Deployment Of A Potential Dose-Reduction Algorithm In Molecular Breast Imaging- Impact Of Reader Experience On Its Optimization

3101 Objectives: Molecular Breast Imaging, performed with Tc-99m sestamibi and small CZT-based detectors is a valuable technique for detection of breast cancers occult on mammography. Previous work has shown that a noise reduction algorithm (ClearMBI) allows for a 50% reduction in image acquisition time, while yielding improved image quality relative to a standard acquisition. Optimization of the filter settings for ClearMBI in our lab was accomplished by experienced breast radiologists and nuclear medicine technologists. Deployment of this algorithm in sites with little or no experience with MBI may not achieve these optimal results. Hence, the objective of this study was to determine if this optimization was dependent on user experience with MBI. Methods: Two groups of readers were selected. Group 1 comprised a total of 4 breast imagers/scientists and 4 nuclear medicine technologists each with more than 2 years’ experience with MBI (>2000 patient studies/reader). Group 2 comprised 2 nuclear medicine physicians and 4 nuclear medicine technologists with no experience with MBI. Each reviewer was presented with a test set of 20 MBI exams (each containing a lesion) representing a range of count densities. For each exam, only the original half-dose image was displayed, and the reviewer was allowed to freely adjust the ClearMBI filter setting until their preferred lesion conspicuity was achieved. The average filter setting as a function of counts per pixel in the MBI images was obtained for each group and compared to the previously derived equation that was used to validate the algorithm. Results: Figure 1 shows the correlation between the selected value for the denoising filter and the mean counts per pixel in the MBI images. There was close agreement on the optimal filter setting as a function of image counts for the experienced readers and technologists, with correlation coefficients of R = 0.91 and 0.82, respectively. By comparison, the nuclear medicine physicians tended to over smooth the MBI images, whereas the technologists were more in line with the experienced users, but tended to under denoise the images. Both inexperienced groups had poorer correlation coefficients (R = 0.73, 0.72). Lack of consistency in use of the algorithm may reduce acceptance by physicians and hinder its adoption into clinical practice. Implementation of ClearMBI into a clinical practice, that is initiating an MBI service, should include a nomogram that enables a consistent application of the algorithm based on the count density in the MBI images and is not subject to results. Figure 1. Correlation between denoising filter setting and count density in the MBI images, for the 4 different groups (experienced physicians / technologists, and inexperienced physicians / technologists). Conclusions: Implementation of the ClearMBI algorithm is strongly dependent on user experience. The algorithm would allow institutions experienced in MBI to reduce image acquisition time by 50% while yielding improved image quality relative to a standard acquisition. For new installations with little or no experience in MBI, it is recommended that staff use a nomogram based on the previously published filter settings in order to provide more consistent image quality across all users.