EEMD Domain AR Spectral Method for Mean Scatterer Spacing Estimation of Breast Tumors From Ultrasound Backscattered RF Data.

We present a novel method for estimating the mean scatterer spacing (MSS) of breast tumors using ensemble empirical mode decomposition (EEMD) domain analysis of deconvolved backscattered radio frequency (RF) data. The autoregressive (AR) spectrum from which the MSS is estimated is obtained from the intrinsic mode functions (IMFs) due to regular scatterers embedded in RF data corrupted by the diffuse scatterers. The IMFs are chosen by giving priority to the presence of an enhanced fundamental harmonic and the presence of a greater number of higher harmonics in the AR spectrum estimated from the IMFs. The AR model order is chosen by minimizing the mean absolute percentage error (MAPE) criterion. In order to ensure that the backscattered data is indeed from a source of coherent scattering, we begin by performing a non-parametric Kolmogorov-Smirnov (K-S) classification test on the backscattered RF data. Deconvolution of the backscattered RF data, which have been classified by the K-S test as sources of significant coherent scattering, is done to reduce the system effect. EEMD domain analysis is then performed on the deconvolved data. The proposed method is able to recover the harmonics associated with the regular scatterers and overcomes many problems encountered while estimating the MSS from the AR spectrum of raw RF data. Using our technique, a mean absolute percentage error (MAPE) of 5.78% is obtained while estimating the MSS from simulated data, which is lower than that of the existing techniques. Our proposed method is shown to outperform the state of the art techniques, namely, singular spectrum analysis, generalized spectrum (GS), spectral autocorrelation (SAC), and modified SAC for different simulation conditions. The MSS for in vivo normal breast tissue is found to be 0.69 ± 0.04 mm; for benign and malignant tumors it is found to be 0.73 ± 0.03 and 0.79 ± 0.04 mm, respectively. The separation between the MSS values of normal and benign tissues for our proposed method is similar to the separations obtained for the conventional methods, but the separation between the MSS values for benign and malignant tissues for our proposed method is slightly higher than that for the conventional methods. When the MSS is used to classify breast tumors into benign and malignant, for a threshold-based classifier, the increase in specificity, accuracy, and area under curve are 18%, 19%, and 22%, respectively, and that for statistical classifiers are 9%, 13%, and 19%, respectively, from that of the next best existing technique.

Computer-Aided Diagnosis for Breast Ultrasound Using Computerized BI-RADS Features and Machine Learning Methods.

This work identifies effective computable features from the Breast Imaging Reporting and Data System (BI-RADS), to develop a computer-aided diagnosis (CAD) system for breast ultrasound. Computerized features corresponding to ultrasound BI-RADs categories were designed and tested using a database of 283 pathology-proven benign and malignant lesions. Features were selected based on classification performance using a "bottom-up" approach for different machine learning methods, including decision tree, artificial neural network, random forest and support vector machine. Using 10-fold cross-validation on the database of 283 cases, the highest area under the receiver operating characteristic (ROC) curve (AUC) was 0.84 from a support vector machine with 77.7% overall accuracy; the highest overall accuracy, 78.5%, was from a random forest with the AUC 0.83. Lesion margin and orientation were optimum features common to all of the different machine learning methods. These features can be used in CAD systems to help distinguish benign from worrisome lesions.

Classification of breast tumour using electrical impedance and machine learning techniques.

