Poster Presentation 51st International Society for the Study of the Lumbar Spine Annual Meeting 2025

INTRODUCTION

In spinal research, paraspinal musculature is increasingly becoming the focus of attention. The quality of the musculature is often associated with pathologies of the spine. Semi-quantitative classifications are often used, which divide muscles into groups based on the proportion of fatty infiltration (FI) to assess the degree of muscle degeneration. Within one group, however, there are diverse patterns in the distribution of fat in the muscle. To create a base for the classification of FI patterns, the aim of the present study was to identify similarities in FI patterns in MRI images of the paraspinal muscles using unsupervised machine learning methods.

METHODS

This retrospective analysis is based on an examination of axial T2-weighted MRI images (see Fig. 1a) at disc level L4/5 of 1,148 patients with disc degeneration. The muscle group consisting of erector spinae and multifidus (see Fig. 1b) was automatically segmented using a U-net [1]. 27 features were extracted, including the mean and standard deviation of local parts of the muscle group, distances of prominent points, and the number and geometric properties of visible structures (see Fig. 1c-f). To identify clusters of similar FI patterns, K-Means was applied to the feature dataset. This study was approved by the responsible ethics committee.

RESULTS

The aim of identifying similar FI patterns could not be achieved. However, the results of the cluster analysis showed a remarkable similarity to common semi-quantitative classifications. To facilitate a comparison between the clustering results presented here and other classification schemes, the Rand score was utilized. The corresponding values for the comparison with a four-level classification [2] are 0.62, and for the comparison with the well-known Goutallier classification [3], 0.73. The observed concordance between the identified clusters is a notable finding.

DISCUSSION

The distribution of FI in the paraspinal musculature is so diverse that it was not possible to identify uniform pattern types using the method employed. However, the presented results exhibit a noteworthy agreement with existing semi-quantitative classification systems. This illustrates the significance of such classifications for the assessment of FI in the clinical practice of radiologists, surgeons, and scientists. In the future, it may be beneficial to consider combining the Goutallier classification with selected image features.