Fine-tuning the H-scan for discriminating changes in tissue scatterers

H-Scan Discrimination for Tumor Microenvironmental Heterogeneity in Melanoma

OBJECTIVE

Melanoma is a form of malignant skin cancer that exhibits significant inter-tumoral differences in the tumor microenvironment (TME) secondary to genetic mutations. The heterogeneity may be subtle but can complicate the treatment of metastatic melanoma, contributing to a high mortality rate. Therefore, developing an accurate and non-invasive procedure to discriminate microenvironmental heterogeneity to facilitate therapy selection is an important goal.

METHODS

In vivo murine melanoma models that recapitulate human disease using synchronous implanted YUMM 1.7 (Yale University Mouse Melanoma) and YUMMER 1.7 (Yale University Mouse Melanoma Exposed to Radiation) murine melanoma lines were investigated. Mice were treated with antibodies to modulate the immune response and longitudinally scanned with ultrasound (US). US radiofrequency data were processed using the H-scan analysis, attenuation estimation and B-mode processing to extract five US features. The measures were used to compare different TMEs (YUMMER vs. YUMM) and responses to immunomodulatory therapies with CD8 depletion or programmed cell death protein 1 (PD-1) inhibition.

RESULTS

Multiparametric analysis produced a combined H-scan parameter, resolving significant differences (i) between untreated YUMMER and YUMM and (ii) between untreated, PD-1-treated and CD8-treated YUMMER. However, more importantly, the B-mode and attenuation measures failed to differentiate YUMMER and YUMM and to monitor treatment responses, indicating that H-scan is required to differentiate subtle differences within the TME.

CONCLUSION

We anticipate that the H-scan analysis could discriminate heterogeneous melanoma metastases and guide diagnosis and treatment selection, potentially reducing the need for invasive biopsies or immunologic procedures.

Hermite-scan imaging for differentiating glioblastoma from normal brain: Simulations and ex vivo studies for applications in intra-operative tumor identificationa)

Hermite-scan (H-scan) imaging is a tissue characterization technique based on the analysis of raw ultrasound radio frequency (RF) echoes. It matches the RF echoes to Gaussian-weighted Hermite polynomials of various orders to extract information related to scatterer diameter. It provides a color map of large and small scatterers in the red and blue H-scan image channels, respectively. H-scan has been previously reported for characterizing breast, pancreatic, and thyroid tumors. The present work evaluated H-scan imaging to differentiate glioblastoma tumors from normal brain tissue ex vivo. First, we conducted 2-D numerical simulations using the k-wave toolbox to assess the performance of parameters derived from H-scan images of acoustic scatterers (15-150 μm diameters) and concentrations (0.2%-1% w/v). We found that the parameter intensity-weighted percentage of red (IWPR) was sensitive to changes in scatterer diameters independent of concentration. Next, we assessed the feasibility of using the IWPR parameter for differentiating glioblastoma and normal brain tissues (n = 11 samples per group). The IWPR parameter estimates for normal tissue (44.1% ± 1.4%) were significantly different (p < 0.0001) from those for glioblastoma (36.2% ± 0.65%). These findings advance the development of H-scan imaging for potential use in differentiating glioblastoma tumors from normal brain tissue during resection surgery.

3-D H-scan ultrasound imaging of relative scatterer size using a matrix array transducer and sparse random aperture compounding

H-scan ultrasound (US) is a high-resolution imaging technique for soft tissue characterization. By acquiring data in volume space, H-scan US can provide insight into subtle tissue changes or heterogenous patterns that might be missed using traditional cross-sectional US imaging approaches. In this study, we introduce a 3-dimensional (3-D) H-scan US imaging technology for voxel-level tissue characterization in simulation and experimentation. Using a matrix array transducer, H-scan US imaging was developed to evaluate the relative size of US scattering aggregates in volume space. Experimental data was acquired using a programmable US system (Vantage 256, Verasonics Inc, Kirkland, WA) equipped with a 1024-element (32 × 32) matrix array transducer (Vermon Inc, Tours, France). Imaging was performed using the full array in transmission. Radiofrequency (RF) data sequences were collected using a sparse random aperture compounding technique with 6 different data compounding approaches. Plane wave imaging at five angles was performed at a center frequency of 8 MHz. Scan conversion and attenuation correction were applied. To generate the 3-D H-scan US images, a convolution filter bank (N = 256) was then used to process the RF data sequences and measure the spectral content of the backscattered US signals before volume reconstruction. Preliminary experimental studies were conducted using homogeneous phantom materials embedded with spherical US scatterers of varying diameter, i.e., 27 to 45, 63 to 75, or 106-126 μm. Both simulated and experimental results revealed that 3-D H-scan US images have a low spatial variance when tested with homogeneous phantom materials. Furthermore, H-scan US is considerably more sensitive than traditional B-mode US imaging for differentiating US scatterers of varying size (p = 0.001 and p = 0.93, respectively). Overall, this study demonstrates the feasibility of 3-D H-scan US imaging using a matrix array transducer for tissue characterization in volume space.

