Deep learning assessment of inflammation in axial spondylarthritis and quantitative analysis of myelin water in multiple sclerosis

Abstract

Magnetic resonance imaging (MRI) is a valuable tool for diagnosing inflammatory diseases, such as axial spondyloarthritis (SpA) and multiple sclerosis (MS). However, interpreting certain types of MRI images, such as short tau inversion recovery (STIR) and fluid-attenuated inversion recovery (FLAIR) scans for SpA and MS, respectively, presents challenges due to the need for experienced physicians and the low reliability between different readers. To address these limitations, this thesis aims to explore the use of deep learning in axial SpA and assess the feasibility of applying multiple inversion recovery (mIR) magnetic resonance fingerprinting (MRF) in MS. These efforts have the potential to improve the accuracy and reliability of MRI interpretation in disease related conditions. The text below outlines several vital studies on applying deep learning or innovative MR techniques in medical imaging for conditions, including SpA and MS. In the first study, a deep learning model was trained using a dataset of 389 participants' sacroiliitis MRIs. The deep learning model exhibited improved performance through “fake-color” imaging, as indicated by satisfactory sensitivity, specificity, and positive predictive value at various evaluation levels. A deep learning-based scoring pipeline was developed using the Spondyloarthritis Research Consortium of Canada system, integrating various deep learning models, including the one from the first study. The high intraclass correlation coefficient values suggested a strong consistency between human readers and the deep learning-based scoring pipeline. In addition to sacroiliitis, the feasibility of a deep learning model in spinal MRI was investigated using STIR MRIs from 247 participants. The findings indicated that the deep learning model's sensitivity, specificity, and positive predictive value at the image and scan levels were comparable to those of a general radiologist. These experiments demonstrated the potential of deep learning in managing SpA. The final study focused on applying mIR MRF with multicompartment analysis to assess myelin water, a biomarker of myelin, in MS patients. Statistical analysis highlighted significant differences (P-value < 0.01) between healthy controls and MS patients, demonstrating that the myelin water fraction (MWF) could provide valuable insights about demyelination. Notably, MWF was able to detect all MS lesions in the white matter, including those that were previously invisible in magnetization-prepared rapid acquisition with gradient echo (MPRAGE).