Blended-Transfer Learning for Compressed-Sensing Cardiac CINE MRI
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Purpose: To overcome the difficulty in building a large data set with a high-quality in medical imaging, a concept of 'blended-transfer learning' (BTL) using a combination of both source data and target data is proposed for the target task. Materials and Methods: Source and target tasks were defined as training of the source and target networks to reconstruct cardiac CINE images from undersampled data, respectively. In transfer learning (TL), the entire neural network (NN) or some parts of the NN after conducting a source task using an open data set was adopted in the target network as the initial network to improve the learning speed and the performance of the target task. Using BTL, an NN effectively learned the target data while preserving knowledge from the source data to the maximum extent possible. The ratio of the source data to the target data was reduced stepwise from 1 in the initial stage to 0 in the final stage. Results: NN that performed BTL showed an improved performance compared to those that performed TL or standalone learning (SL). Generalization of NN was also better achieved. The learning curve was evaluated using normalized mean square error (NMSE) of reconstructed images for both target data and source data. BTL reduced the learning time by 1.25 to 100 times and provided better image quality. Its NMSE was 3% to 8% lower than with SL. Conclusion: The NN that performed the proposed BTL showed the best performance in terms of learning speed and learning curve. It also showed the highest reconstructed-image quality with the lowest NMSE for the test data set. Thus, BTL is an effective way of learning for NNs in the medical-imaging domain where both quality and quantity of data are always limited.
Abstract
Purpose: To overcome the difficulty in building a large data set with a high-quality in medical imaging, a concept of 'blended-transfer learning' (BTL) using a combination of both source data and target data is proposed for the target task. Materials and Methods: Source and target tasks were defined as training of the source and target networks to reconstruct cardiac CINE images from undersampled data, respectively. In transfer learning (TL), the entire neural network (NN) or some parts of the NN after conducting a source task using an open data set was adopted in the target network as the initial network to improve the learning speed and the performance of the target task. Using BTL, an NN effectively learned the target data while preserving knowledge from the source data to the maximum extent possible. The ratio of the source data to the target data was reduced stepwise from 1 in the initial stage to 0 in the final stage. Results: NN that performed BTL showed an improved performance compared to those that performed TL or standalone learning (SL). Generalization of NN was also better achieved. The learning curve was evaluated using normalized mean square error (NMSE) of reconstructed images for both target data and source data. BTL reduced the learning time by 1.25 to 100 times and provided better image quality. Its NMSE was 3% to 8% lower than with SL. Conclusion: The NN that performed the proposed BTL showed the best performance in terms of learning speed and learning curve. It also showed the highest reconstructed-image quality with the lowest NMSE for the test data set. Thus, BTL is an effective way of learning for NNs in the medical-imaging domain where both quality and quantity of data are always limited.
참고문헌 (29)
- Lee VS, "Cardiovascular MRI: physical principles to practical protocols" (2006), Lippincott Williams & Wilkins
- Manning WJ, "Cardiovascular magnetic resonance" (2010), Saunders
- 박진호, "Fast Cardiac CINE MRI by Iterative Truncation of Small Transformed Coefficients", Investigative Magnetic Resonance Imaging, 19(1), pp.19~30 (2015), 대한자기공명의과학회[DOI]
- Yoon JH, "Biases in the assessment of left bentricular dunction by compressed sensing cardiovascular CINE MRI", Investig Magn Reson Imaging, 23, pp.114~124 (2019)
- Lustig M, "Sparse MRI : the application of compressed sensing for rapid MR imaging", Magn Reson Med, 58, pp.1182~1195 (2007)
- 이두희, "Deep Learning in MR Image Processing", Investigative Magnetic Resonance Imaging, 23(2), pp.81~99 (2019), 대한자기공명의과학회
- Wang S, "Accelerating magnetic resonance imaging via deep learning", Proc IEEE Int Symp Biomed Imaging, 2016, pp.514~517 (2016)
- Yang Y, "Deep ADMM-Net for compressive sensing MRI", Adv Neural Inf Process Syst (NIPS), pp.10~18 (2016)
- Yang G, "DAGAN : deep de-aliasing generative adversarial networks for fast compressed sensing MRI reconstruction", IEEE Trans Med Imaging, 37, pp.1310~1321 (2018)
- Goodfellow I, "Generative adversarial networks", Adv Neural Inf Process Syst (NIPS), pp.2672~2680 (2014)
- Radford A, "Unsupervised representation learning with deep convolutional generative adversarial networks", arXiv preprint arXiv:1511.06434, ~ (2015)
- Yu S, "Deep de-aliasing for fast compressive sensing MRI", arXiv preprint arXiv:1705.07137, ~ (2017)
- Zhu J, "How can we make GAN performs better in single medical image super-resolution? A lesion focused multi-scale approach", Proc IEEE Int Symp Biomed Imaging (ISBI), pp.1669~1673 (2019)
- Wang C, "TPSDicyc:improved deformation invariant cross-domain medical image synthesis", Int Workshop on Mach Learn Med Image Reconstr (MLMIR), pp.245~254 (2019)
- Schlemper J, "Stochastic deep compressive sensing for the reconstruction of diffusion tensor cardiac MRI", Int Conf Med Image Comput Comput Assist Interv (MICCAI), pp.295~303 (2018)
- Zhu J, "A ROI focused multi-scale super-resolution method for the diffusion tensor cardiac magnetic resonance", Proc Int Soc Magn Reson Med (ISMRM), pp.1~ (2019)
- Hyun CM, "Deep learning for undersampled MRI reconstruction", Phys Med Biol, 63, pp.135007~ (2018)
- Ronneberger O, "U-net: convolutional networks for biomedical image segmentation", Int Conf Med Image Comput Comput Assist Interv (MICCAI), pp.234~241 (2015)
- Kofler A, "Spatio-temporal deep learning-based undersampling artefact reduction for 2D radial cine MRI with limited training data", IEEE Trans Med Imaging, 39, pp.703~717 (2020)
- Pan SJ, "A survey on transfer learning", IEEE Trans Knowl Data Eng, 22, pp.1345~1359 (2010)
- Park SJ, "Transfer learning for compressedsensing cardiac CINE MRI", Proc Int Soc Magn Reason Med (ISMRM) (2020)
- Oquab M, "Learning and transferring mid-level image representations using convolutional neural networks", Proc IEEE Comput Vis Pattern Revognit (CVPR), pp.1717~1724 (2014)
- Ciresan D C, "Transfer learning for Latin and Chinese characters with deep neural networks", Proc Int Jt Conf Neural Netw (IJCNN), pp.1~6 (2012)
- Meng D, "Liver fibrosis classification based on transfer learning and FCNet for ultrasound images", IEEE Access, 5, pp.5804~5810 (2017)
- Chen A, "Transfer learning for the fully automatic segmentation of left ventricle myocardium in porcine cardiac cine MR images", Int Workshop on Stat Atlases Comput Models Heart (STACOM), pp.21~31 (2017)
- Dar SUH, "A transfer-learning approach for accelerated MRI using deep neural networks", Magn Reson Med, 84, pp.663~685 (2020)
- Andreopoulos A, "Efficient and generalizable statistical models of shape and appearance for analysis of cardiac MRI", Med Image Anal, 12, pp.335~357 (2008)
- Glorot X, "Understanding the difficulty of training deep feedforward neural network", Proc Int Conf Artif Intell Stat (AISTATS), pp.249~256 (2010)
- Seitzer M, "Adversarial and perceptual refinement for compressed sensing MRI reconstruction", Int Conf Med Image Comput Comput Assist Interv (MICCAI), pp.232~240 (2018)