Improving breast cancer diagnosis by incorporating raw ultrasound parameters into machine learning

The improved diagnostic accuracy of ultrasound breast examinations remains an important goal. In this study, we propose a biophysical feature-based machine learning method for breast cancer detection to improve the performance beyond a benchmark deep learning algorithm and to furthermore provide a color overlay visual map of the probability of malignancy within a lesion. This overall framework is termed disease-specific imaging. Previously, 150 breast lesions were segmented and classified utilizing a modified fully convolutional network and a modified GoogLeNet, respectively. In this study multiparametric analysis was performed within the contoured lesions. Features were extracted from ultrasound radiofrequency, envelope, and log-compressed data based on biophysical and morphological models. The support vector machine with a Gaussian kernel constructed a nonlinear hyperplane, and we calculated the distance between the hyperplane and each feature's data point in multiparametric space. The distance can quantitatively assess a lesion and suggest the probability of malignancy that is color-coded and overlaid onto B-mode images. Training and evaluation were performed on in vivo patient data. The overall accuracy for the most common types and sizes of breast lesions in our study exceeded 98.0% for classification and 0.98 for an area under the receiver operating characteristic curve, which is more precise than the performance of radiologists and a deep learning system. Further, the correlation between the probability and Breast Imaging Reporting and Data System enables a quantitative guideline to predict breast cancer. Therefore, we anticipate that the proposed framework can help radiologists achieve more accurate and convenient breast cancer classification and detection.

Power laws prevail in medical ultrasound

Major topics in medical ultrasound rest on the physics of wave propagation through tissue. These include fundamental treatments of backscatter, speed of sound, attenuation, and speckle formation. Each topic has developed its own rich history, lexicography, and particular treatments. However, there is ample evidence to suggest that power law relations are operating at a fundamental level in all the basic phenomena related to medical ultrasound. This review paper develops, from literature over the past 60 years, the accumulating theoretical basis and experimental evidence that point to power law behaviors underlying the most important tissue-wave interactions in ultrasound and in shear waves which are now employed in elastography. The common framework of power laws can be useful as a coherent overview of topics, and as a means for improved tissue characterization.

Disease-Specific Imaging Utilizing Support Vector Machine Classification of H-Scan Parameters: Assessment of Steatosis in a Rat Model

In medical imaging, quantitative measurements have shown promise in identifying diseases by classifying normal versus pathological parameters from tissues. The support vector machine (SVM) has shown promise as a supervised classification algorithm and has been widely used. However, the classification results typically identify a category of abnormal tissues but do not necessarily differentiate progressive stages of a disease. Moreover, the classification result is typically provided independently as a supplement to medical images, which contributes to an overload of information sources in the clinic. Hence, we propose a new imaging method utilizing the SVM to integrate classification results into medical images. This framework is called disease-specific imaging (DSI) that produces a color overlaid highlight on B-mode ultrasound images indicating the type, location, and severity of pathology from different conditions. In this article, the SVM training was performed to construct hyperplanes that can differentiate normal, fibrosis, steatosis, and pancreatic ductal adenocarcinoma (PDAC) metastases in livers based on ultrasound echoes. Also, cluster centroids for specific diseases define unique disease axes, and the inner product between measured features and any disease axis selected by the SVM quantifies the disease progression. The features were measured from 2794 ultrasound frames using the H-scan analysis, attenuation estimation, and B-mode image analysis. The performance of our proposed DSI method was evaluated for a preclinical model of steatosis ( n = 400 frames). The contribution of each feature was assessed, and the results were compared with ground truth from histology. Moreover, the images generated by our DSI were compared with earlier imaging methods of B-mode, H-scan, and histology. The comparisons demonstrate that DSI images yield higher sensitivity to monitor progressive steatosis than B-mode and H-scan and provide a comparable performance with the histology. For the parameter comparison, DSI and H-scan resulted in similar correlation with histology ( r_s = 0.83 ) but higher than attenuation ( r_s = 0.73 ) and B-mode ( r_s = 0.47 ). Therefore, we conclude that DSI utilizing the SVM applied to steatosis can visually represent the classification results with color highlighting, which can simplify the interpretation of classification compared to the traditional SVM result. We expect that the proposed DSI can be used for any medical imaging modality that can estimate multiple quantitative parameters at high resolution.

