Publication
Application of a1H Brain MRS Benchmark Dataset to Deep Learning for Out-of-Voxel Artifacts
Publisher:
Cold Spring Harbor Laboratory
Date:
08-05-2023
DOI:
10.1101/2023.05.08.539813
Abstract: Neural networks are potentially valuable for many of the challenges associated with MRS data. The purpose of this manuscript is to describe the AGNOSTIC dataset, which contains 259,200 synthetic 1 H MRS ex les for training and testing neural networks. AGNOSTIC was intended to be acquisition-agnostic by using 270 basis sets that were simulated across 18 field strengths, 15 echo times, and a range of dwell times. The synthetic ex les were produced to resemble healthy and clinical in vivo brain data with combinations of metabolite, macromolecule, residual water signals, and noise. To demonstrate the utility, we apply 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. Complex OOV signals were mixed into 85% of synthetic ex les to train two separate CNNs for the detection and prediction of OOV signals. AGNOSTIC 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 log 10 normed-MSE of –1.79, an improvement of almost two orders of magnitude. AGNOSTIC is an open-source benchmark dataset for training and testing brain a wide range of 1 H MRS deep learning models. AGNOSTIC is intended to be step towards developing acquisition-agnostic deep learning models by including synthetic ex les from 18 field strengths, 15 echo times, multiple dwell times, healthy and clinical brain concentrations, and a large range of spectral quality (i.e., SNR and linewidth). Two deep learning models were trained using AGNOSTIC to detect and predict out-of-voxel echoes where the models generalized to in vivo data collected with unseen acquisition protocols.