Scripties UMCG - Rijksuniversiteit Groningen
English | Nederlands

Diffusion Weighted Imaging of Breast Lesions Evaluation of the Intra- and Inter-Observer Agreement in the Region of Interest Demarcation Method

(2016) Langius, E. (Eline)

Purpose: Assessment of the intra- and inter-observer agreement in Diffusion Weighted Imaging (DWI) of the breast to determine the Apparent Diffusion Coefficient (ADC) value by using different Regions of interest (ROI). Secondary, the diagnostic value of the minimal pixel ADC value was evaluated.
Materials and Methods: 96 patients with 92 malignant and 15 benign breast lesions of 0.9 cm2 and larger underwent DWI-MRI. ADC values were calculated per ROI for b-value pair 0-1000 s/mm2. ROI1 was manually drawn to encompass the entire lesion, while staying within its borders. Next to the mean ADC (ADCmean), the minimal pixel ADC (ADCmin) was also observed for ROI1. ROI2 was a standardized circle of 0.3 cm2 (30 pixels) positioned in the middle of the lesion. ROI3 (0.3 cm2) and ROI4 (0.6 cm2) were positioned to yield the lowest ADCmean value within the lesion. When the ROIs were copied from the T1 scan to the DWI, the placement often shifted. Therefor the ROIs had to be moved, this was done once by exact measurements using landmarks and five weeks later by free-hand. The agreement between these methods was tested with the Intraclass correlation coefficient (ICC). To determine the intra- and inter-observer agreement three observers drew ROIs in the first 50 lesions. The ICC was used to determine intra- and inter-observer agreement. The correlation between the ADCmin value of ROI1, as revealed in clinical routine MRI analysis, and the ADCmean value of ROI3 was calculated using Pearson correlation. Diagnostic value was assessed by calculating sensitivity and specificity and the McNemar test was used to compare the difference between the two methods.
Results: The free-hand method showed good agreement with using exact measurement to compensate the displacement for all four ROIs (ICC ranging from 0.81 and 0.93). All ROIs and the minimal ADC value of ROI1 showed excellent intra-observer agreement (ICC ranging from 0.88 to 0.93). Inter-observer agreement was good for all four ROIs (ICC ranging from 0.78 to 0.88). The agreement was fair for the minimal pixel ADC value (0.63).
The ADCmin value within ROI1 was significantly lower than the ADCmean in ROI3, with respective mean values of 0.550 × 10-3 and 0.811 × 10-3 mm2/s. A high correlation between the two measurements was found (r = 0.89). Using a threshold ADC value of 0.715 × 10-3 mm2/s for the ADCmin in ROI1 and 0.100 × 10-3 mm2/s for the ADCmean in ROI3, both showed high sensitivity (96.7%; 98.9%) and moderate to good specificity (80%; 67.7%), respectively. The difference between these values was not significant (p = 0.13).
Conclusion: All four ROIs showed a good repeatability when drawn by three different observers. The ADCmin value within a ROI encompassing the breast lesion on DWI has a similar diagnostic value as a small ROI positioned within the lesion at lowest ADCmean value. Therefore this minimum pixel ADC value, rather than the more elaborate positioning of a circle, can be used in clinical practice.

To top