Automatic brain extraction and brain tissues segmentation on multi-contrast animal MRI

www.nature.com/scientificreports OPEN Automatic brain extraction and brain tissues segmentation on multi‑contrast animal MRI Jamil Nour Eddin , Hugo Dorez  & Valentina Curcio * For many neuroscience applications, brain extraction in MRI images is the first pre-processing step of a quantification pipeline. Once the brain is extracted, further post-processing calculations become faster, more specific and easier to implement and interpret. It is the case, for example, of functional MRI brain studies, or relaxation time mappings and brain tissues classifications to characterise brain pathologies. Existing brain extraction tools are mostly adapted to work on the human anatomy, this gives poor results when applied to animal brain images. We have developed an atlas-based Veterinary Images Brain Extraction (VIBE) algorithm that encompasses a pre-processing step to adapt the atlas to the patient’s image, and a subsequent registration step. We show that the brain extraction is achieved with excellent results in terms of Dice and Jaccard metrics. The algorithm is automatic, with no need to adapt the parameters in a broad range of situations: we successfully tested multiple MRI contrasts (T1-weighted, T2-weighted, T2-weighted FLAIR), all the acquisition planes (sagittal, dorsal, transverse), different animal species (dogs and cats) and canine cranial conformations (brachycephalic, mesocephalic, dolichocephalic). VIBE can be successfully extended to other animal species, provided that an atlas for that specific species exists. We show also how brain extraction, as a preliminary step, can help to segment brain tissues with a K-Means clustering algorithm. Brain extraction, also known as skull stripping, is a preliminary image post-processing technique that is fundamental for multiple applications in neuroscience and quantitative image analysis for clinical and research purposes. To perform quantitative analysis on the brain, a segmentation of the brain parenchyma is needed: functional Magnetic Resonance Imaging (fMRI), for example, is a technique that highlights activated regions in the brain, when the subject is stimulated with a specific stimulus pattern. To ensure that the considered activation signals are only those located in the brain, the analysis needs a preliminary skull stripping step. Skull stripping also enhances performances in the operation of inter-subjects brain normalisation, used to draw comparisons in fMRI results among different subjects1. Other examples are the calculations for quantitative MRI mappings applied to brain lesions characterisation, such as T1 and T2 relaxation times or quantitative susceptibility m aps2,3. The brain extraction operation selects the brain as the only region of interest, leading to faster and more specific analysis, focused only on the brain itself. However, automatic brain extraction tools for the animal brain are still limited, and mostly dedicated to non-domestic animals such as rodents, macaque and marmoset4–6. We believe that a brain extraction tool that can be extended to multiple animal species, including domestic animals such as dogs and cats, can push the development of new quantitative neuroimaging tools to be integrated in veterinary research, clinical and pre-clinical practice. One of the most successful brain extraction technique for humans is the atlas-based s egmentation7. A brain atlas is elastically registered to the full anatomical image, and used to extract the brain parenchyma from surrounding non-brain tissues. Many popular automatic brain extraction tools include atlases in their process, such as 3dSkullStrip, part of the AFNI s uite8, the Hybrid Watershed A lgorithm9, combining watershed for initialisation and atlas registration, or the Skull Stripping function part of the Insight Toolkit (ITK), a popular library for medical image p rocessing10. Since these tools use human atlases, they can not be adapted to animals, due to the variability of their anatomy depending on the species. Other popular tools, even if not explicitly based on human atlases, make use of hyperparameters tuned on the human skull conformation. It is the case of the Brain Extraction Tool (BET)11, available as part of the FMRIB Software Library (FSL), and of F reeSurfer12, that makes use of Bayesian priors. The Brain Surface Extraction (BSE) tool integrated in the B rainSuite13,14 is based on edge detection and morphological operations, such as erosion. The animal skulls, such as dogs’ and cats’, have a larger amount of tissues surrounding the brain, and the boundary between the brain and the skull is not always well contrasted. This makes it hard to separate non-brain tissues based only on intensity and morphological operations. HawkCell, Marcy‑l’Étoile 69280, France. *email: Scientific Reports | (2023) 13:6416 | https://doi.org/10.1038/s41598-023-33289-7 1 Vol.:(0123456789) www.nature.com/scientificreports/ The problem arises especially in the rostral portion of the brain (towards the nasal cavity and the buccal cavity), in the ventral portion of the brain (towards the nasopharynx), and laterally (towards the masticatory muscles). For these reasons, an automatic brain extraction algorithm based on animal morphology that can be used in veterinary clinical and research routines is needed. For it to be adapted to different animal morphologies, it needs to be based on atlases of a specific animal species. Nowadays, animal digital atlases are being created, that can be integrated in image processing pipelines. They are available for the d og15–20, the cat21,22, the domestic p ig23, the 24–26 27 28,29 30,31 32,33 34 sheep , the h orse , the b aboon , the m acaque , the m armoset and the rat brain . We have developed VIBE (Veterinary Images Brain Extraction): an atlas-based brain extraction algorithm adapted to the animal’s anatomy. We are showing the application and robustness of VIBE in particular on domestic dogs and cats, but the principle can be applied to every animal species, provided that an MRI atlas exists. We are also showing that an atlas built on a specific MRI contrast can be successfully used for brain extraction of MRI volumes of multiple contrasts, resolutions and acquisition orientations. We quantitatively assess the quality of the brain extraction problem over a cohort of cats and dogs of different races and brain conformations. As an application, we also show how brain extraction as a preliminary step can simplify the brain tissues segmentation task, and leads to precise separation of the cerebrospinal fluid, white and gray matter. Methods Study population. All the animals chosen to be part of this retrospective study were patients of the veterinary clinic VetAgro Sup (Marcy-l’Étoile, France), who were prescribed an MRI examination, performed by HawkCell (Marcy-l’Étoile, France). All the images were anonymized to assure the patients’ confidentiality. All procedures were approved by the Comité d’éthique de (...truncated)