Files
Abstract
In brain tumor diagnosis and treatment, manual segmentation is the “gold standard” approach, through which an expert annotates tumor regions manually. This process is becoming increasingly infeasible as patient data volumes exceed quantities which can be reliably segmented in reasonable periods of time. Additionally, brain tumors exhibit wide variation in type, extent and location, further complicating task. This renders manual segmentation a time-consuming and labor-intensive undertaking, shown to yield inconsistent results. Automated models, implemented with deep learning architectures have demonstrated a faster, more consistent segmentation approach. While the benefits of automated models have been established, they have historically failed to be integrated into clinical practice. Research indicates that bridging the clinical gap requires establishing “trust” and “transparency” between end users, clinicians, and these automated tools. This paper proposes color space visualization of automated segmentation results, designed to improve standard segmentation practices through collaborative effort between automation and expert knowledge.