Quantification of tumor uptake with molecular breast imaging (MBI)

Objectives Molecular Breast Imaging (MBI) uses a small dual-headed semiconductor-based gamma camera in a mammographic configuration to obtain high resolution functional images of 99mTc-sestamibi uptake in the breast. Currently, MBI images are purely qualitative. We have developed a novel scatter and attenuation correction algorithm to quantify uptake with MBI; the accuracy and robustness of this methodology is presented.

Methods A 7 cm thick rectangular phantom containing 99mTc-water simulating breast tissue and fillable spheres simulating tumors were imaged with a GE 750b MBI system. Six sphere sizes ranging in diameter from 9mm to 27mm were imaged with 3 different sphere to background ratios (SBR) of 3.5, 2.6, and 1.7 and located at 3 different depths of 2, 4, 6 cm within the water-bath (for a total of 54 unique tumor scenarios). Phantom images were also acquired in-air under scatter- and attenuation-free conditions which provided a ground truth. The projection images were decay-corrected to a global time point and scatter-corrected using a custom dual-energy-window technique (Bache et al, J Nucl Med 56(3):45, 2015). To estimate true counts, T, from each tumor, the geometric mean (GM) of the counts within a prescribed ROI from the two projection images was calculated as T=√(C_1 C_2 e^μx F), where C is the counts within the square ROI circumscribing each sphere on detectors 1 and 2, µ is the linear attenuation coefficient of water, and x is the detector separation. We have introduced the factor F to account for the background activity; 4 unique F definitions corresponding to standard GM, background subtracted GM, MIRD Primer 16 factor, and a novel “volumetric” factor were investigated. Error in quantified T was estimated with respect to in-air conditions and calculated as 100 x (calculated counts - known counts)/(known counts). The quantitative accuracy of T using the different GM definitions was calculated as a function of SBR, depth, and sphere size. The effects of ROI size and threshold-based ROI (to more accurately mimic clinical conditions where the tumor size is not known) on the quantitative accuracy was also investigated.

Results Mean errors (95%CI range) for all 54 unique scenarios (3 SBRs x 6 sphere sizes x 3 depths) were 167% (-127% - 460%), -16.3% (-37.7% - 5.1%), 18.9% (-16.9% - 54.7%), and 3.8% (-24.4 - 16.9%) for the standard GM, background subtraction GM, MIRD 16 formalism, and our volumetric GM, respectively. Standard GM results are omitted for the remainder of the discussion due to gross inaccuracies. The MIRD 16 GM varied the most under differing SBR with COV of 209%, while the background subtracted GM and our volumetric GM was least sensitive to SBR with COVs of 25% and 19%, respectively. Our volumetric GM was the least sensitive to variations in sphere diameter with COV of 28%, compared to 37% and 110% for the background subtracted GM and MIRD 16 GM, respectively. All 3 non-standard GM formalisms were insensitive to depth with COV of 3% - 6%. Varying ROI size with respect to sphere size showed a predictable trend with all 3 non-standard GM formalisms; smaller ROI size led to lower estimates for true counts. However, our volumetric GM outperformed the other approaches. The mean error (95% CI) for each formalism over 5 ROI sizes ranging from 80% to 120% of sphere size was -15.2% (-33.6% - 3.2%), 15.7% (-23.3% - 54.7%), and 0.84% (-28.2% - 29.9%). The accuracy of FWTM threshold-defined ROIs was -1.37% for our volumetric GM.

Conclusions Using our novel geometric mean formalism, we were able to obtain accurate estimates of tumor uptake in MBI images (mean error <5% with 95%CI ~20%) under a variety of SBRs, tumor sizes, and depths. Our results show that ROI size may play a role in quantitative accuracy, however this systematic bias may be alleviated by using a thresholding technique for tumor delineation. Research Support: GE Healthcare.