Clusters of Ultrasound Scattering Parameters for the Classification of Steatotic and Normal Livers

The study of ultrasound tissue interactions in fatty livers has a long history with strong clinical potential for assessing steatosis. Recently we proposed alternative measures of first- and second-order statistics of echoes from soft tissues, namely, the H-scan, which is based on a matched filter approach, to quantify scattering transfer functions and the Burr distribution to model speckle patterns. Taken together, these approaches produce a multiparameter set that is directly related to the fundamentals of ultrasound propagation in tissue. To apply this approach to the problem of assessing steatotic livers, these analyses were applied to in vivo rat livers (N=21) under normal feeding conditions or after receiving a methionine- and choline-deficient diet that produces steatosis within a few weeks. Ultrasound data were acquired at baseline and again at weeks 2 and 6 before applying the H-scan and Burr analyses. Furthermore, a classification technique known as the support vector machine was then used to find clusters of the five parameters that are characteristic of the different steatotic liver conditions as confirmed by histologic processing of excised liver tissue samples. With the in vivo multiparametric ultrasound measurement approach and determination of clusters, steatotic can be discriminated from normal livers with 100% accuracy in a rat animal model.

H-scan trajectories indicate the progression of specific diseases

PURPOSE

The ability of ultrasound to assess pathology is increasing with the development of quantitative parameters. Among these are a set of parameters derived from the recent H-scan analysis of subresolvable scattering. The emergence of these quantitative measures of tissue/ultrasound interactions now enables a study of the unique trajectories of multiparametric features in multidimensional space, representing the progression of specific diseases over time. We develop the mathematical and visual tools that are effective for classifying, quantifying, and visualizing the steady progression of several diseases from independent studies, all within a uniform framework.

METHODS

After applying the H-scan analysis of ultrasound echoes, we trained a support vector machine (SVM) to classify the unique trajectories of progressive liver disease from fibrosis, steatosis, and pancreatic ductal adenocarcinoma (PDAC) metastasis. Our approaches include the development of trajectory maps and disease-specific color imaging stains.

RESULTS

The multidimensional SVM image classification reached 100% accuracy across the three different studies.

CONCLUSION

H-scan trajectories can be useful to track the progression of multiple classes of diseases, improving diagnosis, staging, and assessing the response to therapy.

Scattering Signatures of Normal versus Abnormal Livers with Support Vector Machine Classification

Fifty years of research on the nature of backscatter from tissues has resulted in a number of promising diagnostic parameters. We recently introduced two analyses tied directly to the biophysics of ultrasound scattering: the H-scan, based on a matched filter approach to distinguishing scattering transfer functions, and the Burr distribution for quantification of speckle patterns. Together, these analyses can produce at least five parameters that are directly linked to the mathematics of ultrasound in tissue. These have been measured in vivo in 35 rat livers under normal conditions and after exposure to compounds that induce inflammation, fibrosis, and steatosis in varying combinations. A classification technique, the support vector machine, is employed to determine clusters of the five parameters that are signatures of the different liver conditions. With the multiparametric measurement approach and determination of clusters, the different types of liver pathology can be discriminated with 94.6% accuracy.

H-scan, Shear Wave and Bioluminescent Assessment of the Progression of Pancreatic Cancer Metastases in the Liver

The non-invasive quantification of tumor burden and the response to therapies remain an important objective for imaging modalities. To characterize the performance of two newly optimized ultrasound-based analyses, we applied shear wave and H-scan scattering analyses to repeated trans-abdominal ultrasound scans of a murine model of metastatic pancreatic cancer. In addition, bioluminescence measurements were obtained as an alternative reference. The tumor metastases grow aggressively and result in death at approximately 4 wk if untreated, but longer for those treated with chemotherapy. We found that our three imaging methods (shear wave speed, H-scan, bioluminescence) trended toward increasing output measures with time during tumor growth, and these measures were delayed for the group receiving chemotherapy. The relative sensitivity of H-scan tracked closely with bioluminescence measurements, particularly in the early to mid-stages of tumor growth. The correlation between H-scan and bioluminescence was found to be strong, with a Spearman's rank correlation coefficient greater than 0.7 across the entire series. These preliminary results suggest that non-invasive ultrasound imaging analyses are capable of tracking the response of tumor models to therapeutic agents.