Robust Automatic 3D Brain Extraction on T1 Weighted Magnetic Resonance Images for dogs and cats

Digital brain extraction is a pre-processing method that is required for multiple neuroimaging applications. The development of such automated and accurate tools in veterinary neurology is a complex task due to the heterogeneous conformations of skulls (brachycephalic, dolichocephalic, mesocephalic) between species and breeds. A new brain extraction method, VIBE (Veterinary Images Brain Extraction) [1], based on atlases for dogs and cats, has recently been developed. The latter has been validated exclusively on supposedly healthy animals. Based on a cohort of 115 animals (cats and dogs), we propose the very first tool based on deep learning, robust to morphological variations of skull conformations of dogs and cats, across different diagnoses, with a Dice coefficient of 0.97, that outperforms the first atlas-based method VIBE. According to our previous work [1], this tool also outperforms existing human methods proven to perform less well for animals.Clinical relevance- This establishes a robust segmentation tool for canine and feline brain extraction with a Dice coefficient of 0.97.