When a breast lump is detected through palpation, mammography or ultrasonography, the final test for characterization of the tumour, whether it is malignant or benign, is biopsy. This is invasive and carries hazards associated with any surgical procedures. The present work was undertaken to study the feasibility for such characterization using non-invasive electrical impedance measurements and machine learning techniques. Because of changes in cell morphology of malignant and benign tumours, changes are expected in impedance at a fixed frequency, and versus frequency of measurement. Tetrapolar impedance measurement (TPIM) using four electrodes at the corners of a square region of sides 4 cm was used for zone localization. Data of impedance in two orthogonal directions, measured at 5 and 200 kHz from 19 subjects, and their respective slopes with frequency were subjected to machine learning procedures through the use of feature plots. These patients had single or multiple tumours of various types in one or both breasts, and four of them had malignant tumours, as diagnosed by core biopsy. Although size and depth of the tumours are expected to affect the measurements, this preliminary work ignored these effects. Selecting 12 features from the above measurements, feature plots were drawn for the 19 patients, which displayed considerable overlap between malignant and benign cases. However, based on observed qualitative trend of the measured values, when all the feature values were divided by respective ages, the two types of tumours separated out reasonably well. Using K-NN classification method the results obtained are, positive prediction value: 60%, negative prediction value: 93%, sensitivity: 75%, specificity: 87% and efficacy: 84%, which are very good for such a test on a small sample size. Study on a larger sample is expected to give confidence in this technique, and further improvement of the technique may have the ability to replace biopsy.

Using nearest neighbors for accurate estimation of ultrasonic attenuation in the spectral domain.

Attenuation is a key diagnostic parameter of tissue pathology change and thus may play a vital role in the quantitative discrimination of malignant and benign tumors in soft tissue. In this paper, two novel techniques are proposed for estimating the average ultrasonic attenuation in soft tissue using the spectral domain weighted nearest neighbor method. Because the attenuation coefficient of soft tissues can be considered to be a continuous function in a small neighborhood, we directly estimate an average value of it from the slope of the regression line fitted to the 1) modified average midband fit value and 2) the average center frequency shift along the depth. To calculate the average midband fit value, an average regression line computed from the exponentially weighted short-time Fourier transform (STFT) of the neighboring 1-D signal blocks, in the axial and lateral directions, is fitted over the usable bandwidth of the normalized power spectrum. The average center frequency downshift is computed from the maximization of a cost function defined from the normalized spectral cross-correlation (NSCC) of exponentially weighted nearest neighbors in both directions. Different from the large spatial signal-block-based spectral stability approach, a costfunction- based approach incorporating NSCC functions of neighboring 1-D signal blocks is introduced. This paves the way for using comparatively smaller spatial area along the lateral direction, a necessity for producing more realistic attenuation estimates for heterogeneous tissue. For accurate estimation of the attenuation coefficient, we also adopt a reference-phantombased diffraction-correction technique for both methods. The proposed attenuation estimation algorithm demonstrates better performance than other reported techniques in the tissue-mimicking phantom and the in vivo breast data analysis.

Direct and gradient-based average strain estimation by using weighted nearest neighbor cross-correlation peaks.

In this paper, two novel approaches, gradient-based and direct strain estimation techniques, are proposed for high-quality average strain imaging incorporating a cost function maximization. Stiffness typically is a continuous function. Consequently, stiffness of proximal tissues is very close to that of the tissue corresponding to a given data window. Hence, a cost function is defined from exponentially weighted neighboring pre- and post-compression RF echo normalized cross-correlation peaks in the lateral (for displacement estimation) or in both the axial and the lateral (for direct strain estimation) directions. This enforces a controlled continuity in displacement/strain and average displacement/strain is calculated from the corresponding maximized cost function. Axial stress causes lateral shift in the tissue. Therefore, a 1-D post-compression echo segment is selected by incorporating Poisson's ratio. Two stretching factors are considered simultaneously in gradient-based strain estimation that allow imaging the lesions properly. The proposed time-domain gradient-based and direct-strain-estimation-based algorithms demonstrate significantly better performance in terms of elastographic signal-to-noise ratio (SNRe), elastographic contrast-to-noise ratio (CNRe), peak signal-to-noise ratio (PSNR), and mean structural similarity (MSSIM) than the other reported time-domain gradient-based and direct-strain-estimation techniques in finite element modeling (FEM) simulation and phantom experiments. For example, in FEM simulation, it has been found that the proposed direct strain estimation method can improve up to approximately 2.49 to 8.71, 2.2 to 6.63, 1.5 to 5, and 1.59 to 2.45 dB in the SNRe, CNRe, PSNR, and MSSIM compared with the traditional direct strain estimation method, respectively, and the proposed gradient-based algorithm demonstrates 2.99 to 16.26, 18.74 to 23.88, 3 to 9.5, and 0.6 to 5.36 dB improvement in the SNRe, CNRe, PSNR, and MSSIM, respectively, compared with a recently reported time-domain gradient-based technique. The range of improvement as noted above is for low to high applied strains. In addition, the comparative results using the in vivo breast data (including malignant or benign masses) also show that the lesion size is better defined by the proposed gradient-based average strain estimation technique.

