Program Information
Automated Pulmonary Fibrosis Segmentation Using 3D Multi-Scale Convolutional Encoder-Decoder Approach in Thoracic CT for Rhesus Macaque with Radiation-Induced Lung Damage
D Yang1*, G Lasio2 , b zhang3 , K Prado4 , W D'Souza5 , D Metaxas6 , T Macvittie7 , J Zhou8 , (1) Rutgers, the State University of New Jersey, Piscataway, NJ, (2) University of Maryland, School of Medicine, Bel Air, MD, (3) University of Maryland School of Medicine, Baltimore, MD, (4) University of Maryland School of Medicine, Baltimore, MD, (5) University of Maryland School of Medicine, Baltimore, MD, (6) Rutgers, the State University of New Jersey, Piscataway, NJ, (7) University of Maryland School of Medicine, Baltimore, MD, (8) University of Maryland School of Medicine, Millburn, NJ
Presentations
SU-H2-GePD-J(A)-1 (Sunday, July 30, 2017) 3:30 PM - 4:00 PM Room: Joint Imaging-Therapy ePoster Lounge - A
Purpose: To develop automated pulmonary fibrosis (PF) segmentation using 3D multi-scale convolutional encoder-decoder (3D MSCED) approach in thoracic CT for Rhesus Macaque with radiation-induced lung damage.
Methods: 152 thoracic computed tomography (CT) scans for Rhesus Macaque with radiation-induced lung damage were collected and PF in each scan was manually segmented in which 142 scans were used as training data and 10 scans were used as testing data to assess the performance of the method.First, the compromised lung volume with acute radiation-induced PF was segmented using previous published robust atlas-based active volume model. Next, 3D MSCED segmentation method was developed which merges the higher spatial information from low-level features with the high-level object knowledge encoded in upper network layers. It includes a bottom-up feed-forward convolutional neural networks and a top-down learning mask refinement process.
Results: The quantitative results of our segmentation method achieved mean dice score of (0.77, 0.82), mean accuracy of (0.996, 0.999), and mean relative error of (0.38, 0.44) with 95% CI. The running time during inference is less than 1 second with the whole CT volume as input on a single computing GPU.
Conclusion: The qualitative and quantitative comparisons show that our proposed method can achieve better segmentation accuracy with less variance in 10 testing data. It will be useful in image analysis applications for lung lesions diagnosis and radiotherapy assessment in thoracic computed tomography.
Funding Support, Disclosures, and Conflict of Interest: Partial funding for this study was provided by Aeolus Pharmaceuticals, Inc. under Biomedical Advanced Research and Development Authority (BARDA) contract no. HHSO100201100007C and in part from the National Institute of Allergy and Infectious Diseases (NIAID), through the MCART, Radiation/Nuclear Medical Countermeasure Product Development Support Services under contract no. HHSN272201000046C.
Contact Email: