Twenty patients were scanned with three directional DWI and six directional DTI. The Wilcoxon signed-rank test for the three directional and six directional median nADC values showed no significant difference for the median (p = 0.94), 25th (p = 0.55), or 75th (p = 0.60) percentile nADC values within the non-enhancing region. There was also a very strong correlation between the three and six directional median ADC (r = 0.97, p <0.001) and nADC (r = 0.964, p <0.001) values. Therefore, the patients scanned with three directional DWI and the patients scanned with the six directional DTI were analyzed together.
The median, 25th, and 75th percentiles were calculated for the nADC and nFA values, within the NEL. Normalization was performed in order to combine three and six directional data. The differences between the groups were maintained whether or not the normalization was performed. The correlation coefficient between the ADC and nADC values was r
= 0.9725, p
<0.0001. Other studies have suggested ADC varies with age (24
) and brain or tumor volume (25
). In this study, the correlation between ADC and age showed a weak but significant correlation for all patient data combined (r
= −0.315, p
<0.022), but no significant correlation for patients with ODs (r
= −0.352, p
<0.099), patients with ACs (r
= −0.192, p
<0.477), or patients with OAs (r
= 0.025, p
<0.932). The correlation between age and ADC values in NAWM was also assessed for any of the subtypes separately and gave OD (r
= 0.0608, p
<0.78), AC (r
= −0.337, p
<0.20), OA (r
= −0.154, p
<0.60) or for all patients combined (r
= −0.099, p
<0.48) showing no correlation. The correlation was also assessed for the FA values in NAWM with OD (r
= 0.076, p
<0.79), AC (r
= −0.126, p
<0.77), OA (r
= 0.383, p
<0.35) and all patients combined (r
= 0.119, p
<0.53) showing no correlation. There was no correlation between ADC and tumor volume for patients with ODs (r
= −0.01, p
<0.96), AC (r
= −0.30, p
<0.23), OA (r
= 0.29, p
<0.32) and all subtypes combined (r
= 0.045, p
<0.73). The tumor volume mean ± standard deviation (std) were 56.6 ± 63.3, 49.7. ± 39.7, and 47.1 ± 33.6 cc for patients with OD, AC, and OAs. A Wilcoxon rank-sum test shows no significant difference in volume between the groups, OD versus
= 0.7426), OD versus
= 0.9127), and AC versus
There was a larger variation in the NAWM values for the FA than the ADC, with a variation of 25% in FA as compared to 13% in ADC. The variation in FA likely arose from the difference in noise from three to six directions and field strengths and the known variation across patients. Therefore, normalization was more important for FA values than for ADC values. Notably, although there was a larger variation in the NAWM FA, there was no significant difference between subtypes (see ) within the NAWM.
Descriptive statistics for the NAWM ADC values and non-enhancing lesion nADC values for patients with oligodendroglioma (OD), astrocytoma (AC), and oligoastrocytoma (OA) subtypes followed by the p-value from the Wilcoxon rank-sum test
The median, 25th, and 75th percentiles for nADC and nFA for the NEL region are shown in . The median nADC OD values were significantly lower (p <0.0001) than the median nADC AC values, with the OAs falling in the middle range, as shown in . This also held true for the 25th and 75th percentile nADC values showing significant differences with p <0.0004 and p <0.0001, respectively. The median nFA OD values were significantly higher (p = 0.005) than the median nFA AC values, with the OA nFA values falling in the middle range, as shown in . This also held for the 25th and 75th percentile nFA values with significant differences of p = 0.004 and p = 0.006, respectively.
Boxplots of the median (a) nADC values and (b) nFA values within the NEL for patients with grade II oligodendroglioma (OD), astrocytoma (AC), and oligoastrocytoma (OA).
In addition to the median, 25th, and 75th percentiles the std and the coefficient of variability (cvb) defined here as the (std/median).*100 were analyzed. The median stds and cvb for the ADC within the NEL of grade II ODs, ACs, and OAs were 230, 283, 265 × 10−3 mm2/s and 19, 20, 18%, respectively. The OD has a significantly lower std than that of the AC (p = 0.015), but when normalized by the median, there are no significant differences between any of the subtypes.
The data were also analyzed by summing the nADC histograms and nFA histograms of each patient per subtype, shown in , along with the NAWM histograms which overlap for nADC and nFA. The grade II OD nADC values were lower than the grade II AC, with the grade II OA falling in between. The nFA values in the grade II ODs were higher than those in the grade II OAs and ACs. The overlap of the histograms between subtypes was greater in the nFA than in the nADC.
Sum of (a) nADC histograms and (b) nFA histograms within the NAWM of all patients and the NEL of all the patients with oligodendrogliomas (OD), astrocytoma (AC), and oligoastrocytoma (OA).
The data were also analyzed by examining the median nFA and the median nADC. There appears to be a significant correlation of all subtypes (r = −0.80, p <0.0001, n = 33), and separately for patients with OD (r = −0.739, p = 0.0025, n = 14) and AC (r = −0.9227, p = 0.0011, n = 8) but no correlation for OA (r = −0.2621, p <0.530, n = 8). High FA values were found to correlate to low ADC values for all subtypes (r = −0.791, p <00001), and separately for patients with OD (r = −0.585, p = 0.0172) and AC (r = −0.876, p = 0.002) but no correlation for OA (r = −0.427, p = 0.2912). Visual inspection of maps of the 25th, 50th, and 75th percentiles revealed that the median values typically reflected the central tumor while the edge of the tumor contained the 25th percentile nADC and 75th percentile nFA values. Although the correlation across patients was strong, the correlation within patients was not as strong. There was a large range of correlation coefficient values, ranging from −0.12 to −0.81, with a median correlation of −0.43, −0.49, −0.45, and −0.44 for patients with ODs, ACs, OAs, and all subtypes, respectively. The within-patient NAWM correlation coefficients were significantly lower than those within the tumor, with a median correlation of −0.17, −0.18, −0.23, and −0.18 for patients with ODs, ACs, OAs, and all subtypes, respectively.
Logistic regression analysis was used to examine the ability of the nADC and nFA in subtyping patients with OD versus AC. The most optimal cutoff value for the median nADC was 1.84, misclassifying 2 of 23 ODs, 3 of 16 ACs with an overall accuracy of 87%. The most optimal cutoff value for the median nFA was 0.41, misclassifying 1 of 14 ODs, 1 of 9 ACs, with an overall accuracy of 91%. The same two patients misclassified from the nFA analysis alone were misclassified when both the nADC and nFA were both included in the analysis.