Purpose: Computational modelling by use of neural networks was performed to assess morphology and composition of paraspinal muscles in individuals for spinal surgery to better assist with decision making in management. Methodology: Pre-operative T2-weighted MRI axial images from 27 participants who had either lumbar microdiscectomy (n=14) or lateral lumbar interbody fusion (n=13) were obtained. Convolutional neural network models were trained for segmentation of the multifidus, erector spinae, and psoas major muscles on each side. Cross-sectional volume and fatty infiltration (%) measures of each paraspinal muscle were calculated and analysed. Results: There was an increased fatty infiltration in all paraspinal muscles in participants who were to undergo lateral lumbar interbody fusion, compared to participants for microdiscectomy. In both groups, multifidus muscle had much higher fatty infiltration (36%), followed by erector spinae muscle (30%), then psoas major muscle (12%). Conclusion: Computational modelling showed a global increase in fatty infiltration specific to the muscles producing extensor torque on the lumbar spine in individuals with low back pain. This was accentuated in advanced spinal disease requiring a lateral lumbar interbody fusion. Further study, including predictive modelling, is warranted to assess correlation between composition of paraspinal muscles to post-operative outcome, and hence to enhance surgical decision making in individuals with low back pain.