Soft Null Hypotheses: A Case Study of Image Enhancement Detection in Brain Lesions

• Activity Number: 672
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Haochang Shou*+ and Russell Shinohara and Han Liu and Daniel S. Reich and Ciprian Crainiceanu
• Companies: Johns Hopkins Bloomberg School of Public Health and The Johns Hopkins University and The Johns Hopkins University and National Institute of Neurological Disorders and Stroke and The Johns Hopkins University
• Keywords: hypothesis testing ; soft null distribution ; principal components analysis ; multiple sclerosis ; DCE-MRI ; contrast enhancement

Abstract

The cornerstone of statistical testing is a well-defined null hypothesis, which we refer to as a "hard null hypothesis". In contrast, many scientific null hypotheses of interest are defined qualitatively. We argue that statistical thinking can help in the transition to hard hypotheses through the examination of a sequence of soft nulls. The work is motivated by a population study of multiple sclerosis (MS) patients using dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) to identify active brain lesions with abnormal blood flow. During each visit, a contrast agent is administered intravenously and multiple images are taken over time to show which areas of the brain have altered magnetic properties due to the presence of the contrast agent, observed as enhancement. Our goal is to identify and quantify enhancement properties of lesions. Because the brain also contains many types of tissues that normally enhance, defining abnormal lesion enhancement is heavily dependent on the definition of "normal", or "null" enhancement. We provide a sequence of soft null hypotheses that describe non-enhancement from various perspectives and develop techniques for testing.