Clinical imaging increasingly makes use of longitudinal image studies to examine subject-specific changes due to pathology, intervention, therapy, neurodevelopment, or neurodegeneration. Moreover, dynamic organ changes as seen in cardiac imaging or functional changes as measured in perfusion imaging, just to name a few, by definition result in time-series image data presenting volumetric image data over time. Expressions such as development, degeneration, disease progress, recovery, monitoring or heart cycle inherently carry the aspect of a dynamic process – suggesting that imaging at multiple time points may be applied. The detection and characterization of changes from baseline due to disease, trauma, or treatment require novel image processing and visualization tools for qualitative and quantitative assessment of change trajectories. Whereas longitudinal analysis of scalar data is well known in the statistics community, its extension to high-dimensional image data, shapes or functional changes poses significant challenges. Cross-sectional analysis of longitudinal data does not provide a model of growth or change that considers the inherent correlation of repeated images of individuals, nor does it tell us how an individual patient changes relative to a change over time of a comparable healthy or disease-specific population, an aspect which is highly relevant to decision making and therapy planning. Although successful early results were presented for image regression in aging studies, such standard regression is not optimal for longitudinal data because repeated observations from the same individual would ignore the correlation between repeated measurements and thus violate the Gauss-Markov assumption of independence. Moreover, individual change trajectories often need to be interpreted in relationship to a population growth model, which in turn is the hidden group model given a representative set of individual trajectories, and require a common framework based on the use of hierarchical linear (or nonlinear) models (HLM). Other typical driving applications are concerned with registration of serial data of the cardiac cycle, sampled at different time points, or measuring object shape changes via shape regression, both requiring new image registration and modeling approaches.

The special nature of longitudinal or repeated, time-series data of individual subjects, with the inherent correlation of structure and function across the sequence of images, resulted in the development of a variety of new image processing and analysis approaches tackling the challenging issues of registration, segmentation and analysis in the presence of geometric and contrast changes over time. New methodologies are rapidly evolving, often focusing on the specific application at hand.

This workshop will cover a comprehensive discussion of multiple approaches and new advances for spatio-temporal image processing of longitudinal image data but also aims at a dialogue to define the generic nature of algorithms, methods, modeling approaches, and statistical analysis for optimal analysis of such data. This in turn will lead to a discussion on solutions achieved so far, summary of the state-of-the-art and open issues to be tackled in future research.

The primary focus of this workshop will be on the generic nature of algorithms, methods, modeling approaches and statistical analysis. We are specifically looking for papers that introduce novel and innovative methodology for spatiotemporal analysis of images, with methods driven by challenging applications. We will invite paper submissions related to the following topics:

List of topics:

•  Regression of sequences of time-series image data
• Modeling of shape changes due to development, degeneration, pathology, or evolution
• Modeling of growth and change trajectories, predictive modeling of “disease”
• Modeling of space occupying and infiltration patterns as seen in tumor staging
• Physiological modeling of pathology evolution to extract model parameters characterizing
• Joint segmentation of time-series image data
• Joint registration of time-series data
• Group discrimination via multi-dimensional longitudinal statistical data analysis
• Others

Paper submissions on novel methodology development for spatiotemporal analysis might be driven by typical challenging applications as listed below.

Novel image analysis approaches driven by:

• Analysis of heart shape across heart cycle
• Study of early brain growth trajectory
• Study of healthy aging and/or aging in disease
• Study of perfusion changes in time-series image data
• Monitoring of therapeutic intervention (baseline and multiple follow-up measures)
• Staging of disease, e.g., MS lesion and brain atrophy evolution or tumor growth
• Others