Multiparametric MRI dataset for susceptibility-based radiomic feature extraction and analysis

Abstract

Multiple sclerosis (MS) is a progressive demyelinating disease impacting the central nervous system. Conventional Magnetic Resonance Imaging (MRI) techniques (e.g., T₂w images) help diagnose MS, although they sometimes reveal non-specific lesions. Quantitative MRI techniques are capable of quantifying imaging biomarkers in vivo, offering the potential to identify specific signs related to pre-clinical inflammation. Among those techniques, Quantitative Susceptibility Mapping (QSM) is particularly useful for studying processes that influence the magnetic properties of brain tissue, such as alterations in myelin concentration. Because of its intrinsic quantitative nature, it is particularly well-suited to be analyzed through radiomics, including techniques that extract a high number of complex and multi-dimensional features from radiological images. The dataset presented in this work provides information about normal-appearing white matter (NAWM) in a cohort of MS patients and healthy controls. It includes QSM-based radiomic features from NAWM and its tracts, and MR sequences necessary to implement the pipeline: T₁w, T₂w, QSM, DWI. The workflow is outlined in this article, along with an application showing feature reliability assessment.

Fig. 1

Scheme of the acquisition and processing pipeline. The brain MRI protocol provided: T₁w, T₂w, DWI and T₂*w images (in this figure: patient with MS, F/38 years old): 1) T₁w images were used to obtain white matter segmentation using 5ttgen from MRtrix3 2) LPA algorithm from FSL was applied to T₂w images to obtain MS lesion mask. Merging WM segmentation with the MS lesion mask, the normal-appearing-white-matter segmentation mask was obtained. We run LST only on patients with MS; for controls, NAWM corresponds to WM from MRtrix3. 3) DWI images were pre-processed as explained in the specific section and diffusion tractography imaging automatic pipeline was applied, obtaining six white matter tracts (arcuate fasciculus, cortico-spinal tract, frontal aslant tract, inferior frontal-occipital fasciculus, optic radiation, uncinate fasciculus); VOIs from tractography reconstruction were merged with MS lesion mask to exclude damaged tissue. 4) T₂*w images were processed to obtain QSM reconstructions, as explained in the specific section. All the images/masks were registered in the T₁w space. Radiomic features were extracted from QSM images in 14 volumes (six white matter tracts and total normal appearing white matter, left and right hemisphere).

Fig. 2

Example of one exam (patient with MS, F/29 years old) excluded because of the quality of QSM image, showing movement artifacts and not considered suitable for the analysis: in the first and second row, magnitude and phase raw data from the first echo time; in the third row, the QSM reconstruction.

Fig. 3

Seed, inclusion and exclusion regions for AF, CST and FAT. Regions were defined on the MNI152 standard brain and used for streamlining generation and selection to reconstruct WM tracts. Region descriptions are in Table 2. In the last row, the appearance of each tract reconstruction is shown in the MNI152 space, obtained by averaging tracts of 30 healthy controls (AF = Arcuate Fasciculus, CST = Cortico-Spinal Tract, FAT = Frontal Aslant Tract, OB WAY = Obligatory Waypoint(s), Excl = Exclusion mask(s), MNI152 = Montreal Neurological Institute’s 152, WM = White Matter).

Fig. 4

Seed, inclusion and exclusion regions for IFOF, OR and UF. Regions were defined on the MNI152 standard brain and used for streamlining generation and selection to reconstruct WM tracts. Region descriptions are in Table 2. In the last row, the appearance of each tract reconstruction is shown in the MNI152 space, obtained by averaging tracts of 30 healthy controls (IFOF = Inferior Frontal-Occipital Fasciculus, OR = Optic Radiation, UF = Uncinate fasciculus, OB WAY = Obligatory Waypoint(s), Excl = Exclusion mask(s), MNI152 = Montreal Neurological Institute’s 152, WM = White Matter).

Fig. 5

Schematic diagram of the dataset organization. A list of the participants to the dataset is provided and for each subject: anatomical T₁w and T₂w; DWI after the correction for EPI distortions and susceptibility effects, eddy currents and signal dropout, with b-values, b-vectors and the registration matrix to T₁w; for QSM, original magnitude and phase maps for the five echo times, final QSM reconstruction registered in T₁w space, with registration matrix to T₁w; VOIs (AF, CST, FAT, IFOF, OR, UF) for left and right hemisphere;.csv files containing the 107 radiomic features for each region. Images were anonymized and de-identified (T₁w = T₁-weighted, T₂w = T₂-weighted, DWI = Diffusion-Weighted Imaging, QSM = Quantitative Susceptibility Mapping, VOI = Volume of Interest).