2.5D Deep Learning for CT Image Reconstruction using a Multi-GPU implementation

TL;DR

This paper introduces a fast deep learning-based CT reconstruction method called DL-MBIR, which approximates traditional MBIR images using a multi-GPU residual neural network, offering similar quality with reduced computational cost.

Contribution

The paper presents a novel multi-GPU deep residual neural network approach for CT image reconstruction that approximates MBIR with improved efficiency and introduces 2.5D and 3D variations.

Findings

2.5D method achieves similar quality to 3D with less computation.

DL-MBIR trained on GPUs produces high-quality reconstructions.

Multi-GPU implementation accelerates the reconstruction process.

Abstract

While Model Based Iterative Reconstruction (MBIR) of CT scans has been shown to have better image quality than Filtered Back Projection (FBP), its use has been limited by its high computational cost. More recently, deep convolutional neural networks (CNN) have shown great promise in both denoising and reconstruction applications. In this research, we propose a fast reconstruction algorithm, which we call Deep Learning MBIR (DL-MBIR), for approximating MBIR using a deep residual neural network. The DL-MBIR method is trained to produce reconstructions that approximate true MBIR images using a 16 layer residual convolutional neural network implemented on multiple GPUs using Google Tensorflow. In addition, we propose 2D, 2.5D and 3D variations on the DL-MBIR method and show that the 2.5D method achieves similar quality to the fully 3D method, but with reduced computational cost.