A parallel way of data decomposition approach for ANN based image reconstruction in e-MRI on a multi-core computer system

Informatics in Medicine Unlocked

关键词: Electron magnetic resonance tomography;    Image reconstruction;    Filtered back projection;    Artificial neural networks;    Multi-core processors;   

DOI  :  10.1016/j.imu.2017.10.003

来源: DOAJ

【 摘 要 】

This paper presents the performance of sequential data decomposition and parallel data decomposition strategies applied on a Back-Propagation Artificial Neural Network (BP-ANN) algorithm. The application system is developed for reconstruction of two-dimensional spatial images from continuous wave electron magnetic resonance imaging (CW-EMRI) tomography data, on a multi-core computer. The BP-ANN learns the relationship between the ‘ideal’ images that are reconstructed using filtered back projection (FBP) technique and the corresponding projection data from various temporal data of in vivo objects. In an earlier work, it has been reported that as the exemplar sizes are too large, the training time is too long and image PSNR (Peak Signal to Noise Ratio) values are too low. Hence, in the present work, we propose that the exemplar datasets are decomposed into subsets. Using these subsets, artificial sub neural nets (subnets) are constructed and training is carried out on a multi-core system. Consequently, the sequential approach of the proposed method yields better PSNR images. However, it consumes more training time. But when the parallel approach is applied the computational training time becomes reduced. The parallel approach of BP-ANN is able to simplify reconstruction tasks and is seen improving both in accuracy and efficiency. The performance results are tabulated for different exemplar subset sizes, different subnet sizes and the number of multi-core processors. The parallel approach is further explored for image reconstruction from ‘noisy’ and ‘limited-angle’ datasets.

【 授权许可】

Unknown