
Proceedings Paper
Region-specific fully convolutional networks for segmentation of the rectal wall on post-chemoradiation T2w MRIFormat | Member Price | Non-Member Price |
---|---|---|
$17.00 | $21.00 |
Paper Abstract
Detailed localization of the rectal wall after chemoradiation on standard-of-care post-chemoradiation (CRT) MRIs could enable more targeted follow-up interventions, but it is a challenging and laborious task for radiologists. This may be because the primary tumor site (i.e. primary" wall) and the remaining adjacent" wall areas depict visually overlapping intensity characteristics as a result of chemoradiation-induced noise and treatment effects. In this study, we present initial results for developing and optimizing fully convolutional networks (FCNs) to automatically segment the rectal wall on post-CRT MRIs. Our cohort comprised 50 post-CRT, T2-weighted MRIs from rectal cancer patients with expert annotations of the entire length of the rectal wall (with separate indications for extent of primary wall as well as adjacent wall). The FCN framework was designed to provide a pixel-wise segmentation of the rectal wall while utilizing the original T2w intensity images, and was tested on 20% of the cohort that was held-out from training. Our results showed that (a) the best-performing FCN for segmenting primary wall areas utilized a training set comprising primary wall sections alone (median DSC = 0.71), while (b) optimal segmentations of adjacent wall areas were achieved by an FCN trained on both primary and adjacent wall sections (median DSC = 0.68). Notably, the primary wall FCN performed poorly when applied to adjacent wall and vice versa; perhaps indicating that fundamental physiological differences exist between these wall areas that must be accounted for within automated CN segmentation approaches. FCNs may hence have to be optimized on a region-specific basis to obtain detailed, accurate delineations of the entire rectal wall on post-CRT T2w MRI, towards more targeted excision surgery and adjuvant therapy.
Paper Details
Date Published: 8 March 2019
PDF: 7 pages
Proc. SPIE 10951, Medical Imaging 2019: Image-Guided Procedures, Robotic Interventions, and Modeling, 1095134 (8 March 2019); doi: 10.1117/12.2513055
Published in SPIE Proceedings Vol. 10951:
Medical Imaging 2019: Image-Guided Procedures, Robotic Interventions, and Modeling
Baowei Fei; Cristian A. Linte, Editor(s)
PDF: 7 pages
Proc. SPIE 10951, Medical Imaging 2019: Image-Guided Procedures, Robotic Interventions, and Modeling, 1095134 (8 March 2019); doi: 10.1117/12.2513055
Show Author Affiliations
Thomas DeSilvio, Case Western Reserve Univ. (United States)
Jacob Antunes, Case Western Reserve Univ. (United States)
Prathyush Chirra, Case Western Reserve Univ. (United States)
Kaustav Bera, Case Western Reserve Univ. (United States)
Jacob Antunes, Case Western Reserve Univ. (United States)
Prathyush Chirra, Case Western Reserve Univ. (United States)
Kaustav Bera, Case Western Reserve Univ. (United States)
Jay Gollamudi, Univ. Hospitals Cleveland Medical Ctr. (United States)
Raj Paspulati, Univ. Hospitals Cleveland Medical Ctr. (United States)
Conor P. Delaney, Cleveland Clinic (United States)
Satish E. Viswanath, Case Western Reserve Univ. (United States)
Raj Paspulati, Univ. Hospitals Cleveland Medical Ctr. (United States)
Conor P. Delaney, Cleveland Clinic (United States)
Satish E. Viswanath, Case Western Reserve Univ. (United States)
Published in SPIE Proceedings Vol. 10951:
Medical Imaging 2019: Image-Guided Procedures, Robotic Interventions, and Modeling
Baowei Fei; Cristian A. Linte, Editor(s)
© SPIE. Terms of Use
