound imaging. AIP Adv. 2017;7(1):015201.View ArticleGoogle Scholar
- Nilsen CI, Holm S. Wiener beamforming and the coherence factor in ultrasound imaging. IEEE Trans Ultrason Ferroelectr Freq Control. 2010;57(6):1329–46.View ArticleGoogle Scholar
- Li P-C, Li M-L. Adaptive imaging using the generalized coherence factor. IEEE Trans Ultrason Ferroelectr Freq Control. 2003;50(2):128–41.View ArticleGoogle Scholar
- Zhao J, Wang Y, Yu J, Guo W, Li T, Zheng Y-P. Subarray coherence based postfilter for eigenspace based minimum variance beamformer in ultrasound plane-wave imaging. Ultrasonics. 2016;65:23–33.View ArticleGoogle Scholar
- Wang Y-H, Li P-C. Snr-dependent coherence-based adaptive imaging for high-frame-rate ultrasonic and photoacoustic imaging. IEEE Trans Ultrason Ferroelectr Freq Control. 2014;61(8):1419–32.View ArticleGoogle Scholar
- Deylami AM, Jensen JA, Asl BM. An improved minimum variance beamforming applied to plane-wave imaging in medical ultrasound. In: IEEE international ultrasonics symposium (IUS), 2016. p. 1–4. New Jersey: IEEE; 2016.Google Scholar
- Rodriguez-Molares A, Torp H, Denarie B, Løvstakken L. The angular apodization in coherent plane-wave compounding [correspondence]. IEEE Trans Ultrason Ferroelectr Freq Control. 2015;62(11):2018–23.View ArticleGoogle Scholar
- Li J, Chen X, Wang Y, Li W, Yu D. Eigenspace-based generalized sidelobe canceler beamforming applied to medical ultrasound imaging. Sensors. 2016;16(8):1192.View ArticleGoogle Scholar
- Asl BM, Mahloojifar A. Eigenspace-based minimum variance beamforming applied to medical ultrasound imaging. IEEE Trans Ultrason Ferroelectr Freq Control. 2010;57(11):2381.View ArticleGoogle Scholar
- Szabo TL. Diagnostic ultrasound imaging: inside out. Cambridge: Academic Press; 2004.Google Scholar
- Jensen JA. Simulation of advanced ultrasound systems using field ii. In: IEEE international symposium on biomedical imaging: nano to macro, 2004. p. 636–9. New Jersey: IEEE; 2004.Google Scholar
- Zeng X, Wang Y, Yu J, Guo Y. Correspondence-beam-domain eigenspace-based minimum variance beamformer for medical ultrasound imaging. IEEE Trans Ultrason Ferroelectr Freq Control. 2013;60(12):2670–6.View ArticleGoogle Scholar
- Ozaktas HM, Arikan O, Kutay MA, Bozdagt G. Digital computation of the fractional fourier transform. IEEE Trans Sign Process. 1996;44(9):2141–50.View ArticleGoogle Scholar