Optimal Control for the Analysis of Image Sequences
Principal Investigator:Marc Niethammer
Funding Agency: National Science Foundation
Agency Number:ECCS-0925875
Abstract
Research Goals Today’s imaging technology allows us to visualize and explore phenomena ranging from the dynamic interaction of proteins within a cell to measurements of brain structure, function, and tissue properties, in vivo. Imaging modalities from microscopy to magnetic resonance imaging provide increasingly better, and increasingly complex data, with researchers and practitioners requiring accurate and efficient data analysis. The shear amount of data created is breathtaking and continues to grow constantly, preventing manual analysis and requiring computer-assisted or computer-automated methods for information extraction and quantification.
Increased spatial resolution allows us to see mitochondria, developing embryos, and nerve fiber tracts. Increased temporal resolution will unlock even more information and necessitates image analysis algorithms specifically designed to uncover subtle temporal changes. Analysis of time-series of images is already important for applications as different as object surveillance, target tracking, and studies of brain development.
Our goal is to develop a general toolbox of methods for the analysis of time-varying image sequences. We choose the study of brain development as our test application, which exhibits a variety of challenging subproblems ranging from appearance changes throughout neurodevelopment, over data sampled sparsely in time, to the analysis of sets of image time-series. In the context of biomedical imaging, the techniques developed will ultimately lead to new insight into disease progression through in vivo monitoring, will enable early disease detection, and will also be a cornerstone to facilitate personalized medicine, which needs to account for individual developmental differences and age effects.
Our research will focus on an optimal control approach for the analysis of dynamically changing image sequences, by treating image evolution as a dynamical system. We will develop methods for longitudinal, cross-sectional, and random designs. Specifically, we will explore (1) the dynamic modeling of time-varying image data, (2) their optimal interpolation, (3) their optimal approximation and smoothing, (4) their optimal filtering, (5) as well as image regression, (6) the analysis of estimation results, and (7) efficient solution approaches based on optimal control theory.
Educational Goals Image analysis research spans a variety of disciplines and touches on various interest subtypes, ranging from the biology high-school student interested in the underlying application, over the computer science or mathematics student interested in the algorithmic details, to the medical researcher interested in answering medical questions through image analysis technology. Unfortunately, frequently gaps exist between algorithm development, algorithm deployment, and algorithm usage in practice. An important reason is lack of communication. The PI’s pedagogical goal is to reduce the communication barriers between research fields in the following ways: (1) by offering image analysis courses which include student group collaborations with research groups within biology and medicine, (2) by providing opportunities for undergraduates and summer students for hands-on image analysis projects, (3) by teaching non computer science majors about the fundamentals and practical aspects of image analysis, (4) and by open-source
dissemination of image analysis software.
Intellectual Merit Several novel methods for the analysis of time-varying images will be explored and developed within the research proposed. They have general applicability. The research will have immediate impact on current imaging studies and will form the basis for future applications. To assure extendability of the developed methods, source code will be made available to the community. This will allow for easy adaptations and the creation of customized image analysis solutions.
Broader Impact The developed methods will have broad applicability, from natural image tracking, to video processing and medical image analysis. Example uses will range from the analysis of microscopy images to monitor spatio-temporal phenomena in individual cells to the study of structural brain changes by magnetic resonance imaging.