SPECT Image Reconstruction by a Learnt Neural Network

Conclusions: We have developed a deep neural network that can produce high-resolution and low-noise images for SPECT, given the SPECT projection data and attenuation map. To reduce the possibility of overfitting, multiple data sources and a total of 24,000 phantom images were involved in training. The resulting network is more reliable than those previously developed by using only tens of or up to a few hundred datasets. Validation by using clinical data demonstrated that the deep learning technology for SPECT image reconstruction is feasible.
Source: Journal of Nuclear Medicine - Category: Nuclear Medicine Authors: Tags: Image Generation (Poster Session) Source Type: research