Spatio-Functional segmentation for spectral images

In head-and-neck radiology, detecting tumors on 3D images through Computed Tomgraphy (CT) scans is already a challenging task for radiologists and is too hard for algorithms to be precise.

With the current Dual-Energy CT acquisition systems using two energy spectra of X-rays, an additional information is recorded: the energy decay curve of the element at each voxel.

We make use of this 4th dimension (in functional domain) to increase the quality of the image analysis (in spatial domain).


The major steps to achieve our tumor segmentation are:

  • spatio-functional clustering - to assemble consistent body regions,
  • classification of each cluster - to predict the outcome of tumor or healthy tissues.

This tumor detection project focuses now on developing a statistical method that comprises both functional and spatial data inside its core optimization. In a 2nd time it will focus on deep learning to accomplish the classification.

PhD Student:

Segolene Brivet


Supervisors and Collaborators:

Mark Coates

Peter Savadjiev (Department of Diagnostic Radiology, McGill)

Reza Forghani (Department of Diagnostic Radiology, McGill)

Faïcel Chamroukhi (University of Caen, France)