Background: Despite chemotherapeutic advances, surgery remains a putative treatment modality for breast cancer. The type of surgery chosen, and its ultimate clinical and cosmetic consequences, depends on a surgeon’s ability to accurately assess a tumor’s size, distribution, and position in the breast relative to anatomical landmarks. Moreover, in cases in which the tumor-to-breast volume ratio is substantial, neoadjuvant chemotherapy (NAC) offers the potential to shrink a tumor such that enhanced cosmesis or breast-conserving surgery becomes possible. Therefore, doctors must not only build an accurate picture of the current tumor shape, size and location, but anticipate how they might change over time and during treatment.
SimBioSys’ TumorSightTM offers doctors the ability to simulate the way a patient’s tumor might respond to various drug therapies, volumetrically and morphologically. These predictions are critical for surgical planning, in which a patient’s eligibility for breast-conserving surgery depends on the spatial distribution of the tumor. Primarily, this means consideration of (1) tumor multicentricity, (2) the extirpative tumor-to-breast volume ratio, and (3) the distance from the tumor to nipple. TumorSight provides physicians with estimates of each of these three key metrics, in 3D, over time, and as a function of regimen, to facilitate accurate evaluation, visualization and forecasting of NAC response with respect to surgical planning.
Methods: TumorSight was used to create 3D models across multiple patient cohorts using DCE-MRI. These included segmentations for skin, chest wall, nipple, adipose tissue, glandular tissue, vasculature, and tumor. Volumetric and morphological changes in the tumor and adjacent tissues were simulated throughout a chemotherapy treatment regimen, as described previously. The resulting tumor distributions were analyzed and compared to distributions in post-treatment MRI segmentations. Metrics of interest in this study were the distance of closest approach between the tumor and nipple, tumor volume-to-breast volume ratio, tumor convex hull-to-breast volume ratio, and whether the tumor was monocentric.
Results: In a set of N=292 patients, both the predicted tumor volume and tumor volume from a post-treatment MRI segmentation were computed. We took the difference in these volumes, normalized by the total breast volume (also computed from an MRI segmentation), to assess how well TumorSight predicts the final tumor volume-to-breast volume ratio after treatment. We found that the median difference in this metric was 3.26e-07, and 5th and 95th percentiles were 0.0035 and 0.0023, respectively. Similar performance was seen when using the volume of a convex boundary enclosing the tumor and a 1 cm margin (to simulate operative conditions). In a set of N=164 patients, the accuracy of our prediction of whether the tumor will be monocentric after treatment was assessed. We found that our simulations were 78.6% accurate in predicting monocentricity. Finally, in a set of N=134 patients, TumorSight predicted a distribution of changes in the closest distance between the nipple and tumor, with a median of 1.06 cm and standard deviation of 1.64 cm. In all studies, there was no difference in median values seen between cancer subtypes.
Conclusion: The ability to accurately measure, visualize, and predict post-NAC metrics such as tumor-to-nipple distance, monocentricity, and tumor-to-breast volume ratio is critical to surgical planning in breast cancer. To assist in this planning, TumorSight provides a surgeon and their patient with a realistic three-dimensional representation of tumor volume, shape, and location in the breast, and reliably predicts how these metrics change during treatment.
