Dynamic MRI of Invasive Breast Cancer: Assessment of Three Region-of-interest Analysis Methods

Purpose: In this study, three region-of-interest (ROI) analysis methods based on operator-defined and semiautomated sampling of pharmacokinetic breast maps of contrast uptake are described. The observer variability and impact of the methods on the estimated enhancement characteristics of invasive cancer are also presented.

Method: Fifty-four women with invasive breast cancer underwent dynamic Gd-DTPA-enhanced MRI. ROIs were drawn by two observers on parametric images obtained from compartmental modeling of the dynamic data. Three methods were used: (a) An irregular ROI was drawn to include as much of the enhancing part of the tumor as possible (large ROI); (b) a 12 pixel circular ROI was placed at the most rapidly enhancing part of the large region (small ROI); and (c) a computer algorithm interrogated the large region pixel by pixel using a 9 pixel square mask and selected the region with the highest mean parameter value (semiautomated ROI).

Results: Significant observer variability and bias were found in the enhancement measurements using the large ROI method. There was no observer bias associated with the other methods, but the variability of the small ROI method was substantial. An almost perfect observer agreement was achieved using the semiautomated method. The small and semiautomated ROI methods produced significantly higher enhancement ratios than the large ROI method, especially in grade III carcinomas.

Conclusion: Variability is inherent in subjective ROI analysis, but the semiautomated method of ROI selection and sampling of parameter images of the breast is an efficient and reliable alternative that may allow better standardization of the MR technique.