Z-Score transformation of ADC values: A way to universal cut off between malignant and benign lymph nodes

Purpose To determine the possibility of a universal cut off value between benign and malignant lymph nodes in patients with tumour by Z-Score transformation method. Materials and methods Diffusion weighted imaging, ADC measurements of malignant and benign lymph nodes of 6 studies (4 body parts), conducted for 5 times, in two institutions with variable technical details were analyzed in their original value as well as the standardized Z-Score value. The standardized Z-Score value was obtained by subtracting the population mean of the control group from an individual raw score and then dividing the difference by the population standard deviation of the control group. General cut off values were obtained by both Mega-analysis by receiver operator characteristic curve analysis, when data from the 6 studies were combined and Meta-analysis with weighting coefficients and cut off values of the six individual studies. Sensitivity, specificity and accuracy with cut offs from individual studies, meta-analysis and mega-analysis were calculated. Kappa test was performed to assess the consistency of diagnostic test accuracy, between optimized cut offs of individual studies and the proposed universal cut offs obtained from meta-analysis and mega-analysis. Results The ADC values of benign and malignant lymph nodes are significantly different, but with large overlap across the studies. The overlap can be minimized by Z-Score transformation. The result of ROC analysis of the collective Z-Score transformed ADC values of 6 studies was superior to that of the collective original ADC values (sensitivity: 87.4% versus 67.2%, specificity: 90.5% versus 87.9%, accuracy: 89.6% versus 81.4%). The universal Z-Score cut off from Meta-analysis is also better than the original ADC cut off (sensitivity: 82.8% versus 76.3%, specificity 92.6% versus 62.9%, accuracy 89.6% versus 67.1%). Applied to the individual studies, the universal transformed Z-Score cut offs produced superior consistency with the individual optimal cut offs (individual and meta Z-Score: 0.7228–0.9793; individual and mega Z-Score: 0.7111–0.9169) compared with the universal original ADC cut offs (individual and meta ADC: 0.3030–1.0000; individual and mega ADC 0.3268–0.9618). Conclusion Z-Score transformation could minimize inter-study variations due to heterogeneity of MR systems and sequence parameters, and provide a more consistent universal cut off value between benign and malignant nodes across studies.