q-Space Deep Learning for Twelve-Fold Shorter and Model-Free Diffusion MRI Scans
Diffusion MRI uses a multi-step data processing pipeline. With certain steps being prone to instabilities, the pipeline relies on considerable amounts of partly redundant input data, which requires long acquisition time. This leads to high scan costs and
- PDF / 280,504 Bytes
- 8 Pages / 439.363 x 666.131 pts Page_size
- 66 Downloads / 167 Views
chnische Universit¨ at M¨ unchen, Garching, Germany 2 University of Freiburg, Freiburg, Germany Max Planck Institute of Psychiatry, Munich, Germany 4 GE Global Research, Garching, Germany [email protected]
Abstract. Diffusion MRI uses a multi-step data processing pipeline. With certain steps being prone to instabilities, the pipeline relies on considerable amounts of partly redundant input data, which requires long acquisition time. This leads to high scan costs and makes advanced diffusion models such as diffusion kurtosis imaging (DKI) and neurite orientation dispersion and density imaging (NODDI) inapplicable for children and adults who are uncooperative, uncomfortable or unwell. We demonstrate how deep learning, a group of algorithms in the field of artificial neural networks, can be applied to reduce diffusion MRI data processing to a single optimized step. This method allows obtaining scalar measures from advanced models at twelve-fold reduced scan time and detecting abnormalities without using diffusion models.
1
Introduction
Advanced diffusion MRI models such as DKI [1] and NODDI [2] are preferable over traditional diffusion MRI models because they provide more accurate characterization of tissue microstructure. However, they require long acquisition times. This can be problematic in clinical applications due to high scan costs or if the patient is uncooperative, uncomfortable or unwell. In diffusion MRI, a number of diffusion-weighted images (DWIs) for different diffusion weightings and directions (constituting the so-called three-dimensional q-space) are acquired. The task in quantitative diffusion MRI is to find a mapping from a limited number of noisy signal samples to rotationally invariant scalar measures that quantify microstructural tissue properties. This inverse problem is solved in each image voxel. The classical approach consists of fitting [3] a diffusion model and calculating rotationally invariant measures from the fitted model parameters. Another approach to calculate scalar measures is approximation, particularly machine learning. Simulations of simplified tissue models with extensive sets of diffusion weightings [4, 5] indicate that standard model fitting procedures can be replaced by approximation methods. On the basis of c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part I, LNCS 9349, pp. 37–44, 2015. DOI: 10.1007/978-3-319-24553-9_5
38
V. Golkov et al.
these observations, we apply deep learning [6–9] for accurate approximation and present a deep learning framework for different inputs (full and subsampled sets of regular DWIs, non-diffusion contrasts) and outputs (denoising, missing DWI reconstruction, scalar measure estimation, tissue segmentation). We term this framework q-space deep learning (q-DL). Scalar measure estimation from twelve-fold shortened acquisition is demonstrated on two advanced models: DKI and NODDI. By shortening the acquisition duration of advanced models by an order of magnitude, we strongly improve their potential for clinical use. Re
Data Loading...
