“…Fully automatic algorithms based on active statistical shape modelling have been applied to cochlear segmentation 25 , 26 , but to attain an accurate statistical shape in different scenarios, a large amount of annotated data would be required 27 . Other proposed solutions such as atlas-based frameworks 11 , 12 and iterative random-walks algorithm with shape prior integration produced encouraging results for cochlea segmentation but are computationally expensive 28 , 29 . Moreover, with shape priors and atlas-based methods, segmentation might fail if the analysed image diverges from the average shape model, and this is often the case in malformations.…”