Octree Representation Improves Data Fidelity of Cardiac CT Images and Convolutional Neural Network Semantic Segmentation of Left Atrial and Ventricular Chambers

Kunal Gupta; Nitesh Sekhar; Davis M Vigneault; Anderson R Scott; Brendan Colvert; Amanda Craine; Adhithi Raghavan; Francisco J Contijoch

doi:10.1148/ryai.2021210036

Octree Representation Improves Data Fidelity of Cardiac CT Images and Convolutional Neural Network Semantic Segmentation of Left Atrial and Ventricular Chambers

Radiol Artif Intell. 2021 Sep 29;3(6):e210036. doi: 10.1148/ryai.2021210036. eCollection 2021 Nov.

Authors

Kunal Gupta¹, Nitesh Sekhar¹, Davis M Vigneault¹, Anderson R Scott¹, Brendan Colvert¹, Amanda Craine¹, Adhithi Raghavan¹, Francisco J Contijoch¹

Affiliation

¹ Departments of Computer Science Engineering (K.G., N.S.), Bioengineering (D.M.V., A.R.S., B.C., A.C., A.R., F.J.C.), and Radiology (F.J.C.), University of California, San Diego, 9500 Gilman Dr, MC 0412, La Jolla, CA 92093; and Department of Internal Medicine, Scripps Mercy Hospital, San Diego, Calif (D.M.V.).

Abstract

Purpose: To assess whether octree representation and octree-based convolutional neural networks (CNNs) improve segmentation accuracy of three-dimensional images.

Materials and methods: Cardiac CT angiographic examinations from 100 patients (mean age, 67 years ± 17 [standard deviation]; 60 men) performed between June 2012 and June 2018 with semantic segmentations of the left ventricular (LV) and left atrial (LA) blood pools at the end-diastolic and end-systolic cardiac phases were retrospectively evaluated. Image quality (root mean square error [RMSE]) and segmentation fidelity (global Dice and border Dice coefficients) metrics of the octree representation were compared with spatial downsampling for a range of memory footprints. Fivefold cross-validation was used to train an octree-based CNN and CNNs with spatial downsampling at four levels of image compression or spatial downsampling. The semantic segmentation performance of octree-based CNN (OctNet) was compared with the performance of U-Nets with spatial downsampling.

Results: Octrees provided high image and segmentation fidelity (median RMSE, 1.34 HU; LV Dice coefficient, 0.970; LV border Dice coefficient, 0.843) with a reduced memory footprint (87.5% reduction). Spatial downsampling to the same memory footprint had lower data fidelity (median RMSE, 12.96 HU; LV Dice coefficient, 0.852; LV border Dice coefficient, 0.310). OctNet segmentation improved the border segmentation Dice coefficient (LV, 0.612; LA, 0.636) compared with the highest performance among U-Nets with spatial downsampling (Dice coefficients: LV, 0.579; LA, 0.592).

Conclusion: Octree-based representations can reduce the memory footprint and improve segmentation border accuracy.Keywords CT, Cardiac, Segmentation, Supervised Learning, Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning Algorithms© RSNA, 2021.

Keywords: CT; Cardiac; Convolutional Neural Network (CNN); Deep Learning Algorithms; Machine Learning Algorithms; Segmentation; Supervised Learning.

2021 by the Radiological Society of North America, Inc.

Abstract

Grants and funding