Post-operative stereotactic radiotherapy to the resection cavity is standard of care for patients that underwent neurosurgery for brain cancer. Current image-guided radiotherapy requires acquisition of MR scans that allow manual delineation of tumor target and organs at risk (OARs), and thereafter, acquisition of a CT scan for radiotherapy dose planning and patient position verification at the linear accelerator during treatment. In this project we will develop image analysis algorithms to allow fast delineation of tumor target and OARs. Moreover, we will develop algorithms for the synthesis of CT scan to allow MR-only MRI workflow for radiotherapy interventions shortly after neurosurgical procedures.
Researchers
Inproceedings
|
1. | L.D. van Harten, J.M. Wolterink, J.J.C. Verhoeff, I. Išgum Automatic online quality control of synthetic CTs Inproceedings In: SPIE Medical Imaging, pp. 113131M, 2020. Abstract | Links | BibTeX @inproceedings{vanharten2020automatic,
title = {Automatic online quality control of synthetic CTs},
author = {L.D. van Harten, J.M. Wolterink, J.J.C. Verhoeff, I. Išgum},
url = {https://spie.org/MI/conferencedetails/medical-image-processing#2549286},
doi = {10.1117/12.2549286},
year = {2020},
date = {2020-03-10},
booktitle = {SPIE Medical Imaging},
volume = {11313},
pages = {113131M},
abstract = {Accurate MR-to-CT synthesis is a requirement for MR-only workflows in radiotherapy (RT). While deep learning-based approaches have shown impressive results in this field, assessing the quality of the results as part of the clinical workflow is problematic. We use an ensemble of synthetic CT (sCT) generators and assess their disagreement as a measure of uncertainty of the results. We show that this uncertainty can be used for online quality control to detect input images that are outside the expected distribution of MR images, and to identify sCT images that potentially contain errors.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Accurate MR-to-CT synthesis is a requirement for MR-only workflows in radiotherapy (RT). While deep learning-based approaches have shown impressive results in this field, assessing the quality of the results as part of the clinical workflow is problematic. We use an ensemble of synthetic CT (sCT) generators and assess their disagreement as a measure of uncertainty of the results. We show that this uncertainty can be used for online quality control to detect input images that are outside the expected distribution of MR images, and to identify sCT images that potentially contain errors. |
2. | L.D. van Harten, J.M. Wolterink, J.J.C. Verhoeff, I. Išgum Exploiting clinically available delineations for CNN-based segmentation in radiotherapy treatment planning Inproceedings In: SPIE Medical Imaging, pp. 113131F, 2020. Abstract | Links | BibTeX @inproceedings{vanharten2020exploiting,
title = {Exploiting clinically available delineations for CNN-based segmentation in radiotherapy treatment planning},
author = {L.D. van Harten, J.M. Wolterink, J.J.C. Verhoeff, I. Išgum},
url = {https://spie.org/MI/conferencedetails/medical-image-processing#2549653},
doi = {10.1117/12.2549653},
year = {2020},
date = {2020-03-10},
booktitle = {SPIE Medical Imaging},
volume = {11313},
pages = {113131F},
abstract = {This work investigates whether clinically obtained segmentations could provide a possible source of data to train a CNN for segmentation of organs-at-risk (OARs) in radiotherapy treatment planning. In such data, delineations of structures deemed irrelevant to the target clinical use may be lacking. To overcome this issue, we use multi-label instead of multi-class segmentation. We empirically assess how many clinical delineations would be sufficient to train a CNN for the segmentation of OARs and find that increasing the training set size beyond a limited number of images leads to sharply diminishing returns. Moreover, we find that by using multi-label segmentation, missing structures in the reference standard do not have a negative effect on overall segmentation accuracy.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This work investigates whether clinically obtained segmentations could provide a possible source of data to train a CNN for segmentation of organs-at-risk (OARs) in radiotherapy treatment planning. In such data, delineations of structures deemed irrelevant to the target clinical use may be lacking. To overcome this issue, we use multi-label instead of multi-class segmentation. We empirically assess how many clinical delineations would be sufficient to train a CNN for the segmentation of OARs and find that increasing the training set size beyond a limited number of images leads to sharply diminishing returns. Moreover, we find that by using multi-label segmentation, missing structures in the reference standard do not have a negative effect on overall segmentation accuracy. |