In this study we have performed measurements of strain magnitude and brightness intensity across the ultrasound depicted border of glial tumours, with subsequent analysis of differences in contrast between the image modalities. The results of the analyses show a significantly higher contrast between tumour tissue and presumed normal tissue in the strain images, as compared to the B-mode images. From Table we observe that the mean contrast for all B-mode measurements is 0.39 while it is 0.60 for the strain magnitude measurements, which is 54
% higher. One interpretation of the results could be that ultrasound strain imaging should be the preferred image modality to use during surgery of brain tumours, since the strain images provide better discrimination (higher contrast) between the tumour tissue and the normal brain tissue. However, in a clinical setting there are still several key issues to solve before the surgeons can use ultrasound strain imaging as a practical tool for identification of the resectable tumour tissue. With the processing parameters applied in this study the strain images generally appear noisier than the conventional B-mode images. This is partly introduced by the processing of the data where e.g. the differentiation of the calculated time delays in the axial direction typically introduces strain values with alternating polarity and a spiking appearance in the strain image. The processing is also prone to decorrelation of the echo signal due to low signal levels (hypoechoic regions in the B-mode image) or "out of plane" tissue motion causing loss of temporally coherent signals. This may cause the processing to produce false results with abnormally high strain values. Also, our method for estimation of time delays assumes that the delay is smaller than the sampling time Ts, i.e. that the tissue velocity is low compared to the number of frames acquired per second. If this assumption is not met the processing may produce incorrectly high strain values.
In our processed strain images we have indeed seen that noise can be present in parts of the image. This is typically seen in regions with low intensity in the B- mode image, for example when imaging homogenous tissue like the brain stem and deeper white brain matter that appear hypoechoic compared to other brain tissue. However, our measurements are intentionally performed in the transition zone from tumour to presumed normal brain tissue. In this short distal range we expect the data to be least influenced by noise, with the B-mode intensity ranging from the hyperechoic tumour to the isoechoic areas with presumed normal tissue. The inspection of the strain magnitude curves did not indicate any abrupt change of signal level within the spatial distance analysed, as could be expected if the strain processing produced invalid results.
It can be argued that the measurements performed in the transition zone from tumour to normal tissue impose a selection bias for the contrast analysis. This is the region that is of interest to the surgeon, but it is also the region where we should expect the strain images to be least affected by noise. The contrast measurements are only valid for analysis of image contrast between glial tumour tissue and adjacent normal tissue. It should not be interpreted to represent differences in contrast resolution between the image modalities in general.
The methodology for the analysis of image contrast in the peripheral parts of tumour involves a subjective assessment of the approximate position of the depicted tumour border and manual reading of the displayed strain magnitude and brightness curves. Even if the implemented method of analysis is not fully automatic, the measurements were obtained by following a standardized procedure, as outlined in the Methods section. Quantitative image quality measures will usually imply some subjective decisions about where to perform the analysis in the image. It is therefore difficult to establish a method without some kind of manual intervention. However, the calculation of additional measures like e.g. the contrast-to-noise ratio (CNR), or signal-to-noise ratio (SNR) would increase the robustness of the image assessment and should be considered in future studies [12
]. It would also have been interesting to address intra- and interobserver variability of the measurements, which was not performed in this study.
As discussed above there are different factors that may have affected the measurements. However, we have found the obtained measurements to be quite robust and we believe that the differences in contrast found between ultrasound strain magnitude and B-mode intensity should represent actual differences between the image modalities.
Ultrasound strain imaging in brain surgery is a quite novel approach and we have not found other studies performing a similar comparison between strain images and conventional B-mode images. It is therefore difficult to compare our results with previous findings. Some studies have however explored the use of strain ratio for diagnostic purposes, but the similar ratio for B-mode intensity has not been reported. The strain ratio is a quantitative index but should not be considered as an objective diagnostic
parameter as its value may be heavily dependent on which regions are selected for comparison and is therefore prone to variations between observers and within the patient population, which has also been pointed out by others [13
]. It should be noted that the contrasts calculated in our study are not intended to serve a diagnostic purpose; the sole purpose is the pairwise comparison between the ultrasound modalities.
The diagnostic value
of ultrasound strain imaging of brain tumours has not been assessed in this study. This would require a comparison between image findings and histology, which was not available for the current study. Glial tumours are diffuse infiltrating and tumour cells are likely to be present also beyond the border zone seen in the ultrasound B-mode image [14
]. Scattered tumour cells are likely to be present in the isoechoic regions interpreted to be mainly normal brain tissue, but to a substantially less extent than in the hyperechoic regions. The calculated contrasts should therefore represent differences in magnitude (strain/brightness) between areas in the brain predominated by glial tumour cells and areas predominated by normal brain cells, respectively.
The results obtained should provide a rationale for further technical developments and investigations of methods for real-time intraoperative ultrasound strain imaging of brain tumours. The ultrasound strain magnitude images possess a higher contrast between tumour and normal brain tissue in the peripheral parts of the tumour than the conventional B-mode images. This suggest that the surgeon may use imaging of strain to improve detection of remaining tumour towards the end of surgery, compared to using conventional ultrasound imaging alone.