Elucidating the effect of tumor and background region-of-interest selection on the performance metrics used to assess fluorescence imaging.

Abstract

Significance: The development of fluorescent contrast agents for fluorescence-guided surgery is rapidly growing with many agents being designed for tumor visualization. Although efforts have been made to standardize the sensitivity of imaging system detection methods for these contrast agents, guidelines to evaluate tumor contrast agent performance, especially the selection of tumor and background regions of interest (ROIs), differ widely across studies. We examine how systematically changing tumor and background ROIs affects common metrics of contrast agent performance.

Aim: We aim to elucidate the influence of changing tumor and background brain regions of interest on fluorescent contrast agent performance.

Approach: Mice with orthotopic brain tumors were administered a non-targeted fluorescent contrast agent 40 min prior to sacrifice and then imaging of the specimen using whole-body fluorescence cryotomography. The reconstructed 3D fluorescence volumes were then used to compute contrast and diagnostic performance metrics [tumor-to-background brain ratio (TBR), contrast-to-noise (CNR), and area under the receiver operating characteristic curve (AUC)] while systematically varying tumor and normal brain ROIs.

Results: ROI selection had a significant impact on the reported values of metrics used to evaluate fluorescence imaging strategies. The use of contralateral background ROIs, commonly used in the field, produced elevated and favorable performance metric values. These metrics decreased as background ROIs approached regions adjacent to the tumor boundary. TBR changed by a factor of 5, CNR by a factor of 7, and AUC by over 10%, largely depending on the proximity of the background region to the tumor.

Conclusions: Background ROI selection has a significant impact on the performance metrics commonly used in the field. Future studies should carefully select ROIs relevant to the application and include clear descriptions of these regions.