Apparent Diffusion Coefifcient Histogram Of MR Diffusion Weighted Imaging: Evaluation On Pathological Features Of Uterine Cervix Cancer

The size, tumor histologic types, differentiation, lymphovascular space invasion andlymph node involvement impact the prognosis of cervical cancer. It is of great clinicalsignificance to non-invasively evaluate the pathological features of cervical cancer(advanced stage in particular) for guiding and monitoring chemoradiation by means ofmedical imaging. Water diffusion restriction caused by cell membrane and cellularityalteration can be detected by DWI; celluar structure change can therefore be predictedduring the pathological process by the technique. Literatures have documented that DWIwas applied to the early detection, staging, differentiating, treatment outcome assessment,and even pathological types prediction of cervical cancer. ADC histogram reveals thedistribution of ADC values of whole tumor voxel-by-voxel, which indicates theheterogeneity of diffusion of tissue. This technique has been documented as useful forgrading gliomas, differentiating paediatric brain tumors, sub-classification of low-gradeglioma histologic subtypes, and even distinguishing biologically different regions of braintumors. Some parameters obtained from the histogram may also serve as biomarkers oftreatment response of tumors. Furthermore, scan-rescan reproducibility of the whole brainADC histogram has been verified.The first part of this study investigates the correlation between pathological featuresand ADC distribution of cervical cancer, while the second part focuses on the repeatabilityanalysis of the measurement by the technique.Part1Evaluation of pathological features of cervicalcancer by apparent diffusion coefficient histogram ofMR DWIObjective: This study employs apparent diffusion coefficient (ADC) histogramobtained from MR diffusion weighted imaging (DWI) to evaluate the heterogeneity ofdiffusion in cervical cancer, whose correlation with pathological features and diagnosticperformance in differentiating between malignant and benign tissues are analyzed.Methods: The study was sanctioned by the institutional ethics board, and all patients’agreements with signatures were acquired. From June2011to July2013,73patients (age:33-69y, mean±SD:50.5±8.6y) with FIGO IB-IIIB stages cervical cancer wereincluded in the patient group. Histological diagnosis was obtained by surgery in34patients, and by biopsy in39patients.38patients (age:38-61y, mean±SD:44.3±6.5y) withuterine leiomyoma who underwent radical hysterectomy were included in the controlgroup. All patients both in the2groups were examined by routine MR sequences, DWIand dynamic contrast enhanced imaging (DCE-MR) before histological comfirmation.Data analysis was performed by a radiologist with8years experience in pelvic MRimaging diagnosis. With reference to T2WI and DCE-MR, area of tumor (patient group) orcervical canal (control group) was defined on each section of sagittal ADC maps togenerate ADC histogram automatically by post-processing software (Siemens Syngo).ADC histogram of entire tumor or cervical canal was built with all sections combinedusing SPSS16.0. According to histological results, ADCmean, ADCmedian, the25thpercentile of ADC, the75th percentile of ADC, skewness and kurtosis of the histogramwere compared between groups of histological types, differentiation, lymphovascularspace invasion, FIGO staging, tumor size and ages. Levene’s test was employed for testinghomogeneity of variance. Student’s t-test or One-Way ANOVA, and non-parametric testincluding Mann-Whitney U-test (2groups) and Kruskal-Wallis test (3groups) was used. Atest of equilibrium (Chi-squared test) for patient constituent ratio was implemented forthose groups (≥2) showing statistical significance at every parameter of histogram.Significance level=0.05. Receiver operating characteristic curve (ROC) was employed forthe diagnostic performance of ADCmean, ADCmedian, the25th percentile of ADC, the75th percentile of ADC, skewness and kurtosis in differentiating between cervical cancerin IB stage and the control group. For those groups of pathological types, differentiationand lymphovascular space invasion, ROC analysis was performed for parameters showingstatistical significance. The optimal cut-point was calculated according to Yoden index(sensitivity+specificity-1). Diagnostic accuracy was classified as excellent