Segmenting extremely small but important subcortical brain structures, such as the amygdala's subregions, would be useful in neuroimaging studies of many neurological disorders yet it remains very challenging. No reliable segmentation tools is currently available for amygdala subregions. We developed a 3D fully convolutional neural network to segment the amygdala and its subregions with high accuracy. Our method yields excellent segmentations and is much more efficient than multi-atlas based methods.

This work was supported by NARSAD: Brain and Behavior grant 24103 (to BN) and NIH grant funding NINDS R01 NS092870, NIMH P50 MH100031 and the Waisman Center U54 IDDRC from the Eugene Kennedy Shriver National Institute of Child Health and Human Development (U54 HD090256). NA was also supported in part by the BRAIN Initiative R01-EB022883-01, CPCP U54-AI117924-03, the Alzheimer's Disease Connectome Project (ADCP) UF1-AG051216-01A1 and R56-AG052698-01.