Tissue Characterization through Ultrasonic Backscatter

Characterization of tissue microstructure with ultrasonic backscatter.

The histogenetic classification (epithelial vs. mesenchymal) and the assessment of benign (adenoma vs. benign mesenchymal) versus malignant (carcinoma vs. sarcoma) behavior of tumors are based on established guidelines following light microscope evaluation of stained tissue sections. These guidelines [Cotran et al., 1994] include the evaluation of:

1. Overall growth pattern of the tumor in relation to the normal tissue (mm resolution);
2. Presence or absence of a capsule around the tumor or a scirrhous response associated with the tumor (hundreds of µm resolution);
3. Organization and growth patterns of the cells within the tumor (10-20 µm resolution);
4. Penetration of cells through basement membrane, into supporting connective tissue, and/or invasion into blood or lymphatic vessels (10-20 µm resolution);
5. Morphology of individual cells within the tumor (nucleus, nucleolus, cytoplasm) (1-5 µm resolution);
6. Presence of mitotic figures, abnormal mitotic figures, or abnormally nucleated cells (1-5 µm resolution).

Conventional ultrasound B-mode images of tumors and surrounding healthy tissues are useful in identifying the first microscopic feature listed by Cotran and often times able to distinguish the second microscopic feature. In order to image the remaining microscopic features, ultrasound B-mode images must be constructed utilizing a much higher frequency range, i.e. 300 MHz and above to image the smallest microscopic features listed above. Operating at these high frequencies is not always feasible due mainly to the large attenuation of the echo signals that results in greatly limited depth of penetration. Quantifying the smaller microscopic features is important to properly identifying the type of tumor (cancerous vs. noncancerous). At present, accurate detection and classification of solid tumors is obtained through light microscope evaluation of biopsied and stained samples, an invasive technique.

This study quantifies small-scale structures through ultrasound backscattering from small particles or perturbations in tissues and solid tumors for the purposes of noninvasive in vivo classification of tumors. Parametric ultrasound B-mode images will be constructed utilizing the properties of the scatterers in the tissues. Specifically, the spatial distribution of the average scatterer diameter and scatterer concentration will be determined to form diagnostic quality images, that is, a parametric image. A composite image may be constructed that superimposes the parameterized image for selected regions of interest (ROIs) on conventional B-mode images.

The noninvasive in vivo experiments of this study focus on forming and interpreting ultrasonic parametric images from tumors grown in rats and from surrounding healthy tissues. The initial B-mode ultrasonic images are constructed over the frequency ranges of 2-12 MHz. The parameterization of tissue structures from backscatter are conducted over the frequency ranges of 5-12 MHz and then superimposed on the existing B-mode images for enhancement. The conventional B-mode images resolve the first two microscopic features listed above. It has been noted [Insana and Hall, 1990] that the most accurate measurements of small-scale structure through scattering occur when the ka value (acoustic wavenumber times the effective scatterer radius) is between 0.5 and 1.2. The frequencies, and associated acoustic wavenumbers, chosen for the parameterization allow for the third and fourth microscopic features listed by Cotran to be measured.

B-mode images of the scanned area are constructed. Regions of interest in the B-mode images are examined for the spectral content of the backscattered RF echoes. The average scatterer size and concentration are determined by comparing the measured normalized power spectrum from each ROI to a theoretical power spectrum. The theoretical power spectrum is modeled according to the shape and distribution of scatterers postulated for the tissue. In soft tissue scattering, the spatial distribution of scatterers is assumed to be statistically stationary so that the distribution can be described in terms of a stochastic function, the spatial autocorrelation. In most cases of soft-tissue scattering, the Gaussian correlation function and associated form factor have been shown to correctly describe tissue structures.

