Spectrum-analysis and neural networks for imaging to detect and treat prostate cancer.

Conventional B-mode ultrasound currently is the standard means of imaging the prostate for guiding prostate biopsies and planning brachytherapy to treat prostate cancer. Yet B-mode images do not adequately display cancerous lesions of the prostate. Ultrasonic tissue-type imaging based on spectrum analysis of radiofrequency (rf) echo signals has shown promise for overcoming the limitations of B-mode imaging for visualizing prostate tumors. This method of tissue-type imaging utilizes nonlinear classifiers, such as neural networks, to classify tissue based on values of spectral parameter and clinical variables. Two- and three-dimensional images based on these methods demonstrate potential for guiding prostate biopsies and targeting radiotherapy of prostate cancer. Two-dimensional images are being generated in real time in ultrasound scanners used for real-time biopsy guidance and have been incorporated into commercial dosimetry software used for brachytherapy planning. Three-dimensional renderings show promise for depicting locations and volumes of cancer foci for disease evaluation to assist staging and treatment planning, and potentially for registration or fusion with CT images for targeting external-beam radiotherapy.

Three-dimensional ultrasound analyses of the prostate.

Although conventional ultrasonography has proven to be clinically useful for depicting many types of cancerous lesions, it cannot distinguish reliably between cancerous and noncancerous tissue of the prostate. Therefore, conventional transrectal ultrasonography (TRUS) is used primarily for general evaluations of the gland and for guiding biopsies based on clearly imaged anatomic features such as the capsule, seminal vesicles, and urethra. Spectrum analysis extracts ultrasound signal parameters associated with biopsy-proven tissue types, and these parameters are then classified using neural network tools such as learning vector quantization, radial basis, and multilayer perceptron algorithms. Classification of cancerous and noncancerous prostate tissue using neural networks produces receiver operating characteristic (ROC) curves of 0.87 +/- 0.04 compared with 0.64 +/- 0.04 for conventional ultrasonography. To image the prostate using these methods, parameter values are computed at each pixel location, then translated into a score for the likelihood of cancer using a look-up table generated using the best classification algorithm. The score for cancer likelihood is expressed as a gray-scale or color value, and the resulting image may be useful to guide biopsies or therapy. Changes in parameter or score values over time potentially can be used to assess progression of disease or efficacy of therapy.

Interlaboratory comparison of ultrasonic backscatter, attenuation, and speed measurements.

In a study involving 10 different sites, independent results of measurements of ultrasonic properties on equivalent tissue-mimicking samples are reported and compared. The properties measured were propagation speed, attenuation coefficients, and backscatter coefficients. Reasonably good agreement exists for attenuation coefficients, but less satisfactory results were found for propagation speeds. As anticipated, agreement was not impressive in the case of backscatter coefficients. Results for four sites agreed rather well in both absolute values and frequency dependence, and results from other sites were lower by as much as an order of magnitude. The study is valuable for laboratories doing quantitative studies.

In vitro B-mode ultrasonographic criteria for diagnosing axillary lymph node metastasis of breast cancer.

Axillary lymph node status is an important factor for staging and treatment planning in breast cancer. Our study was performed in vitro on a node-by-node basis to evaluate the ability of B-mode ultrasonographic images to distinguish metastatic from nonmetastatic nodes. Immediately prior to histologic examination, individual dissected axillary nodes were scanned in a water bath using a 10 MHz B-mode ultrasonographic transducer. Four B-mode features (size, circularity, border demarcation, and internal echo) were evaluated for their ability to distinguish metastatic from nonmetastatic lymph nodes. Lymph node metastasis was indicated by (1) a large size (i.e., a length of the longest axis of 10 mm or greater); (2) a circular shape (i.e., the ratio of the shortest axis to the longest axis between 0.5 and 1.0); (3) a sharply demarcated border compared with surrounding fatty tissue; and (4) a hypoechoic internal echo, with obliteration of the fatty hilum. The sensitivity and specificity were compared for all combinations of features. We examined 84 histologically characterized axillary nodes from 27 breast cancer patients, including 64 nonmetastatic and 20 metastatic nodes. Of the criteria cited, circular shape was the best single feature for distinguishing metastatic from nonmetastatic nodes (sensitivity, 65%; specificity, 73%). The best combination of sensitivity (85%) and specificity (73%) was obtained using the criterion that a lymph node contained cancer when at least three positive features were present. The present in vitro study demonstrated that the sensitivity and specificity of B-mode ultrasonography for diagnosing lymph node metastasis were lower than 90%. Therefore, B-mode ultrasonography may not be an optimal noninvasive screening method for diagnosing axillary lymph node metastasis in breast cancer patients, particularly under in vivo clinical conditions.

