Vignesh Arasu MD, Presenter: Nothing to Disclose

David Newitt PhD, Abstract Co-Author: Nothing to Disclose

Dorota J. Wisner MD, PhD, Abstract Co-Author: Nothing to Disclose

John Kornak PhD, Abstract Co-Author: Nothing to Disclose

Bonnie N. Joe MD, PhD, Abstract Co-Author: Investigator, Sentinelle Medical, Inc

Nola Michiko Hylton PhD, Abstract Co-Author: Research grant, Hologic, Inc

Signal enhancement ratio (SER), a three time-point contrast kinetics marker for cancer, has been shown to retrospectively identify 45% of false positives on BIRADS 4/5 breast MRI lesions while preserving sensitivity when using empiric thresholds for percent enhancement (PE) and SER kinetics. This study investigated 1) how variation in PE and SER thresholds affect diagnostic accuracy and 2) how these thresholds can be optimized.

A retrospective review was performed of BIRADS 4/5 breast MRI lesions in 2008 yielding a sample size of 72 biopsied lesions (18 malignant, 54 benign) in 64 patients. Images were acquired using gadolinium-enhanced 1.5T MRI with a high spatial-resolution sequence. Three time-points were acquired every 3.5 minutes: precontrast (S0), early postcontrast (S1), and late postcontrast (S2), where S represents signal intensity. ROI box volumes were drawn around lesions by researchers guided by the radiology report for lesion location but blinded to pathology. Voxels were first filtered by PE threshold and then by an SER kinetic threshold (defined by the ratio [S1-S0]/[S2-S0]), and finally summed to give a filtered lesion volume. PE threshold was varied between 50-200% and SER kinetic threshold between 0-2.0. The primary predictor used was the SER volume fraction, defined as the filtered lesion volume (PE=50-200%, SER=0-2.0) divided by the total lesion volume (PE=50%, SER=0). An ROC analysis was performed to identify optimal thresholds ranges when sensitivity is preserved at > 90%.

Peak AUC was observed at 0.84, corresponding to PE=90% and SER=0.8. Within 20% of the peak AUC, thresholds ranged from PE=60-130% and SER=0-1.4. When identifying thresholds optimized to sensitivity of 90-100%, SER was observed to reliably operate at AUC > 0.75. In this optimized area, thresholds ranged from PE=60-110% and SER=0-1.4. False positives were identified in 35-51% of lesions at >90% sensitivity and identified in 17-47% of lesions at 100% sensitivity.

These findings suggest that SER is able to operate robustly over a broad range of PE or SER kinetic thresholds to maintain both 1) high overall diagnostic accuracy and 2) high sensitivity while still identifying a clinically significant number of false positives.

Three-time point kinetic models can decrease false positives while maintaining high sensitivity in breast MRI using thresholds robust over a broad range.

Arasu, V, Newitt, D, Wisner, D, Kornak, J, Joe, B, Hylton, N, Threshold Optimization of Enhancement Kinetics Criteria on Breast MRI.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11003480.html