Direct mean strain estimation for elastography using nearest-neighbor weighted least-squares approach in the frequency domain.

Ultrasound elastography is emerging with enormous potential as a medical imaging modality for effective discrimination of pathological changes in soft tissue. It maps the tissue elasticity or strain due to a mechanical deformation applied to it. The strain image most often calculated from the derivative of the local displacement field is highly noisy because of the de-correlation effect mainly due to unstable free-hand scanning and/or irregular tissue motion; consequently, improving the SNR of the strain image is still a challenging problem in this area. In this paper, a novel approach using the nearest-neighbor weighted least-squares is presented for direct estimation of the 'mean' axial strain for high quality strain imaging. Like other time/frequency domain reported schemes, the proposed method exploits the fact that the post-compression rf echo signal is a time-scaled and shifted replica of the pre-compression rf echo signal. However, the elegance of our technique is that it directly computes the mean strain without explicitly using any post filter and/or previous local displacement/strain estimates as is usually done in the conventional approaches. It is implemented in the short-time Fourier transform domain through a nearest-neighbor weighted least-squares-based Fourier spectrum equalization technique. As the local tissue strain is expected to maintain continuity with its neighbors, we show here that the mean strain at the interrogative window can be directly computed from the common stretching factor that minimizes a cost function derived from the exponentially weighted windowed pre- and post-compression rf echo segments in both the lateral and axial directions. The performance of our algorithm is verified for up to 8% applied strain using simulation and experimental phantom data and the results reveal that the SNR of the strain image can be significantly improved compared to other reported algorithms in the literature. The efficacy of the algorithm is also tested with in vivo breast data known to have malignant or benign masses from histology.

Robust strain-estimation algorithm using combined radiofrequency and envelope cross-correlation with diffusion filtering.

In ultrasound elastography, the strain in compressed tissue due to external deformation is estimated and is smaller in harder than softer tissue. With increased stress, the nonaxial motions of tissue elements increase and result in noisier strain images. At high strain, the envelope of the rf signal exhibits robustness to signal decorrelation. However, the precision of strain estimates using envelope signals is much worse compared to that using the rf signals. In this paper, we propose a novel approach for robust strain estimation by combining weighted rf cross-correlation and envelope cross-correlation functions. An applied strain-dependent piecewise-linear-weight is used for this purpose. In addition, we introduce nonlinear diffusion filtering to further enhance the resulting strain image. The results of our algorithm are demonstrated for up to 10% applied strain using a finite-element modelling (FEM) simulation phantom. It reveals that the elastographic signal-to-noise ratio (SNRe) and the elastographic contrast-to-noise ratio (CNRe) of the strain images can be improved more significantly than with other algorithms used in this paper. In addition, comparative results in terms of the mean structural similarity (MSSIM) using in vivo breast data show that the strain image quality can be improved noticeably by the proposed method than with the techniques employed in this work.

Comparison of template-matching and singular-spectrum-analysis methods for imaging implanted brachytherapy seeds.

