The field of neuroimaging statistics is concerned with elucidating meaningful conclusions from high-dimensional imaging objects, often in the form of single-dimensioned summary statistics. Ideally, these summaries should provide interpretable biomarker measurements that can guide patient diagnoses or treatment decisions while minimizing information loss associated with dimension reduction. This dissertation is focused on (1) exploring methods for analyzing previously developed imaging biomarkers and (2) developing new imaging biomarkers using both well-established and novel imaging analysis techniques. We approach this problem in three ways: in our first project, we assess how previously developed imaging biomarkers can best be incorporated into downstream analyses in the context of a clinical trial. This work conceptualizes imaging biomarkers as measurements which intrinsically contain historical information on a patient and examines the effect of incorporating these predictors on the statistical power in a clinical trial analysis. For our second project, we develop a radiomic predictor that automatically identifies an important prognostic biomarker in multiple sclerosis, relying on quantification of imaging patterns potentially associated with brain atrophy and more severe disease courses. In our third project, we construct a coordinate system and framework for multiple sclerosis lesions analyses for more sensitive and specific biomarker development. We use dimension reduction and flexible nonparametric modelling to assess the diagnostic value of this method. These methods lay the groundwork for improving future work developing and utilizing imaging biomarkers with imaging statistics.