when the areaunder the ROC curve (AUC) was0.9-1, good0.8-0.9, fair0.7-0.8, poor0.6-0.7, andfailed0.5-0.6.Results: In comparison between cervical adenocarcinoma and squamous cellcarcinoma, ADCmean (1183.43±185.15×106mm2s1vs1081.67±158.30×106mm2s1), ADCmedian (1143.81±213.37×106mm2s1vs1019.33±150.86×106mm2s1), the25th percentile of ADC (988.71±187.34×106mm2s1vs870.01±140.20×106mm2s1) and skewness (0.73±0.56vs1.02±0.37) showed statisticallydifferent (P <0.05), with an AUC of0.66,0.66,0.69and0.60respectively. Between wellor moderately differentiated and poorly differentiated, ADCmedian (1088.28±187.55× 106mm2s1vs1001.89±152.32×106mm2s1) and skewness (0.82±0.47vs1.12±0.34)were statistically different (P <0.05), with an AUC of0.64and0.71. No parameter wassignificantly different between patients with and without lymphovascular space invasion.Differences in the skewness approached significance when the patients were classified asIB/IIA and IIB/IIIA-B (0.80±0.49vs1.07±0.36), and different levels of tumor size(0.78±0.56vs0.90±0.34vs1.15±0.38)(P <0.05). In groups with different ages,ADCmean (1166.17±171.97×106mm2s1vs1128.56±172.43×106mm2s1vs1029.70±145.39×106mm2s1), ADCmedian (1107.00±200.48×106mm2s1vs1078.05±165.22×106mm2s1vs969.47±150.53×106mm2s1) and the75th percentile of ADC(1338.14±174.10×106mm2s1vs1294.51±207.30×106mm2s1vs1183.83±176.98×106mm2s1) showed statistically different (P <0.05). In comparison between IB stagecervical cancer and control groups, ADCmean (1099.01±206.41×106mm2s1vs1621.28±249.72×106mm2s1), ADCmedian (1045.27±212.08×106mm2s1vs1639.63±242.08×106mm2s1), the25th percentile of ADC (898.79±191.27×106mm2s1vs1420.91±238.76×106mm2s1), the75th percentile of ADC (1261.58±226.84×106mm2s1vs1841.11±265.37×106mm2s1), skewness (0.80±0.51vs-0.26±0.41)and kurtosis (1.38±1.27vs0.33±0.26) revealed statistically different (P <0.05), and thecorresponding AUC was0.94,0.97,0.96,0.95,0.93,0.80.Conclusion: Pathological features of cervical cancer can be non-invasively assessedby ADC histogram. The heterogeneity of diffusion revealed by ADC histogram may reflectthe celluar structure changes due to the variation of pathological type, differentiation andtumor size.Part2Analysis of Measurement Repeatability ofCervical Cancer ADC HistogramObjective: To analyze the measurement repeatability of cervical cancer ADChistogram.Methods:20patients (age:42-50y，mean±SD:45.9±2.4y) were randomlyenrolled from the patient group for the analysis. Imaging data of each patient wasevaluated twice with3days interval by a radiologist with8years of pelvic MR imagingdiagnosis. With reference of T2WI and DCE-MR, area of tumor was defined on eachsection of sagittal ADC maps to generate ADC histogram automatically by post-processingsoftware (Siemens Syngo). ADC histogram of entire tumor was built with all sections combined using SPSS16.0, by which ADCmean, ADCmedian, the25th percentile of ADC,the75th percentile of ADC, skewness and kurtosis were then calculated. Difference (d),mean, standard deviation (SD), within-subject SD (Sw) and repeatability for the2measurement were obtained using Excel (version:2003). Bland-Altman plot wasgenerated by MedCalc (Version:9.6.2.0) using the difference and average values betweenthe2measurement, and the95%limits of agreement were obtained.Results:1. Repeatability of tumor size:95%limits of agreement (-14.7,13.3) contain100%(20/20) of the difference scores, and Repeatability=13.05.2. ADCmean:95%limitsof agreement (-33.8×106mm2s1,68.1×106mm2s1) contain95%(19/20) of the differencescores, and Repeatability=59.90×106mm2s1.3. the25th percentile of ADC:95%limitsof agreement (-42.1×106mm2s1,52.9×106mm2s1) contain95%(19/20) of the differencescores, and Repeatability=47.48×106mm2s1.4. ADCmedian:95%limits of agreement(-52.4×106mm2s1,40.0×106mm2s1) contain100%(20/20) of the difference scores, andRepeatability=46.62×106mm2s1.5. the75th percentile of ADC:95%limits ofagreement (-49.6×106mm2s1,46.6×106mm2s1) contain100%(20/20) of the differencescores, and Repeatability=46.95×106mm2s1.6. Skewness:95%limits of agreement(-0.28,0.25) contain95%(19/20) of the difference scores, and Repeatability=0.26.7.Kurtosis:95%limits of agreement (-1.5,2.0) contain90%(18/20) of the difference scores,and Repeatability=1.8.Conculsion: Measurements of ADC value and skewness from cervical cancer ADChistogram were found to be repeatable; however, there might be bias for the measurementof kurtosis.