Neuroimaging Big Data Analysis using Machine Learning: Novel Directions, Opportunities and Challenges
Affiliation: CBICA University of Pennsylvania
Abstract: Machine learning has played a key role in the rapid evolution of the neuroimaging field over the past few decades. Multivariate learning techniques enabled accurate segmentation of normal and abnormal anatomy for quantitative analyses, development of individually-based imaging indices for establishing diagnostic and prognostic biomarkers, and identifying imaging patterns of subtle brain changes that characterize neurodegenerative diseases and conditions. However, clinical translation of these novel methods and techniques has often been slow and limited. A major reason for this gap was the limited scope of machine learning applications to small clinical samples. In recent years, multiple efforts towards “neuroimaging big data analysis” aimed to address these limitations by pooling large multi-study samples. These efforts were driven by important advances in the field. Particularly, novel deep learning techniques achieved state-of-the-art performance with extensive training on large datasets, and they have been a major new direction in neuroimaging research.
In this talk, I will briefly present the iSTAGING (Imaging-based coordinate SysTem for AGing and NeurodeGenerative diseases) consortium, an international consortium that pools scans from approximately 32,000 individuals aged 45 and older. I will discuss challenges for harmonizing and integrating imaging and clinical data from various sites, and opportunities for fully leveraging the potential of recent advances in machine learning, specifically new semi-supervised methods for obtaining a deeper understanding of the “heterogeneity” of imaging signatures and deep learning applications.