In vitro diagnosis of axillary lymph node metastases in breast cancer by spectrum analysis of radio frequency echo signals.

Axillary lymph node status is of particular importance for staging and managing breast cancer. Currently, axillary lymph node dissection is performed routinely in cases of invasive breast cancer because of the lack of accurate noninvasive methods for diagnosing lymph node metastasis. We investigated the diagnostic ability of ultrasonic tissue characterization based on spectrum analysis of backscattered echo signals to detect axillary lymph node metastasis in breast cancer in vitro compared with in vitro B-mode imaging. Immediately after surgery, individual lymph nodes were isolated from axillary tissue. Each lymph node was scanned in a water bath using a 10-MHz instrument, and radio frequency data and B-mode images were acquired. Spectral parameter values were calculated, and discriminant analysis was performed to classify metastatic and nonmetastatic lymph nodes. Forty histologically characterized axillary lymph nodes were enrolled in this study, including 25 nonmetastatic and 15 metastatic lymph nodes. A significant difference existed in the spectral parameter values (slope and intercept) for metastatic and nonmetastatic lymph nodes. Spectral parameter-based discriminant function classification of metastatic vs. nonmetastatic lymph nodes provided a sensitivity of 93.3%, specificity of 92.0%, and overall accuracy of 92.5%. In comparison, B-mode ultrasound images of in vitro lymph nodes provided a sensitivity of 73.3%, specificity of 84.0%, and overall accuracy of 80.0%. Receiver operating characteristic (ROC) analysis comparing the efficacy of both methods gave an ROC curve area of 0.9888 for spectral methods, which was greater than the area of 0.8980 for B-mode ultrasound. Hence, this in vitro study suggests that the diagnostic ability of spectrum analysis may prove to be markedly superior to that of B-mode ultrasound in detecting axillary lymph node metastasis in breast cancer. Because of these encouraging results, we intend to conduct an investigation of the ability of spectral methods to classify metastatic axillary lymph nodes in vivo.

In vitro investigation of lymph node metastasis of colorectal cancer using ultrasonic spectral parameters.

Lymph node involvement is one of the major factors affecting the prognosis of colorectal cancer. Various imaging methods, including ultrasound and computed tomography, are not sufficiently sensitive or specific for reliably determining lymph node involvement. We investigated the feasibility of using ultrasonic tissue characterization (UTC) based on spectrum analysis of backscattered echo signals for diagnosing lymph node metastasis of colorectal cancer in vitro. Forty lymph nodes, including 17 metastatic and 23 nonmetastatic nodes, from 11 colorectal cancer operations were investigated. Lymph nodes were scanned using a clinical instrument; B-mode imaging was performed for each lymph node, and radiofrequency (RF) data were acquired. The UTC parameters, slope and intercept, were calculated from the normalized power spectrum of the backscattered echo signals from each lymph node. The mean values of UTC parameters of metastatic and nonmetastatic lymph nodes were compared. The accuracy of UTC in distinguishing metastatic from nonmetastatic lymph nodes was calculated using discriminant analysis. Receiver operating characteristic (ROC) analysis was performed to compare the classification efficacy of UTC and B-mode ultrasound. UTC parameters demonstrated a significant difference in parameter values between metastatic and nonmetastatic lymph nodes. The overall accuracy in diagnosing the lymph node metastasis was 87.5% for UTC and 77.5% for B-mode ultrasound. ROC analysis produced an ROC curve area of 0.92 or 0.89 for UTC (depending on the performance-assessment algorithm) and 0.84 for B-mode ultrasound, which indicated that UTC performed markedly better than B-mode ultrasound in diagnosing metastatic lymph nodes. The advantages of UTC over conventional B-mode ultrasound in discriminating metastatic lymph nodes from nonmetastatic lymph nodes are extremely encouraging, and warrant an in vivo UTC study.

Determination of carotid plaque risk by ultrasonic tissue characterization.

