Deep learning methods for automatic segmentation of lower leg muscles and bones from MRI scans of children with and without cerebral palsy

• Published in: NMR in biomedicine 2021-12, Vol.34 (12), p.e4609-n/a
• Main Authors: Zhu, Jiayi, Bolsterlee, Bart, Chow, Brian V. Y., Cai, Chengxue, Herbert, Robert D., Song, Yang, Meijering, Erik
• Format: Article
• Language: English
• Subjects: Adolescent; Biological products; Bone and Bones - diagnostic imaging; Bones; Cerebral palsy; Cerebral Palsy - diagnostic imaging; Child; Child, Preschool; Children; Deep Learning; Entropy; Female; Humans; Image processing; Image segmentation; Leg; Leg - diagnostic imaging; Legs; lower leg; Magnetic resonance imaging; Male; MRI; muscle segmentation; Muscle, Skeletal - diagnostic imaging; Muscles; Paralysis; Performance evaluation; Sample Size; Two dimensional models
• ISSN: 0952-3480; 1099-1492
• DOI: 10.1002/nbm.4609

Cerebral palsy is a neurological condition that is known to affect muscle growth. Detailed investigations of muscle growth require segmentation of muscles from MRI scans, which is typically done manually. In this study, we evaluated the performance of 2D, 3D, and hybrid deep learning models for automatic segmentation of 11 lower leg muscles and two bones from MRI scans of children with and without cerebral palsy. All six models were trained and evaluated on manually segmented T1‐weighted MRI scans of the lower legs of 20 children, six of whom had cerebral palsy. The segmentation results were assessed using the median Dice similarity coefficient (DSC), average symmetric surface distance (ASSD), and volume error (VError) of all 13 labels of every scan. The best performance was achieved by H‐DenseUNet, a hybrid model (DSC 0.90, ASSD 0.5 mm, and VError 2.6 cm3). The performance was equivalent to the inter‐rater performance of manual segmentation (DSC 0.89, ASSD 0.6 mm, and VError 3.3 cm3). Models trained with the Dice loss function outperformed models trained with the cross‐entropy loss function. Near‐optimal performance could be attained using only 11 scans for training. Segmentation performance was similar for scans of typically developing children (DSC 0.90, ASSD 0.5 mm, and VError 2.8 cm3) and children with cerebral palsy (DSC 0.85, ASSD 0.6 mm, and VError 2.4 cm3). These findings demonstrate the feasibility of fully automatic segmentation of individual muscles and bones from MRI scans of children with and without cerebral palsy. Manual segmentation of muscles from MRI scans is tedious, subjective, and time‐consuming and requires expert anatomical knowledge. This study demonstrates, for the first time, that deep learning models can segment muscles from MRI scans of children with and without cerebral palsy with performance similar to that of human raters. Hybrid models that combine two‐ and three‐dimensional features performed particularly well. The automated segmentation methods will enable large‐scale investigations of typical and disordered muscle growth.