Cardio-pulmonary substructure segmentation of CT images using convolutional neural networks Conference Paper
| Authors: | Haq, R.; Hotca, A.; Apte, A.; Rimner, A.; Deasy, J. O.; Thor, M. |
| Title: | Cardio-pulmonary substructure segmentation of CT images using convolutional neural networks |
| Conference Title: | 1st International Workshop - Artificial Intelligence in Radiation Therapy |
| Abstract: | Radiotherapy doses to some cardio-pulmonary substructures may be critical factors in the observed early mortality following radiotherapy for nonsmall cell lung cancer patients. Our goal is to provide an open-source tool to automatically segment cardio-vascular substructures for consistent outcomes analyses, and subsequently for radiation treatment planning of thoracic patients. Here, we built and validated a multi-label Deep Learning Segmentation (DLS) framework for accurate auto-segmentation of cardio-pulmonary substructures. The DLS framework utilized a deep neural network architecture to segment 12 cardio-pulmonary substructures from Computed Tomography (CT) scans of 217 patients previously treated with thoracic RT. The model was robust against variability in image quality characteristics, including the presence/absence of contrast. A hold-out dataset of additional 24 CT scans was used for quantitative evaluation of the final model against expert contours using Dice Similarity Coefficients (DSC) and 95th Percentile of Hausdorff Distance (HD95). DLS contours of an additional 10 CT scans were reviewed by a radiation oncologist to determine the number of slices in need of adjustment for each of the non-overlapping substructures. The DLS model reduced segmentation time per patient from about one hour of manual segmentation to 10 s. Quantitatively, the highest accuracy was observed for the Heart (median DSC = (0.96(0.91 - 0.93 )) and HD95 = (4.3 mm(3.8 mm − 5.5 mm)). The median DSC for the remaining structures was 0.80 - 0.92. The expert judged that, on average, 85% of the contours were equivalent to state-of-the-art manual contouring and did not require any modifications. © Springer Nature Switzerland AG 2019. |
| Keywords: | radiotherapy; computerized tomography; medical imaging; radiation therapy; radiation treatment planning; non small cell lung cancer; image segmentation; semantics; neural networks; computed tomography scan; convolution; convolutional neural network; semantic segmentation; deep neural networks; quantitative evaluation; network architecture; convolutional neural networks; similarity coefficients; cardio-pulmonary; dynamic light scattering |
| Journal Title | Lecture Notes in Computer Science |
| Volume: | 11850 |
| Conference Dates: | 2019 Oct 17 |
| Conference Location: | Shenzhen, China |
| ISBN: | 0302-9743 |
| Publisher: | Springer |
| Date Published: | 2019-01-01 |
| Start Page: | 162 |
| End Page: | 169 |
| Language: | English |
| DOI: | 10.1007/978-3-030-32486-5_20 |
| PROVIDER: | scopus |
| DOI/URL: | |
| Notes: | Part of the MICCAI 2019 Conference -- Source: Scopus |
Altmetric
Citation Impact
BMJ Impact Analytics
Related MSK Work