Faculty, Staff and Student Publications
Publication Date
1-1-2024
Journal
Physics and Imaging in Radiation Oncology
DOI
10.1016/j.phro.2023.100526
PMID
38179210
PMCID
PMC10765294
PubMedCentral® Posted Date
December 2023
PubMedCentral® Full Text Version
Post-print
Abstract
One of the key contributions of this study is the reappropriation of standard DL outputs as a quality indicator to identify cases that clinicians should review further. The authors achieve this by applying an empirically derived threshold to the softmax output of their DL network, computing the mean of the thresholded score map (termed the HiS metric), and correlating it with standard geometric quality indices. When juxtaposed with a mean entropy — a commonly used measure of model output uncertainty — HiS consistently demonstrated a stronger correlation with the geometric indices, suggesting its superior ability to stratify cases needing additional review. We applaud the authors' efforts for their novel contributions and would like to note some potential caveats that could pave the way for future research directions.
Published Open-Access
yes
Recommended Citation
Wahid, Kareem A; Sahlsten, Jaakko; Jaskari, Joel; et al., "Harnessing Uncertainty in Radiotherapy Auto-Segmentation Quality Assurance" (2024). Faculty, Staff and Student Publications. 770.
https://digitalcommons.library.tmc.edu/uthgsbs_docs/770
Included in
Bioinformatics Commons, Biomedical Informatics Commons, Medical Sciences Commons, Oncology Commons, Radiation Medicine Commons, Radiology Commons