Functional tumor diameter measurement with molecular breast imaging: development and clinical application

Abstract

Purpose: Molecular breast imaging (MBI) is used clinically to visualize the uptake of 99mTc-sestamibi in breast cancers. Here, we use Monte Carlo simulations to develop a methodology to estimate tumor diameter in focal lesions and explore a semi-automatic implementation for clinical data. Methods: A validated Monte Carlo simulation of the GE Discovery NM 750b was used to simulate >75,000 unique spherical/ellipsoidal tumor, normal breast, and image acquisition conditions. Subsets of this data were used to 1) characterize the dependence of the full-width at half-maximum (FWHM) of a tumor profile on tumor, normal breast, and acquisition conditions, 2) develop a methodology to estimate tumor diameters, and 3) quantify the diameter accuracy in a broad range of clinical conditions. Finally, the methodology was implemented in patient images and compared to diameter estimates from physician contours on MBI, mammography, and ultrasound imaging. Results: Tumor profile FWHM was determined be linearly dependent on tumor diameter but independent of other factors such as tumor shape, uptake, and distance from the detector. A linear regression was used to calculate tumor diameter from the FWHM estimated from a background-corrected profile across a tumor extracted from a median-filtered single-detector MBI image, i.e., diameter = 1.2 mm + 1.2 × FWHM, for FWHM ≥ 13 mm. Across a variety of simulated clinical conditions, the mean error of the methodology was 0.2 mm (accuracy), with >50% of cases estimated within 1-pixel width of the truth (precision). In patient images, the semi-automatic methodology provided the longest diameter in 94% (60/64) of cases. The estimated true diameters, for oval lesions with homogeneous uptake, differed by ± 5 mm from physician measurements. Conclusion: This work demonstrates the feasibility of accurately quantifying tumor diameter in clinical MBI, and to our knowledge, is the first to explore its implementation and application in patient data.