Enhanced B-mode images are created by relating small pixels to average scatterer properties like the average scatterer diameter and average scatterer concentration. Figure 1 shows a conventional B-mode image of a rat with a spontaneous mammary tumor along with an enhanced B-mode image with superimposed pixels related to the average scatterer diameter. Similarly, Figure 2 represents an enhanced B-mode image with superimposed pixels corresponding to the average scatterer concentration. The enhanced B-mode images are made by assuming an attenuation of .7 dB/MHz/cm. The enhanced B-mode images show structures not seen in the ordinary B-mode image and could lead to improved detection and classification of diseased tissues.

Figure 1. Conventional B-mode image of rat chest with spontaneous mammary tumor and enhanced B-mode with
average scatterer diameter information. (Attenuation assumed to be .7 dB/MHz/cm).

Figure 2. Conventional B-mode image of rat chest with spontaneous mammary tumor and enhanced B-mode with
average scatterer concentration information. (Attenuation assumed to be .7 dB/MHz/cm)

A novel estimation scheme was utilized in this work [Oelze et al., 2001] to obtain scatterer parameters assuming a Gaussian form factor. The new estimation scheme reduces the calculation time required over previous estimation schemes [Chaturvedi and Insana, 1996] while maintaining the same expected variance in the estimate. Acoustic evaluation of scatterer properties from the Gaussian form factor will be verified by light microscope evaluation of the tissue microstructure. Ultrasound waves detect changes in the mechanical properties of the tissues while light microscopy detects electromagnetic (dielectric) changes in the tissues. Scattering particles are expected to have both mechanical and electromagnetic differences with surrounding tissues. There will not be a one-to-one correlation between mechanical and electromagnetic property changes in the scattering particles, but the optical characteristics should give an approximation of shape and size that can be related to acoustic estimations.

In total, eight rats that had developed spontaneous mammary tumors were scanned and processed. A comparison of the scatterer property estimates made inside the tumors and outside the tumors showed a distinct difference. Figure 3 shows a histogram of the estimated scatterer properties inside and outside the tumors. In the case of the average scatterer diameters, diameters estimated from inside the tumor were shown to be larger than estimates made outside the tumors. On average there was a 30% increase in diameter sizes estimated inside the tumors. ANOVA statistics were used to test for significant differences between estimates inside and outside and were found to be significant in a majority of cases. For the estimated average concentration, concentrations inside the tumors were seen to be less than concentrations outside the tumors in each rat. Test of significance showed that in all rats (except rat 1), there was a significant difference between estimates made inside and outside the tumors.

Figure 3. Comparison of the estimates of the average scatterer diameter (left) and
average scatterer concentration (right) made inside and outside the tumors.

Further distinctions can be seen by looking at a plot of the feature analysis for the eight rats. Figure 4 shows a plot of the feature analysis for the eight rats (average scatterer diameter versus average scatterer concentration). A clear separation can be seen between estimates made inside the tumor and estimates made outside the tumor. In Fig. 4 the lower right quadrant of the graph is made up of the estimates outside the tumor while the upper left quadrant describes the estimates made inside the tumor. The distinct separation between estimates made inside and outside the tumors may be useful to characterizing and classifying different tissues for disease diagnosis.

Figure 4. Feature analysis plot for the estimates of scatterer diameter versus
scatterer concentration made inside and outside the tumors.

After optimizing the scattering estimation measurement, average scatterer size and concentration estimates for ROIs inside the tumor can be compared with estimates from surrounding tissues. A database of scattering properties for tissues inside tumors and in normal tissues will be gathered for further analysis. The mean value and variance about the mean value of scattering properties will then be used to determine if there exist absolute threshold values for scattering properties in detecting and possibly identifying disease. If no absolute thresholds can be determined, then a relative measure must be employed. The relative measure will take a case-by-case study of scattering properties from each animal by comparing diseased tissues with the normal tissues. Light microscopic data will be important to determine whether the scattering measurement and theory correctly predict the scattering parameters.

Continued
1 | 2 | 3 | 4 | 5 | Next >