This in vitro study investigated the ability of ultrasonic tissue characterization (UTC) to discriminate between plaques from asymptomatic and symptomatic patients and to compare UTC findings with quantitative measurements of plaque morphology. A total of 34 plaque specimens removed at carotid endarterectomy were scanned transversely at intervals of 1 mm, and compared to tissue cross-sections examined by optical microscopy employing computer-assisted planimetry. UTC was performed by spectral analysis of backscattered radiofrequency signals. The slope, intercept and total power parameters of the spectrum were evaluated. Discriminant analysis was used to compare the ability of the UTC spectral parameters and morphological constituents to correctly classify plaques according to their symptom group membership. UTC correctly classified 88.2% of the plaques. Thrombus was present in 93.9% of the plaques, and there was little difference in the morphological constituents of plaques from asymptomatic and symptomatic patients. Morphological constituents correctly classified 60.7% of the plaques. We conclude, in this preliminary study, that UTC can discriminate between carotid plaques from asymptomatic and symptomatic patients with moderate accuracy, despite a similarity in their morphological composition. UTC discrimination is not related to differences in the type or amount of morphological constituents in the plaques.

Carotid plaque typing by multiple-parameter ultrasonic tissue characterization.

We evaluated the ability of ultrasonic tissue characterization (UTC), based on backscattered echo signals, to distinguish among the components of advanced carotid plaques. We performed spectral analysis of echo signals acquired from human carotid endarterectomy specimens in vitro to calculate three parameters of the calibrated power spectrum: slope, intercept and total power for fibrous, lipid pool and thrombus constituents of plaque. Plaque constituents were identified histologically. We evaluated classification efficacy by discriminant function analysis. Slope and intercept parameters alone provided correct classification in 92.5%, 57.6% and 72.4% of fibrous, lipid pool and thrombus plaque components, respectively. Slope, intercept and total power used in combination improved classification of the three tissue types to 93.0%, 69.7% and 81.0%. The overall proportion of correctly classified tissue regions increased from 84.5% to 88.0% by the combined use of the three parameters. The improvement in classification that occurred when we included total power as a third parameter suggests that ultrasound plaque components may not consist solely of small, randomly distributed isotropic scatterers. Our ability to identify plaque thrombi provides motivation for future studies of parameter-based imaging methods for identifying such plaque that presents an increased risk of embolic neurologic ischemic events.

Statistical framework for ultrasonic spectral parameter imaging.

This study examines the statistics of ultrasonic spectral parameter images that are being used to evaluate tissue microstructure in several organs. The parameters are derived from sliding-window spectrum analysis of radiofrequency echo signals. Calibrated spectra are expressed in dB and analyzed with linear regression procedures to compute spectral slope, intercept and midband fit, which is directly related to integrated backscatter. Local values of each parameter are quantitatively depicted in gray-scale cross-sectional images to determine tissue type, response to therapy and physical scatterer properties. In this report, we treat the statistics of each type of parameter image for statistically homogeneous scatterers. Probability density functions are derived for each parameter, and theoretical results are compared with corresponding histograms clinically measured in homogeneous tissue segments in the liver and prostate. Excellent agreement was found between theoretical density functions and data histograms for homogeneous tissue segments. Departures from theory are observed in heterogeneous tissue segments. The results demonstrate how the statistics of each spectral parameter and integrated backscatter are related to system and analysis parameters. These results are now being used to guide the design of system and analysis parameters, to improve assays of tissue heterogeneity and to evaluate the precision of estimating features associated with effective scatterer sizes and concentrations.

Determining the acuteness and stability of deep venous thrombosis by ultrasonic tissue characterization.

PURPOSE: The intent of the study was to determine whether ultrasonic tissue characterization (UTC) could indicate acuteness and stability of deep venous thrombosis (DVT) of the lower extremities. METHODS: Thrombi presenting as filling defects on color Doppler imaging in the common or superficial femoral or popliteal veins in 50 extremities in 45 patients with DVT were studied. Acute DVT was less than 4 days duration, and chronic DVT was greater than 21 days duration. UTC analysis of parameters from the normalized power spectrum of backscattered ultrasound signals from venous filling defects was performed. This spectrum approaches a straight line, and its basic parameters, slope, and Y-intercept are related to scatterer size, concentration, and the square of the scatterer-to-medium acoustic impedances. Ten of the DVT extremities were reexamined at 1 week to assess UTC changes that would indicate thrombus instability. RESULTS: Acute DVT (19 of the 50 extremities) could be distinguished from chronic DVT, mainly on the basis of significantly higher intercept values for the acute group, which were 11.6 relative decibels (dBr) higher than those of the chronic DVT group. Discriminant linear analysis of the two parameters indicated a sensitivity of 94.7% and specificity of 90.3% in correctly diagnosing acute DVT. In a small sample of 10 extremities reexamined at 1 week, acute DVT extremities showed a mean 9.4 dBr decrease in intercept values with no significant change in slope. CONCLUSIONS: UTC distinguished clinically defined acute from chronic DVT. In a small series of extremities, UTC revealed significant instability of acute thrombi in a selected patient population.

