Image and Geometry Processing

Geometric model of a segment of the spine generated from a CT image set.

This new CIBC technical core develops algorithms and software for the processing or analyzing of signals and imaging data in biomedicine. Signals are samplings of any measured quantity of time; biomedical examples include voltage from electrically active tissues or the flow of biological fluids in vessels. Images are samples in space of similar functional quantities as well as structural information. Images may also vary with time and thus share characteristics with signals. We can also represent spatial information in concise geometric forms, often derived from images, and these geometries can also be time dependent. Signals, images, and geometries arise in many biomedical applications and their processing and manipulation is essential in many area of biomedical research. Thus it is natural to have a single core that includes support for signals, images, and geometry and seeks to develop integrated software for the associated operations.

Signal, image, and geometry processing is a field with great breadth and a wide range of maturity. Some areas are quite mature, yet the resulting algorithms are inaccessible to biomedical scientists. The reason can be either because the algorithms are not available as robust programs or are not part of an integrated system for data processing. Some algorithms are available but require specific tuning and adjusting before they operate properly in a particular application domain—biomedical scientists are rarely equipped to carry out the required tuning so that the algorithms remain inaccessible. Other needs do not yet have dependable solutions and are the topics of research in computer science, engineering, and mathematics.

The goal of the CIBC is to improve access of biomedical scientists to advanced algorithms and methods in signal, image, and geometry processing. We will achieve this goal by implementing state of the art algorithms and adjusting them to the needs of particular collaborations and the fields they represent. We will also seek to integrate algorithms into comprehensive processing toolkits that will allow scientists to perform all related steps for the processing and analysis of their data, from viewing raw image data to creating complex geometric models of the underlying anatomy.

An interactive segmentation PowerApp. The tool combines real-time segmentation based on a level-set method with interactive volume rendering. Users can initialize/modify models, sample image statistics, and tune parameters. Left: The segmentation of the cortex from MRI. Right: The segmentation of a mouse liver from MRI.

Probabilistic classification of brain data, showing cerebro-spinal fluid, white matter, and gray matter.

Automatic segmentation of a spiny dendrite. Panels a) and b) show the initial and refined versions, respectively of one portion. Panel c) contains a volume rendering of the complete segmentation.