Summary
| Excerpt |
|---|
In this single institution retrospective study, we reviewed 950 consecutive patients with prostate adenocarcinoma receiving definitive radiotherapy between 2011 and 2019, and identified among them 112 patients with anatomic variations (edge cases) seen on simulation CT and/or MRI imaging. These variations included hip arthroplasty, prostate median lobe hypertrophy, so-called “droopy” seminal vesicles, presence of a urinary catheter, and others. Three commercially available solutions for pelvic anatomy autosegmentation were applied to treatment planning CT scans to generate segmented volumes of the prostateA separate cohort of 19 “normal” cases were randomly selected for inclusion. Prostate, rectum, bladder, and bilateral femoral heads . To quantify the accuracy of software-generated contours, Dice similarity coefficients were calculated for each software-generated contour with a comparator clinician-delineated reference contour.Results: On average, deep learning autosegmentation outperformed both atlas-based and model-basedwere manually segmented on all CT simulation images (where present) and were ultimately used clinically for radiation treatment planning. We leveraged this imaging dataset to assess the comparative performance of deep learning, atlas-based, and model-based autosegmentation methods across both normal and edge case cohorts . However, all methods exhibited lower mean DSC for edge cases compared to the normal cohort, recapitulating certain known algorithm limitations (e.g. in the context of hip prostheses) and identifying novel areas of poor performance.Conclusions: Anatomic variation may present challenges to commercial autosegmentation solutions, and such edge cases should be considered both prior to and throughout any clinical implementation: <INSERT DOI/LINK HERE>. |
Acknowledgements
We would like to acknowledge the individuals and institutions that have provided data for this collection:
...