Per-voxel classification of Modic changes in conventional MRI

Motivation: Modic Changes (MCs) are often heterogeneous and difficult to classify objectively. Hence, new diagnostic tools are required to improve MC classification. Goal(s): This study aims to develop a data-driven model for the classification of MC lesions on a per-lesion and per-voxel level from conventional MR images. Approach: Conventional MR images from 12 patients were used to create an MC classification model by fitting three multivariate normal distribution functions to the MRI MC data which was subsequently used for MC classification. Results: The model reached high accuracy (74-100%), enabling a detailed classification on a per-voxel level and longitudinal tracking of MC transitions. Impact: This study introduces a data-driven model for classifying Modic changes in vertebral bone marrow using MR images. The model achieves high accuracy, enabling detailed classification and tracking of Modic change transitions, potentially improving patient diagnosis and care.