The SabLab Journal Club wiki can be found here.
Thanks to the remarkable progress in hardware technologies, biomedical data are multiplying rapidly, improving in quality, and offering unprecedented means to examine the underlying biology. Today, biomedical data generation has outstripped the development of sophisticated computational tools that can efficiently and effectively extract information from these data. For example, the biggest challenge we are currently facing in examining potential patterns of associations between genotype, environment, anatomy, behavior, and clinical symptoms, is methodological. We are in need of methods that can unveil true, multivariate, and dynamic relationships of modest effect size, by examining high dimensional variables (e.g., millions of image voxels) and usually with limited sample size (e.g., thousands of subjects).
Broadly, our research focuses on developing novel tools for analyzing biomedical imaging scans, often in conjunction with other clinical data types. Our projects are motivated by a range of problems such as mapping and detecting anatomical changes due to pathology; characterizing the temporal dynamics of these alterations; studying pre-symptomatic neuroanatomical abnormalities for early diagnosis; quantifying disease severity; functionally characterizing risk genes associated with neural disorders; and making clinically-useful individual-level predictions, e.g., based on image and genotype data. Our research is also concerned with laying out the theoretical underpinnings of the analytic problems we face in these applications.
Statistical analysis, machine learning, signal/image processing, and computer vision have historically evolved separately to tackle different problems, and thus offer complementary viewpoints. Recently, however, the once-sharp boundaries between these fields have begun to blur. It is becoming increasingly clear that the methods that we desperately need to make sense of large-scale biomedical data will have to draw from multiple domains, including these fields. Our research program is positioned at this intersection, where our central aim is to develop analytic methods that will be instrumental in biomedical research, particularly neurology and neuroscience, where we are faced with challenging clinical and basic science problems.
As enthusiastic believers that software dissemination is an integral part of scholarly work, we will be dedicated to sharing and supporting implementations of the algorithms we develop. This will be a significant effort in our new lab at Cornell.