RESUMO
PURPOSE: AccuBoost is a complex non-invasive brachytherapy procedure for breast treatment. This technique requires a radiation oncologist to manually select applicator grid position and size by overlaying transparencies over a mammographic image to encompass surgical clips and resected tumor bed. An algorithm was developed in MATLAB™ to automate the selection of round applicators based on surgical clip position. MATERIAL AND METHODS: A total of 42 mammograms belonging to 10 patients were retrospectively analyzed. Images were pre-processed by masking imprinted localization grid and regions around the grid. A threshold was applied to isolate high-intensity pixels and generate a binary image. A set of morphological operations including region dilation, filling, clearing border structures, and erosion were performed to segment the different regions. A support vector machine classification model was trained to categorize segmented regions as either surgical clips or miscellaneous objects based on different region properties (area, perimeter, eccentricity, circularity, minor axis length, and intensity-derived quantities). Applicator center position was determined by calculating the centroid of detected clips. Size of the applicator was determined with the smallest circle that encompassed all clips with an isotropic 1.0 cm margin. RESULTS: The clip identification model classified 946 regions, with a sensitivity of 96.6% and a specificity of 98.2%. Applicator position was correctly predicted for 20 of 42 fractions and was within 0.5 cm of physician-selected position for 33 of 42 fractions. Applicator size was correctly predicted for 25 out of 42 fractions. CONCLUSIONS: The proposed algorithm provided a method to quantitatively determine applicator position and size for AccuBoost treatments, and may serve as a tool for independent verifications. The discrepancy between physician-selected and algorithm-predicted determinations of applicator position and size suggests that the methodology may be further improved by considering radiomic features of breast tissue in addition to clip position.
RESUMO
PURPOSE: The purpose of this study was to describe and evaluate methods for calculating a megavoltage computed tomography (MVCT)-derived MR hardware attenuation map (µ-map) and dual-energy CT (DECT) for 511 keV photons. METHODS: Phantom measurements were acquired on a whole-body hybrid PET/MRI system, using a four-channel receive-only MR radiofrequency (RF) breast coil. Two acquisitions were performed: with the phantoms positioned in the four-channel RF breast coil, and without the breast coil. PET attenuation from the breast coil was corrected using three different CT-derived hardware µ-maps: (a) Single-energy CT (SECT), (b) DECT, and (c) MVCT. Each attenuation-corrected (AC) PET volume was evaluated and compared with the acquisition performed without the breast coil. RESULTS: The breast coil attenuated PET photons by 10% overall. Threshold values were applied to the SECT µ-map to reduce the effects of metal artifacts, but overcorrection occurred in more highly attenuated regions. The DECT-derived virtual monochromatic image reduced beam-hardening artifacts, but other metal artifacts remained. Despite the remaining metal artifacts in the DECT image, it led to an improvement in the more attenuated regions. The MVCT images appear to be free from metal artifacts leading to an artifact-free µ-map and a further improvement AC-PET images. CONCLUSIONS: Our MVCT-based approach for creating µ-maps for MR RF coils greatly reduces artifacts produced by metal in a SECT approach. This eliminates the need for other artifact reduction methods, including the application of a threshold of narrow beam attenuation coefficients, or disassembling hardware to remove high-Z components before imaging with a kilovoltage source.