Deep neural network ensemble for on-the-fly quality control-driven segmentation of cardiac MRI T1 mapping
(2021) In Medical Image Analysis 71.- Abstract
Recent developments in artificial intelligence have generated increasing interest to deploy automated image analysis for diagnostic imaging and large-scale clinical applications. However, inaccuracy from automated methods could lead to incorrect conclusions, diagnoses or even harm to patients. Manual inspection for potential inaccuracies is labor-intensive and time-consuming, hampering progress towards fast and accurate clinical reporting in high volumes. To promote reliable fully-automated image analysis, we propose a quality control-driven (QCD) segmentation framework. It is an ensemble of neural networks that integrate image analysis and quality control. The novelty of this framework is the selection of the most optimal segmentation... (More)
Recent developments in artificial intelligence have generated increasing interest to deploy automated image analysis for diagnostic imaging and large-scale clinical applications. However, inaccuracy from automated methods could lead to incorrect conclusions, diagnoses or even harm to patients. Manual inspection for potential inaccuracies is labor-intensive and time-consuming, hampering progress towards fast and accurate clinical reporting in high volumes. To promote reliable fully-automated image analysis, we propose a quality control-driven (QCD) segmentation framework. It is an ensemble of neural networks that integrate image analysis and quality control. The novelty of this framework is the selection of the most optimal segmentation based on predicted segmentation accuracy, on-the-fly. Additionally, this framework visualizes segmentation agreement to provide traceability of the quality control process. In this work, we demonstrated the utility of the framework in cardiovascular magnetic resonance T1-mapping - a quantitative technique for myocardial tissue characterization. The framework achieved near-perfect agreement with expert image analysts in estimating myocardial T1 value (r=0.987, p<.0005; mean absolute error (MAE)=11.3ms), with accurate segmentation quality prediction (Dice coefficient prediction MAE=0.0339) and classification (accuracy=0.99), and a fast average processing time of 0.39 second/image. In summary, the QCD framework can generate high-throughput automated image analysis with speed and accuracy that is highly desirable for large-scale clinical applications.
(Less)
- author
- Hann, Evan ; Popescu, Iulia A. ; Zhang, Qiang ; Gonzales, Ricardo A. LU ; Barutçu, Ahmet ; Neubauer, Stefan ; Ferreira, Vanessa M. and Piechnik, Stefan K.
- publishing date
- 2021-07
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Cardiovascular MRI, Ensemble neural network, Image quality assessment, Segmentation
- in
- Medical Image Analysis
- volume
- 71
- article number
- 102029
- publisher
- Elsevier
- external identifiers
-
- pmid:33831594
- scopus:85103695127
- ISSN
- 1361-8415
- DOI
- 10.1016/j.media.2021.102029
- language
- English
- LU publication?
- no
- additional info
- Funding Information: A patent is filed for this work. EH and RAG acknowledge support for their DPhil studies from the Clarendon Fund, and the Radcliffe Department of Medicine, University of Oxford. EH acknowledges donation of a GPU from NVIDIA for this work. IAP, SKP, VMF and SN acknowledge support from the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre at The Oxford University Hospitals NHS Foundations Trust, University of Oxford, UK. SKP, VMF and SN acknowledge the British Heart Foundation (BHF) Centre of Research Excellence, Oxford. QZ and VMF are supported by the BHF. Publisher Copyright: © 2021
- id
- 37d7e1ff-8fcb-453b-afe9-4d4a234feb0d
- date added to LUP
- 2021-10-28 13:37:12
- date last changed
- 2024-09-09 02:25:09
@article{37d7e1ff-8fcb-453b-afe9-4d4a234feb0d, abstract = {{<p>Recent developments in artificial intelligence have generated increasing interest to deploy automated image analysis for diagnostic imaging and large-scale clinical applications. However, inaccuracy from automated methods could lead to incorrect conclusions, diagnoses or even harm to patients. Manual inspection for potential inaccuracies is labor-intensive and time-consuming, hampering progress towards fast and accurate clinical reporting in high volumes. To promote reliable fully-automated image analysis, we propose a quality control-driven (QCD) segmentation framework. It is an ensemble of neural networks that integrate image analysis and quality control. The novelty of this framework is the selection of the most optimal segmentation based on predicted segmentation accuracy, on-the-fly. Additionally, this framework visualizes segmentation agreement to provide traceability of the quality control process. In this work, we demonstrated the utility of the framework in cardiovascular magnetic resonance T1-mapping - a quantitative technique for myocardial tissue characterization. The framework achieved near-perfect agreement with expert image analysts in estimating myocardial T1 value (r=0.987, p<.0005; mean absolute error (MAE)=11.3ms), with accurate segmentation quality prediction (Dice coefficient prediction MAE=0.0339) and classification (accuracy=0.99), and a fast average processing time of 0.39 second/image. In summary, the QCD framework can generate high-throughput automated image analysis with speed and accuracy that is highly desirable for large-scale clinical applications.</p>}}, author = {{Hann, Evan and Popescu, Iulia A. and Zhang, Qiang and Gonzales, Ricardo A. and Barutçu, Ahmet and Neubauer, Stefan and Ferreira, Vanessa M. and Piechnik, Stefan K.}}, issn = {{1361-8415}}, keywords = {{Cardiovascular MRI; Ensemble neural network; Image quality assessment; Segmentation}}, language = {{eng}}, publisher = {{Elsevier}}, series = {{Medical Image Analysis}}, title = {{Deep neural network ensemble for on-the-fly quality control-driven segmentation of cardiac MRI T1 mapping}}, url = {{http://dx.doi.org/10.1016/j.media.2021.102029}}, doi = {{10.1016/j.media.2021.102029}}, volume = {{71}}, year = {{2021}}, }