Development of representative magnetic resonance imaging–based atlases of the canine brain and evaluation of three methods for atlas-based segmentation
- Autores: Marjorie E. Milne, Christopher Steward, Simon M. Firestone, Sam N. Long, Terrence J. O'Brien, Bradford A. Moffat
- Localización: American Journal of Veterinary Research, ISSN-e 1943-5681, ISSN 0002-9645, Vol. 77, Nº. 4, 2016, págs. 395-403
- Idioma: inglés
- Enlaces
-
Resumen
OBJECTIVE To develop representative MRI atlases of the canine brain and to evaluate 3 methods of atlas-based segmentation (ABS).
ANIMALS 62 dogs without clinical signs of epilepsy and without MRI evidence of structural brain disease.
PROCEDURES The MRI scans from 44 dogs were used to develop 4 templates on the basis of brain shape (brachycephalic, mesaticephalic, dolichocephalic, and combined mesaticephalic and dolichocephalic). Atlas labels were generated by segmenting the brain, ventricular system, hippocampal formation, and caudate nuclei. The MRI scans from the remaining 18 dogs were used to evaluate 3 methods of ABS (manual brain extraction and application of a brain shape–specific template [A], automatic brain extraction and application of a brain shape–specific template [B], and manual brain extraction and application of a combined template [C]). The performance of each ABS method was compared by calculation of the Dice and Jaccard coefficients, with manual segmentation used as the gold standard.
RESULTS Method A had the highest mean Jaccard coefficient and was the most accurate ABS method assessed. Measures of overlap for ABS methods that used manual brain extraction (A and C) ranged from 0.75 to 0.95 and compared favorably with repeated measures of overlap for manual extraction, which ranged from 0.88 to 0.97.
CONCLUSIONS AND CLINICAL RELEVANCE Atlas-based segmentation was an accurate and repeatable method for segmentation of canine brain structures. It could be performed more rapidly than manual segmentation, which should allow the application of computer-assisted volumetry to large data sets and clinical cases and facilitate neuroimaging research and disease diagnosis.