Roles of hematocrit and fibrinogen in red cell aggregation determined by ultrasonic scattering properties.

Parameters of the power spectrum of backscattered echoes were applied to quantitatively evaluate red cell aggregation in vitro. Human red cell suspensions were circulated in a closed loop of tubing, and ultrasonic, radiofrequency, echo-signal data were obtained using a 10-MHz transducer. Data acquisition was performed at 30-s to 1-min intervals for 5 min after flow stoppage. Two parameters of the normalized power spectrum of the echo signals, spectral slope and Y-intercept, were computed, and estimates of two scattering properties, the scatterer size and acoustic concentration were calculated from these parameters using equations based on scattering theory. Size and acoustic concentration were observed as they changed over time after the stoppage of flow. The key findings were that hematocrit affected the rate of cell aggregation while fibrinogen controlled aggregate size and acoustic concentration.

Experimental assessment of spectrum analysis of ultrasonic echoes as a method for estimating scatterer properties.

In vitro experiments using weak scatterers ranging in size from mean longest diameter of 26.9 to 83.0 microns were performed to test the validity of theoretical predictions for scatterer size and concentration derived from normalized power spectrum parameters of ultrasonic backscatterer echoes. Scatterers consisting of cell clusters were suspended in collagen gel and scanned by a 10 MHz transducer system. Optical measurements validated theoretical predictions that (1) slope value is a function solely of scatterer size; (2) intercept value is a function of scatterer size and concentration; and (3) midband fit value increases as scatterer concentration increases, and, to a lesser extent, as size increases. These results were obtained under relatively ideal conditions of minimal attenuation and scatterer spacing (not closer than two scatterer diameters) and were consistent with the assumptions underlying the scattering theory.

Differentiation of breast tumors by ultrasonic tissue characterization.

The ability of ultrasonic tissue characterization to differentiate and classify benign and malignant breast tissues in vivo in patients with palpable breast masses and in vitro in excised breast tissue was evaluated. One-hundred and twenty-four in vivo and 89 in vitro studies were performed using a technique of UTC based on parameters from the power spectrum of backscattered echoes. Sensitivities and specificities for diagnosing carcinoma were 86 and 84% for in vivo studies and 94 and 92% for in vitro studies. These UTC parameters provided threshold values for color-coding breast lesion images. The results of this preliminary investigation suggest that UTC provides a basis for assessing more accurately lesions suspected of being malignant prior to biopsy and possibly for evaluating breast lesions noninvasively.

Age determination of experimental venous thrombi by ultrasonic tissue characterization.

PURPOSE: The ability of ultrasonic tissue characterization based on radiofrequency signal processing to detect compositional differences in thrombi of varying ages was evaluated in vivo. METHODS: Thrombi were produced in 49 jugular veins of 26 anesthetized 18 to 20 kg pigs by partial ligation and application of direct electric current. Thrombi were imaged 30 minutes after formation and 1, 7, and 14 days later with a color Doppler ultrasound scanner that identified the thrombi, and acquired radio frequency data for ultrasonic tissue characterization analysis. Ultrasonic tissue characterization used two parameters from the normalized power spectrum, slope, and intercept, which are related to scatterer size, scatterer concentration, and acoustic-impedance differences between scatterers and surrounding medium. Previous in vitro studies demonstrated that lower slope and higher intercept values correlated with greater cellularity and more-dense fibrin mesh. Histologic examination was performed for each time period. The values of slope and intercept for each timed observation were compared by a multilinear discriminant analysis. RESULTS: There were no statistical differences between day 0 and day 1. Statistically-significant differences in ultrasonic tissue characterization parameters were seen between all other time intervals with p values < 0.01. Older thrombi tended to demonstrate higher slope and lower intercept values. These ultrasonic tissue characterization changes correlated with a red cell and fibrin-mesh density reduction, which was confirmed by histologic findings and was indicative of partial spontaneous thrombolysis. The degree of spontaneous thrombolysis provides an estimate of the age of thrombi. CONCLUSION: Ultrasonic tissue characterization is capable of distinguishing age differences in thrombi in an animal model and has the potential for noninvasive application in clinical diagnosis.

Effect of perfusion and blood content on ultrasonic backscattering of liver tissue.

