Tumour heterogeneity assessment using histogram analysis of IVIM-based diffusion and perfusion characteristics of cervical cancer
Jose Angelo Udal Perucho1, Elaine Yuen Phin Lee1, and Queenie Chan2

1Department of Diagnostic Radiology, The University of Hong Kong, Hong Kong, Hong Kong, 2Philips Healthcare, Hong Kong, Hong Kong

Synopsis

Histogram analysis of intravoxel incoherent motion (IVIM) diffusion-weighted MRI (DWI) could be a promising quantitative approach in assessing tumour heterogeneity. We retrospectively studied thirty-five treatment-naïve patients with cervical cancer who had IVIM MRI examinations to determine to the value of IVIM histogram analysis, as a means of assessing tumour heterogeneity, in relationship with clinical staging. We observed statistically significant differences in most histogram parameters of f (perfusion fraction) between patients with early and locally advanced disease but only three histogram parameters of D (true diffusion coefficient) were statistically different in patients with early and locally advanced cervical cancer.

Purpose

Advanced cervical cancer is hypothesised to be more heterogeneous than early disease leading to diverse intratumoral diffusion and perfusion characteristics. The purpose of this this study was to investigate the value of histogram analysis of intravoxel incoherent motion (IVIM) based diffusion and perfusion characteristics of cervical cancer as a measure of tumour heterogeneity and its relationship with clinical staging.

Methods

MRI images from thirty five patients with squamous cell carcinoma (SCC) of the cervix were retrospectively analysed. These patients were treatment-naïve and part of a previous ethically-approved study. They were recruited for diffusion-weighted (13 b values, b = 0-1,000 s/mm2) and standard pelvic MRI performed on 3.0T Achieva TX scanner, Philips Healthcare.

DWI was acquired using single-shot spin-echo echo-planar imaging in free breathing with background body signal suppression. Bi-exponential analysis was performed on the DWI data to generate parametric maps of the true diffusion coefficient (D) and perfusion fraction (f). Regions of interest were placed on the cervical cancer using ImageJ (ImageJ version 1.49, National Institute of Health) and histogram parameters (skewness, kurtosis, percentiles, interquartile range (IQR) and interdecile range (IDR)) were extracted using Histogram Tool (Histogram Tool, Philips Healthcare).

Clinical staging was defined according to the International Federation Gynecology and Obstetrics (FIGO) staging system. Patients were divided into having early disease (FIGO Ib – IIa, low FIGO group) and locally advanced disease (FIGO IIb – IVa, high FIGO group). Student’s t-test was used to compare IVIM histogram parameters between low and high FIGO groups. Statistical significance assumed at p < 0.05.

Results

There were 7 patients in the low FIGO group and 28 patients in the high FIGO group. The histogram parameters of D and f were tabulated in Tables 1 and 2. There were positive correlations between histogram skewness and kurtosis in both D (R2=0.862) and f (R2=0.909) (Figure 1). In f, all histogram parameters, except the 10th percentile and IQR, showed significant differences between low and high FIGO groups; especially in skewness (p=0.045) and kurtosis (p=0.037). As for D, only the 90th percentile, IDR, and IQR, showed significant differences between the low and high FIGO groups.

Discussion

A normal distribution would have skewness and kurtosis equal to zero, where increasing skewness results in greater asymmetry and increasing kurtosis results in heavier tails. A change of distributions in D and f can be qualitatively assessed as a change in the shape and geometry of the histogram (Figure 2). Based on the hypothesis that advanced cervical cancer is more heterogeneous than early disease, we expect more variable diffusion and perfusion characteristics in cervical cancer with high FIGO stage relative to that of low FIGO stage. Increasing heterogeneity implies simultaneous broadening of peaks and right shifted peaks; this can be quantitatively observed as a positive correlation between histogram skewness and kurtosis (Figure 1).

Previous studies have shown that histogram parameters of DWI and other MRI techniques had high correlation with their respective histological grading and were powerful in predicting treatment response in cervical cancer 1,2; however, studies for the assessment of tumour heterogeneity were relatively sparse.

In the context of cervical cancer, f was able to evaluate the perfusion within the tumour microcirculation 3,4. The vascular density and geometry were shown to be affected in cervical cancer; furthermore, tumour vascularity was found to be prognostic of tumour aggressiveness and disease outcome 5,6.

The differences found in the histogram parameters of f between different FIGO groups were likely reflections of tumour heterogeneity in its microcirculation. The shape of the histogram curve corroborated with the effect of tumour heterogeneity in different stages of disease.

Surprisingly, only three histogram parameters of D (90th percentile, IDR, and IQR) were statistically different between the low and high FIGO groups, while other parameters lacked distinction, in which the reason was unclear. It might suggest that, although histogram analysis of D could measure heterogeneity in tumour cellularity, the effect of tumour heterogeneity of D was not as strong as those exhibited by f or the findings could be related to the small sample size in this study.

Conclusion

Histogram analysis of IVIM-based diffusion and perfusion parameters is a potentially useful technique in quantifying tumour heterogeneity in cervical cancer, especially in f, which exhibited differences in different clinical FIGO stages.

Acknowledgements

References

1. Downey K, Riches SF, Morgan VA, Giles SL, Attygalle AD, Ind TE, Barton DPJ, Shepherd JH, deSouza NM. Relationship Between Imaging Biomarkers of Stage I Cervical Cancer and Poor-Prognosis Histologic Features: Quantitative Histogram Analysis of Diffusion-Weighted MR Images. American Journal of Roentgenology 2013; 200:314-320.

2. Just N. Improving Tumour Heterogeneity MRI Assessment with Histograms British Journal of Cancer 2014; 111:2205-2213.

3. Lee EYP, Yu X, Chu MMY, Ngan HYS, Siu SWK, Soong IS, Chan Q, Khong PL. Perfusion and Diffusion Characteristics of Cervical Cancer Based on Intravoxel Incoherent Motion MR Imaging – A Pilot Study. European Radiology 2014;24:1506-1513.

4. Lee EY, Hui ES, Tse KY, Kwong WK, Chang TY, Chan Q, Khong PL. Relationship between intravoxel incoherent motion diffusion-weighted MRI and dynamic constrast-enhanced MRI in tissue perfusion of cervical cancers. Journal of Magnetic Resonance Imaging 2015;42(2):454-459.

5. Schelenger K, Höckel M, Mitze M. Schäffer U. Weikel W, Knapstein PG, Lambert A. Tumor Vascularity – A Novel Prognostic Factor in Advanced Cervical Carcinoma. Gynecological Oncology 1995;59:57-66.

6. Révész L, Siracka E. A Morphometric Study of Vascularisation in Uterine Cervix Cancers. Cytometry 1984;5:442-444.

Figures

Figure 1. Correlations between histogram parameters, skewness and kurtosis of D and f. Skewness, measures the asymmetry of a distribution while kurtosis, measures the peakedness of a distribution relative the normal distribution.

Figure 2. Smooth curve renderings of representative histogram distributions from the two FIGO groups in (A) Diffusion Coefficient D and (B) Perfusion Fraction f.

Table 1. Histogram parameters of D in low and high FIGO groups

Table 2. Histogram Parameters of f in low and high FIGO groups



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1587