Automating tumor classification with pixel-by-pixel contrast-enhanced ultrasound perfusion kinetics

Abstract

Contrast-enhanced ultrasound (CEUS) enables highly specific time-resolved imaging of vasculature by intravenous injection of ∼2 μm gas filled microbubbles. To develop a quantitative automated diagnosis of breast tumors with CEUS, breast tumors were induced in rats by administration of N-ethyl-N-nitrosourea. A bolus injection of microbubbles was administered and CEUS videos of each tumor were acquired for at least 3 min. The time-intensity curve of each pixel within a region of interest (ROI) was analyzed to measure kinetic parameters associated with the wash-in, peak enhancement, and wash-out phases of microbubble bolus injections since it was expected that the aberrant vascularity of malignant tumors will result in faster and more diverse perfusion kinetics versus those of benign lesions. Parameters were classified using linear discriminant analysis to differentiate between benign and malignant tumors and improve diagnostic accuracy. Preliminary results with a small dataset (10 tumors, 19 videos) show 100% accuracy with fivefold cross-validation testing using as few as two choice variables for training and validation. Several of the parameters which provided the best differentiation between malignant and benign tumors employed comparative analysis of all the pixels in the ROI including enhancement coverage, fractional enhancement coverage times, and the standard deviation of the envelope curve difference normalized to the mean of the peak frame. Analysis of combinations of five variables demonstrated that pixel-by-pixel analysis produced the most robust information for tumor diagnostics and achieved 5 times greater separation of benign and malignant cases than ROI-based analysis.

Figure 1

Integrated intensity of B-mode frames pre and post image registration. The intensity of each pixel was integrated over all frames within video 2 (a) pre and (b) post image registration. With the integrated intensity algorithm, motion within the videos should cause features to blur. Although video 2 had the greatest percent decrease in sum of squared difference scores after motion correction (Table TABLE III.), the two integrated intensity images looked nearly identical; features can be seen sharply in both images.

Figure 2

(Color online) Time-intensity curve preprocessing. Sample processing of the compressed video signal from a single representative pixel from a typical (a) benign and (b) malignant video. TICs for each pixel were either low pass filtered (LPF) or the envelope curve (EC) was detected prior to linearization. LPF allowed characterization of the overall wash-in and wash-out perfusion patterns, while EC and the unprocessed signal allowed detection of transient fluctuations in concentration of the contrast agent.

Figure 3

(Color online) Performance of LDA classifier. Histograms showing results of the LDA classification of five variables in LDA space. The top combination from each of the groups are presented in the order of increasing Fisher discriminant criterion, which quantifies the separation of the classes by the separation of the means normalized by the sum of the standard deviations: (a) nonpixel-by-pixel (18.8); (b) mean based (35.8); (c) standard deviation based (38.4); (d) pixel-by-pixel (106); and (e) all variables (281). Note the differences in the scale of the x-axis. The combinations of variables chosen for the five groups were (a) non-P × P (AUCWO_M, MTT_M, WOT80_M, AUOC_M, TOA_M); (b) mean-based (MTT_M, WOT50_M, DIWO15_M, ISDN_M, AUEC_M); (c) SD-based (FWHM_S, MTT_S, WOT50_S, OEDNM_S, TOA_S); (d) P × P (PGWI_S, ISDN_S, PGWONM_S, PSWINM_S, coverage); (e) all (FWHM_M, PW_M, WOT80_M, OEDN_S, coverage). AUCWO: area under the wash-out curve; MTT: mean transit time; WOT50, 80: wash-out time to 50 and 80% of peak; AUOC: area under the original TIC curve; TOA: time of arrival; DIWO15: drop of intensity wash out to 15 s; ISDN: standard deviation of intensity normalized to peak; AUEC: area under the envelope curve; FWHM: full width at half maximum; OEDNM: envelope curve difference normalized to the mean of the peak frame; PGWI: peak gradient wash-in; PGWONM: peak gradient wash-out normalized to the peak frame; PSWINM: peak slope wash-in normalized to peak frame; PW: peak width; _M: mean within the ROI; _S: standard deviation within the ROI.

Figure 4

(Color online) Correlation of top combination of variables. This figure shows the correlations of the five variables that provided the best discrimination of benign and malignant tumors according to the Fisher discriminant criterion. Out of the five variables, three pairs of variables were highly correlated with each other, while the remaining seven pairs were predominantly uncorrelated. FWHM: full width at half maximum; PW: peak width; WOT80: wash-out time to 80% of peak; OEDN: envelope curve difference normalized to the mean of the peak frame; _M: mean with the ROI; _S: standard deviation with the ROI.

Figure 5

Enhancement differences due to manual bolus injections. (a) Peak enhancement frame from video 15 (rat 8, BR38). (b) Peak enhancement frame from video 16 (rat 8, BR55). These two images were acquired approximately 30 min apart following the same bolus injection and imaging protocol. In video 15, the significantly reduced enhancement was most likely due to operator error during injection of microbubbles.