Background

Centerlines (i.e.skeletons) are an extremely efficient way to represent the size, shape and topology of items in a digital image. They are widely used in computer vision, image processing, pattern recognition, data compression and are becoming commonly used in medical imaging. However, although extracting the centerline from a 2D shape is relatively straight forward, the problem is substantially more complex for 3D objects (especially those that are not smooth and/or uniform).

Technology Overview

University of Manitoba researchers have developed and validated a computationally efficient, fully automated, centerline extraction method that produces one-pixel thin, continuous, and spatially accurate skeletons from complex 3D shapes. The method uses a ‘divide and conquer’ algorithm, in which any 3D object is first sliced into a series of 2D images in X, Y and Z directions. A geometric (Voronoi) algorithm is then applied to each planar image in order to extract the 2D centerlines before recombining the information using an intersection technique that is able to reconstruct the centerline of the original 3D object.

This method requires relatively short processing times for even complex medical images and is compatible with any medical imaging modality (MRI, CT, etc.), provided segmentation is done to define the structure of interest. Post-segmentation, the method could theoretically identify the 3D centerline of the 3D segmentation and be used to quantify a variety of features (e.g., cross-sectional area along the centerline, mean image intensities within the cross-sectional areas along the centerline, etc.).

Further Details

S Younas and CR Figley (2019). Development, Implementation and Validation of an Automatic Centerline Extraction Algorithm for Complex 3D Objects. Journal of Medical and Biological Engineering, 39(2):184-204.

Benefits

This approach has been validated using both standard 3D benchmark objects, as well as more complex anatomical structures segmented from medical images, demonstrating that the method is able to accurately extract centerlines.

Unlike other centerline extraction methods, this algorithm is computationally efficient, requiring relatively short processing times for both the 3D benchmark objects as well as complex medical images and thus can be run on standard desktop computers.

Applications

This centerline extraction method has wide-ranging utility in the field of 3D medical imaging, including data reduction for large microscopy datasets and automated quantification medical images. It is currently being developed for automated quantification of anatomical features from medical imaging data but could be broadly applied to a variety of medical and non-medical applications.

With the 3D medical imaging field, the method can be used for a variety of uses, including (but not limited to):

3D tumor grading/measurements
quantifying tumor, brain, eye or cardiac vasculature
along-track white matter quantification
muscle and/or tendon tracking/measurements.

In addition, the algorithm may also be used to create a database of structures and their centerlines for training a Deep Learning (e.g., convolutional neural network, etc.) algorithm for ultra-fast and extremely computationally-efficient 3D centerline quantification.

Opportunity

The centerline extraction method has been validated and benchmarked using various 3D objects (including anatomical structures segmented from medical imaging data). It is now being developed for quantifying MRI characteristics along-structures (e.g., axons, blood vessels, etc.) and for use in Coronary Artery Imaging using CT angiography image datasets.

The University is interested in discussing licensing and collaboration opportunities with respect to how our automated 3D centerline approach could be integrated into commercial products in both the clinical and non-clinical fields.

Extracting Centerlines From Medical Imaging Data

Patents

IP Status

Seeking