Brachytherapy using small implanted radioactive seeds is becoming an increasingly popular method for treating prostate cancer, in which a radiation oncologist implants seeds in the prostate transperineally under ultrasound guidance. Dosimetry software determines the optimal placement of seeds for achieving the prescribed dose based on ultrasonic determination of the gland boundaries. However, because of prostate movement and distortion during the implantation procedure, some seeds may not be placed in the desired locations; this causes the delivered dose to differ from the prescribed dose. Current ultrasonic imaging methods generally cannot depict the implanted seeds accurately. We are investigating new ultrasonic imaging methods that show promise for enhancing the visibility of seeds and thereby enabling real-time detection and correction of seed-placement errors during the implantation procedure. Real-time correction of seed-placement errors will improve the therapeutic radiation dose delivered to target tissues. In this work, we compare the potential performance of a template-matching method and a previously published method based on singular spectrum analysis for imaging seeds. In particular, we evaluated how changes in seed angle and position relative to the ultrasound beam affect seed detection. The conclusion of the present study is that singular spectrum analysis has better sensitivity but template matching is more resistant to false positives; both perform well enough to make seed detection clinically feasible over a relevant range of angles and positions. Combining the information provided by the two methods may further reduce ambiguities in determining where seeds are located.

Ultrasonic multi-feature analysis procedure for computer-aided diagnosis of solid breast lesions.

We have developed quantitative descriptors to provide an objective means of noninvasive identification of cancerous breast lesions. These descriptors include quantitative acoustic features assessed using spectrum analysis of ultrasonic radiofrequency (rf) echo signals and morphometric properties related to lesion shape. Acoustic features include measures of echogenicity, heterogeneity and shadowing, computed by generating spectral-parameter images of the lesion and surrounding tissue. Spectral-parameter values are derived from rf echo signals at each pixel using a sliding-window Fourier analysis. We derive quantitative acoustic features from spectral-parameter maps of the lesion and adjacent areas. We quantify morphometric features by geometric and fractal analysis of traced lesion boundaries. Initial results on biopsy-proven cases show that although a single parameter cannot reliably discriminate cancerous from noncancerous breast lesions, multi-feature analysis provides excellent discrimination for this data set. We have processed data for 130 biopsy-proven patients, acquired during routine ultrasonic examinations at three clinical sites and produced an area under the receiver-operating-characteristics (ROC) curve of 0.947 +/- 0.045. Among the quantitative descriptors, lesion-margin definition, spiculation and border irregularity are the most useful; some additional morphometric features (such as border irregularity) also are particularly effective in lesion classification. Our findings are consistent with many of the BI-RADS (Breast Imaging Reporting and Data System) breast-lesion-classification criteria in use today.

Radiation-force technique to monitor lesions during ultrasonic therapy.

This report describes a monitoring technique for high-intensity focused ultrasound (US), or HIFU, lesions, including protein-denaturing lesions (PDLs) and those made for noninvasive cardiac therapy and tumor treatment in the eye, liver and other organs. Designed to sense the increased stiffness of a HIFU lesion, this technique uniquely utilizes the radiation force of the therapeutic US beam as an elastographic push to detect relative stiffness changes. Feasibility was demonstrated with computer simulations (treating acoustically induced displacements, concomitant heating, and US displacement-estimation algorithms) and pilot in vitro experimental studies, which agree qualitatively in differentiating HIFU lesions from normal tissue. Detectable motion can be induced by a single 5 ms push with temperatures well below those needed to form a lesion. Conversely, because the characteristic heat diffusion time is much longer than the characteristic relaxation time following a push, properly timed multiple therapy pulses will form lesions while providing precise control during therapy.

Emerging ultrasound technologies for early markers of disease.

Ultrasound has been a popular clinical imaging modality for decades. It is well established as a means of displaying the macroscopic anatomy of soft-tissue structures. While conventional ultrasound methodologies (i.e., B-mode and Doppler methods) are well proven and continue to advance technically on a daily basis, e.g. by extending into higher frequencies and taking advantage of harmonic phenomena in tissues, fundamentally new ultrasound technologies also are emerging and offer exciting promise for making significant improvements in the clinical imaging of disease. These emerging methods include spectrum analysis, elasticity imaging, contrast-agent methods, and advanced flow detection and measurement techniques. Each provides independent information and, used alone, each can provide powerful new imaging capabilities; combined with each other, their capabilities may be even greater in many applications; and all in principle can be used in concert with other imaging modalities to offer the possibility of further improvements in disease detection, evaluation, and monitoring.