Brain MR images segmentation using 3D CNN with features recalibration mechanism for segmented CT generation

The segmentation of MR (magnetic resonance) images is a simple approach to create Pseudo CT which are useful for many medical imaging analysis applications. One the main challenges this process bone brain images. Deep convolutional neural networks (CNNs) have been widely and efficiently applied perform segmentation. aim work propose novel excitation-based CNN by recalibrating network features adaptively enhance segmenting into three tissue classes: bone, soft tissue, air. proposed method combines two types excitation mechanisms namely: (1) spatial squeeze channel block (cSE) (2) (sSE). blocks combined sequentially integrated seamlessly 3D encoder decoder network. novelty emerges in combination improve performance reduce model complexity. evaluated through comparison with computed tomography (CT) as ground truth validated other methods literature that deep approaches image PET attenuation correction. Brain datasets consist 50 patients used evaluate method. classes using precision, recall, dice similarity coefficient (DSC), Jaccard index. presented improves compared baseline where DSC improved from 0.6278 ± 0.0006 0.6437 an improvement percentage 2.53% class. architecture demonstrates promising competitive results reduces complexity thanks sequential blocks.