Universal Soldier: A deep neural net for unsupervised 3D segmentation of tomographic images of bones

The process of identifying the object in an image, and the object background is accomplished by the human brain instantly. We develop such skills with experience, and often take it for granted. In digital imaging and computer vision, the same operation called image segmentation is a bottleneck of quantitative analysis. In particular, automated image segmentation is difficult in 3D, and especially so when the object is of biological origin. Automated segmentation of 3D datasets would abolish these limitations and increase the precision of quantitative image analysis, with high throughput. A way to mimic the precision and accuracy of human visual perception is to train an artificial neural network on a vast variety of examples. We have accrued a big library of 3D images of bones and skeletons, and will train the Universal Soldier Neural Net to recognize, tag and quantify the image components.

Faculty Supervisor:

Natalie Reznikov

Student:

Partner:

Object Research Systems

Discipline:

Computer science

Sector:

Information and cultural industries; Professional, scientific and technical services

University:

McGill University

Program: