U. Scheipers

Ultrasonic multifeature tissue characterization for prostate diagnostics.

A new system for prostate diagnostics based on multifeature tissue characterization is proposed. Radiofrequency (RF) ultrasonic echo data are acquired during the standard transrectal ultrasound (US) imaging examination. Nine spectral, texture, first order and morphologic parameters are calculated and fed into two adaptive neuro-fuzzy inference systems (FIS) working in parallel. The outputs of the FISs are fed into a postprocessing procedure evaluating contextual information before being combined to form a malignancy map in which areas of high cancer probability are marked in red. The malignancy map is presented to the physician during the examination to improve the early detection of prostate cancer. The system has been evaluated on 100 patients undergoing radical prostatectomy. The ROC curve area using leave-one-out cross-validation over patients is A(Z) = 0.86 when distinguishing between hyperechoic and hypoechoic tumors and normal tissue and A(Z) = 0.84 when distinguishing between isoechoic tumors and healthy tissue, respectively. Tumors that are not visible in the conventional B-mode image can be located. Diagnosis of the prostate carcinoma using multifeature tissue characterization in combination with US imaging allows the detection of tumors at an early stage. Also, biopsy guidance and therapy planning can be improved.

Ultrasonic multifeature tissue characterization for the early detection of prostate cancer

The incidence of the prostate carcinoma is one of the highest cancer risks in men in the western world. Its position in cancer mortality statistics is also among the highest. The prostate carcinoma is only curable at an early stage. Therefore, early detection is extremely important. At an early stage the prostate carcinoma is limited to the prostate capsule and can hence be cured performing radical prostatectomy. The different types of diagnostics that are used today (digital rectal examination, transrectal ultrasound and PSA value analysis) lack reliability and are therefore not sufficient. Even a combination of these three methods is not sufficiently reliable. Diagnosis of the prostate carcinoma using multi-feature tissue characterization in combination with ultrasound allows the detection of tumors at an early stage. Also biopsy guidance and planning can be improved. This results in reduced costs for cancer treatment.

A tutorial on the use of ROC analysis for computer-aided diagnostic systems.

The application of the receiver operating characteristic (ROC) curve for computer-aided diagnostic systems is reviewed. A statistical framework is presented and different methods of evaluating the classification performance of computer-aided diagnostic systems, and, in particular, systems for ultrasonic tissue characterization, are derived. Most classifiers that are used today are dependent on a separation threshold, which can be chosen freely in many cases. The separation threshold separates the range of output values of the classification system into different target groups, thus conducting the actual classification process. In the first part of this paper, threshold specific performance measures, e.g., sensitivity and specificity; are presented. In the second part, a threshold-independent performance measure, the area under the ROC curve, is reviewed. Only the use of separation threshold-independent performance measures provides classification results that are overall representative for computer-aided diagnostic systems. The following text was motivated by the lack of a complete and definite discussion of the underlying subject in available textbooks, references and publications. Most manuscripts published so far address the theme of performance evaluation using ROC analysis in a manner too general to be practical for everyday use in the development of computer-aided diagnostic systems. Nowadays, the user of computer-aided diagnostic systems typically handles huge amounts of numerical data, not always distributed normally. Many assumptions made in more or less theoretical works on ROC analysis are no longer valid for real-life data. The paper aims at closing the gap between theoretical works and real-life data. The review provides the interested scientist with information needed to conduct ROC analysis and to integrate algorithms performing ROC analysis into classification systems while understanding the basic principles of classification.

Ultrasonic Tissue Characterization for Prostate Diagnostics: Spectral Parameters vs. Texture Parameters. Sonohistologie für die Prostatadiagnostik: Vergleich von Spektral- und Texturparametern

An ultrasonic multi-feature tissue characterizing system for the detection of prostate cancer is presented. The system is based on the processing of radio frequency (RF) ultrasonic echo data. Data from 100 patients was acquired in a clinical study. Parameters are extracted from the RF echo data and classified using two adaptive network-based fuzzy inference systems (FIS) working in parallel as a nonlinear classifier. Next to spectral parameters, conventional texture parameters are calculated using demodulated and log-compressed echo data. In the first approach, the classifier is trained on both, spectral and texture parameters. In the second approach, the classifier is only trained on texture parameters. Classification results of both approaches are compared and it is demonstrated, that only the use of spectral parameters yields satisfying classification results. Results of a minimum distance classifier (MDC) are presented for comparison with the fuzzy inference system. For the final fuzzy inference systems used in this approach, the area under the ROC curve is between 84% and 86% for the combined approach and between 70% and 74% for the approach based on texture parameters only.

Evaluation of ultrasonic texture and spectral parameters for coagulated tissue characterization

Radiofrequency ablation is a well established, minimally invasive approach for the treatment of tumors. However, at present there is a lack of suitable imaging modalities for accurate online monitoring of the coagulation process. The aim of this work is to evaluate the potential of various tissue characterizing ultrasonic parameters from spectral and spatial domain to differentiate between coagulated and noncoagulated tissue. The calculated parameters include first and second order texture parameters, estimates of attenuation coefficients, spectral parameters (slope, intercept and midband value) and coefficients of autoregressive spectral estimates. As a measure of selectivity of each parameter the area under the receiver operating characteristic (ROC) curve was utilized. The best performing parameters can be used to be processed by a classification system.

Computerized segmentation of blood and luminal borders in intravascular ultrasound

Intravascular ultrasound (IVUS) provides detailed images of normal and abnormal coronary vessel wall morphology and can be used for measuring the lumen area and plaque burden. A prerequisite for this task is the reliable segmentation of IVUS images and discrimination of blood and tissue. At frequencies above 20 MHz the backscatter of blood approaches the same level as backscatter from the vessel wall, which complicates manual segmentation. This work presents an automated scheme for the segmentation of blood in IVUS images. Based on the in vivo acquisition of radio frequency (RF) data, spectral parameters as well as first and second order textural parameters were evaluated. Tissue describing parameters were estimated directly from RF data after dividing each RF frame into numerous regions of interest to allow spatially resolved classification. Parameters originating from different parameter groups were compared with each other and a neuro-fuzzy inference system was trained on up to eight parameters to distinguish blood from tissue using a multi-feature approach. The in vivo results of the multi-feature classifier achieve classification results of A/sub ROC/=0.95 measured as the area under the receiver operating characteristic curve (ROC) and thus prove the reliability of the presented method for the segmentation of blood and tissue with IVUS.

Classification of venous thrombosis combining ultrasound elastography and tissue characterization

Deep venous thrombosis (DVT) is the formation of a blood clot in one of the deep veins of the body, usually in the leg. Common treatment methods include medication with anticoagulants or surgical thrombectomy. Since treatment of DVT succeeds only during the first 7-10 days, exact age determination of DVT is of high importance for an appropriate treatment decision. However, the accuracy of available methods including sonography, phlebography, CT and MRT is often not sufficient. It has been reported that about 30% of all DVT are wrongly staged using common diagnostic modalities and therefore lead to inadequate therapeutic efforts. Therefore alternative and more accurate approaches for staging DVT are desired. Blood clots leading to venous thrombosis undergo an organization process with increasing age. It is known that changes in mechanical stiffness, acoustical properties and appearance in B-mode images accompany the organization process. Therefore several alternative diagnostic approaches, including elastography and ultrasonic tissue characterization, have been proposed in the past. In this work, 22 thrombi of defined age were induced in pigs. Ultrasonic measurements were carried out after surgical resection of the thrombosed vessel segments. Spectral and texture parameters as well as strain estimates obtained using elastography were used to classify thrombosed vessel segments in vitro and thus distinguish between thrombi of age /spl les/ 6 days and age > 6 days. A combination of the best performing parameters was processed by a classification system. Total crossvalidation over specimens was done using Euclidian, Mahalanobis, and maximum-likelihood classifiers. 90% of specimens could be classified correctly using maximum-likelihood classifiers.

Diagnostics of prostate cancer based on ultrasonic multifeature tissue characterization

Ultrasonic multifeature tissue characterization can be used for the computerized detection of prostate cancer tumors. Malignant areas within the prostate can be located with a high degree of accuracy, independent of the diagnostic skills of the operator. Radiofrequency ultrasonic echo data of the prostate are captured using standard ultrasound equipment. Several features describing the histological characteristics of the underlying tissue are estimated after dividing each ultrasound data frame into up to 1000 regions of interest and compensating the echo data for diffraction and system dependent effects. Spectral features, textural features of first and second order, clinical variables and morphological descriptors are applied. Two parallel network-based fuzzy inference systems classify and separate the regions of interest. Subsequent morphological analysis combines clusters within malignancy maps, which consist of conventional grey-scaled B-mode images with areas of high cancer probability marked in red. In a clinical study, RF ultrasonic echo data of 100 patients have been recorded. Prostate slices with histological diagnosis following radical prostatectomies are used as standard. The mean area under the ROC curve is between 0.84, for isoechoic tumors, and 0.86, for hypo- and hyperechoic tumors. Standard deviations are as low as 0.02, for isoechoic tumors, and 0.01, for hyper- and hypoechoic tumors. All three spectral approaches, evaluated conventional Fourier spectrum parameters, generalized spectrum parameters and AR parameters, yield comparable classification rates for the underlying prostate data sets.

Neuro-fuzzy inference system for ultrasonic multifeature tissue characterization for prostate diagnostics

The incidence of the prostate carcinoma is one of the highest cancer risks in men in the western world. Its position in cancer mortality statistics is also among the highest. The different types of diagnostics that are used today lack reliability and are therefore not sufficient. Diagnosis of the prostate carcinoma using multifeature tissue characterization in combination with ultrasound allows the detection of tumors at an early stage and thus can aid the conducting physician in finding a diagnosis. Spatially resolved parameters and contextual information are used for the classification. Next to hypo- and hyperechoic tumors, also isoechoic tumors can be visualized.

Real Time Strain Imaging — a new Ultrasonic Method for Cancer Detection: First Study Results

Prostate tumors can have a higher mechanical hardness than the surrounding tissue. During the digital rectal exam this can be used not only to detect the hypertrophy but also localized hardenings. The examination by digital palpation is inaccurate and even in combination with PSAvalue and a transrectal ultrasonic examination the result is often not reliable. Ultrasound strain imaging is able to measure and visualize the elastic properties of a tissue region and hence is an adequate supplement for commonly used diagnostic procedures. We have developed a real time system for elastographic mechanical tissue assessment of the prostate which can be used for the transrectal ultrasonic examination for navigation and diagnosis. During the examination a sequence of ultrasonic images is acquired while the organ is slightly compressed by the ultrasound probe. Using a numerical analysis of image pairs of the acquired sequence the tissue strain is calculated which represents the spatial elasticity distribution of a specific cross-section of the organ and which are able to distinguish hard areas in the tissue. We present results from several patients. which show, that real time strain imaging is able to detect tumor-like areas which are inconspicuous in the bmode image. The results correspond to the histological specimens. After the evaluation of 130 patients using a prospective study we found the specificity for cancer detection to be approximately 84% and a sensitivity of approximately 76 %. Furthermore the tumor location and extend was correctly predicted in most of the investigated patients using our real time strain imaging. SIGNAL PROCESSING Strain imaging was first described by Ophir in 1991 [1], but could not easily be clinically applied so far, because the described method had no real time capability. To use it in a clinical setting we invented a time efficient algorithm, called “phase root seeking” [2], which in a current system is able to calculate up to 30 strain images per second using a conventional desktop PC. Similar to [1], time shifts are estimated using a discrete number of windows at discrete depths. The time shift τm,k of the k-th window of two A-lines centered around tk = k∆T is estimated by the following iterative formula L k m k m , 1 , 0 , , − τ = τ (1)