Application of a¹H Brain MRS Benchmark Dataset to Deep Learning for Out-of-Voxel Artifacts

Abstract

PurposeNeural networks are potentially valuable for many challenges associated with MRS data. The purpose of this manuscript is to describe the AGNOSTIC dataset, which contains 259,200 synthetic MRS examples. To demonstrate the utility, we use AGNOSTIC to train two Convolutional Neural Networks (CNNs) to address out-of-voxel (OOV) echoes. A Detection Network was trained to identify the point-wise presence of OOV echoes, providing proof of concept for real-time detection. A Prediction Network was trained to reconstruct OOV echoes, allowing subtraction during post-processing.MethodsAGNOSTIC was created using 270 basis sets that were simulated across 18 field strengths and 15 echo times. The synthetic examples were produced to resemblein vivobrain data with combinations of metabolite, macromolecule, and residual water signals, and noise.Complex OOV signals were mixed into 85% of synthetic examples to train two separate U-net CNNs for the detection and prediction of OOV signals.ResultsAGNOSTIC is available through Dryad and all Python 3 code is available through GitHub. The Detection network was shown to perform well, identifying 95% of OOV echoes. Traditional modeling of these detected OOV signals was evaluated and may prove to be an effective method during linear-combination modeling. The Prediction Network greatly reduces OOV echoes within FIDs and achieved a median log10normed-MSE of –1.79, an improvement of almost two orders of magnitude.ConclusionThe AGNOSTIC benchmark dataset for MRS is introduced and various dataset features are described. As an exemplar use of AGNOSTIC, two CNNs were developed to detect and predict OOV echoes.

Simultaneous multi-transient linear-combination modeling of MRS data improves uncertainty estimation

“…Validation can be achieved with precisely controlled synthetic data. Interest in and availability of synthetic data generation is further motivated by the need for realistic training data for MRS machine learning methods [35][36][37][38] 39 , amplitude fluctuations considered during fMRS [10][11][12] or due to motion, and phase cycling need to be carefully designed to recreate sources of transient-to-transient variability. The ultimate application to in-vivo data where the ground truth for all physical parameters is unknown should be done with caution unless their impact is properly understood.…”

“…This resurgence should be accompanied by thorough validation and benchmarking, as it has become clear that even 1D model procedures are very susceptible to small changes in algorithmic configuration 15-Validation can be achieved with precisely controlled synthetic data. Interest in and availability of synthetic data generation is further motivated by the need for realistic training data for MRS machine learning methods [34][35][36][37]…”

“…Validation can be achieved with precisely controlled synthetic data. Interest in and availability of synthetic data generation is further motivated by the need for realistic training data for MRS machine learning methods [34][35][36][37]…”