Search

Introduction: The relevance of spine MRI data to understanding low back pain (LBP) is questioned but it remains one of the few ways to assess pathoanatomical features of the condition. A need for more reliable quantitative interpretation of clinical MRI data has arisen in step with growing interest in the genetic architecture of LBP and associated systemic and immunometablolic factors. Methods: We aim to quantitatively classify disc degeneration (DD) imaging phenotypes using deep learning (DL). We have identified from the Northern Finland Birth Cohort target phenotypes derived from qualitative DD schemas and Modic change size. We will use DL and radiomic analysis to train models classifying these target phenotypes, externally validated on TwinsUK. We will then use the results to refine phenotype definitions. These data-driven phenotypes will be investigated in genome wide association studies (GWAS) to locate associated single nucleotide polymorphisms and immunometabolic pathways. Results: We expect the resultant model to accurately classify DD phenotypes, allowing coding of large cohorts for GWAS and other data-driven approaches identifying biological aspects of LBP. Discussion: For MRI to benefit LBP sufferers, quantitative approaches to interpretation are needed. DL can contribute to effective sub-grouping based on MRI that can impact both research and clinical management. Process evaluation: This study is limited by analysing MRI data independent of psychosocial and systemic variables. Future DL studies should tackle this with multimodal solutions. [ABSTRACT FROM AUTHOR]