The purpose of this study was to determine the effect of blood flow perfusion and red cell content on ultrasonic scattering by liver tissue. Data acquisition for ultrasonic tissue characterization (UTC) employing analysis of the backscattered echoes from the power spectrum was obtained from the same region of pig liver tissue under four conditions: 1) normal perfusion in situ, 2) ischemia in situ in the living pig, 3) ischemia in situ immediately postmortem, and 4) immediately after excision of the liver. Discriminant function analysis was used to evaluate differences in the two basic parameters from the normalized power spectrum: slope and intercept. Normal perfused liver had significantly higher intercept values and lower slope values than liver under the other three conditions. Excised liver showed the lowest intercept and highest slope values (p < 0.01). These experiments indicate that differences in perfusion produce significant differences in ultrasonic scattering by liver tissue (ischemia caused a 3 dB drop in intercept amplitude). Normal or ischemic in vivo and in vitro liver tissue is associated with different patterns of ultrasonic scattering, and scattering data under these various circumstances are not equivalent.

Ultrasonic tissue characterization of experimental venous intimal hyperplasia.

Ultrasonic tissue characterization (UTC) employing slope and Y-intercept parameters from the normalized power spectrum of backscattered echoes was employed in vivo to study compositional changes in the walls of pig jugular veins in which thrombi were experimentally induced. Light microscopy revealed these changes to be intimal hyperplasia with an early predominance of smooth muscle cells and a later mixture of smooth muscle cells and collagen deposits. UTC distinguished intimal hyperplasia from previously reported data from luminal thrombosis UTC. Furthermore, UTC was able to discriminate between early (predominantly smooth muscle cells) and older (smooth muscle cells plus collagen deposits) intimal hyperplasia. The study suggests that intimal hyperplasia in the experimental model used may be organized thrombus and that UTC may be able to follow both the development of wall changes as well as luminal changes occurring in venous thrombosis.

Simulation studies of ultrasonic backscattering and B-mode images of liver using acoustic microscopy data.

Data obtained from a scanning laser acoustic microscope (SLAM) were used to examine several aspects of ultrasonic backscattering from the liver. Phase interferograms from normal and abnormal human-liver specimens were digitized, and a series of algorithms was used to compute images of propagation velocity within the specimens. The propagation velocity images were then employed to simulate A- and B-mode results. These initial simulations were used to investigate how ultrasonic echo signals are related to tissue microstructure. Among the topics examined were B-mode speckling, frequency and beamwidth effects, and angulation dependencies.

Cellularity and fibrin mesh properties as a basis for ultrasonic tissue characterization of blood clots and thrombi.

This in vitro study was designed to evaluate the ability of ultrasonic tissue characterization (UTC) based on power spectrum analysis of backscattered radio-frequency echo signals to distinguish two prominent variables of thrombi: cellularity (primarily red cell content) and fibrin-mesh density. Six types of clots simulating thrombus components were prepared by varying red-cell and platelet concentrations and shear forces during clotting. Data were acquired with a linear-array transducer, digitized, and analyzed in terms of slope and intercept parameters obtained from normalized power spectra of radio-frequency echo signals. Increased cellularity and fibrin-mesh density both produced lower slope and higher intercept values, which permitted statistically significant discrimination of cellularity and mesh density in the six types of clots analyzed. Shearing forces and (to a lesser degree) platelet concentrations increased fibrin-mesh density. This study suggests that UTC based upon the power spectrum of echo signals may be used to detect and follow compositional differences that have clinical relevance in the diagnosis and follow-up of thrombi.

Ultrasonic tissue characterization of blood clots.

Ultrasonic tissue characterization based on an analysis of the power spectrum of backscattered signals obtained with ultrasound was used to distinguish morphologic components of blood clots. The three morphologic features for which discrimination was attempted were loose fibrin, red-cell, and dense fibrin clots. The UTC was able to distinguish the morphologic blood components tested. This in vitro work was based on the analysis of parameters related to ultrasound-tissue interaction and on inferences related to the physical properties of scatterer properties (scatterer size, scatterer concentration, and ratio of scatterer to medium acoustic impedances). The ability to distinguish these blood-clot components suggests that UTC may be able to distinguish red from white thrombi and to assess the structures and changes within thrombi associated with the age of the thrombus, their mechanical properties, and treatment monitoring.

Comparison of theoretical scattering results and ultrasonic data from clinical liver examinations.

A theoretical analysis of soft-tissue ultrasonic scattering has been used to formulate specific results describing spectral parameters for tissue characterization. Results are applicable to clinical liver examinations. Three spectral parameters are mathematically expressed in terms of acoustic attenuation and the effective sizes, concentrations, and relative acoustic impedances of tissue scatters. Results from a clinical data base are shown to agree well with analytical results for each spectral parameter. Agreement is found for: spectral shapes; effects of attenuation; and correlations between parameters. Images of three spectral parameters are presented and their gray-scale features are evaluated with reference to analytical results.