Personalizing in silico models of tumours with in vivo medical images


Year:

Session type:

Nicholas Ayache

INRIA (French Research Institute of Computer Science and Automatic Control), Sophia-Antipolis, France

Abstract

Personalizing in silico models of tumours with in vivo medical images

We developed a computational model of brain gliomas to simulate their evolution through time. The model has three main components:

1. a geometric component, built from pre-operative MR and DT images, which describes the 3-D shape of important head structures (skull, grey matter, white matter fibres, CSF, Falx, etc.). This information is completed by the 3-D surface of the visible boundary of the brain tumour.

2. a biomechanical component, based on a finite-element mesh computed from the previous 3-D shapes, to provide the head structures with inhomogeneous anisotropic linear elastic behaviour (small deformations).

3. a physiopathological component, modelling the evolution of the tumoural cell density on the previous mesh with a reaction diffusion partial differential equation which mimics proliferation and diffusion. The diffusion is privileged in the direction of white matter fibres, and the proliferation increases locally the pressure to induce a brain deformation (mass effect).

This model was confronted to longitudinal series of patient MRIs to show its capacity to reproduce qualitatively the complex shape of evolving tumours and the brain deformation on a few manually adjusted cases. We then developed mathematical and computational methods to automatically identify a number of key parameters of the model, namely the diffusion coefficient (in white and grey matter) and the proliferation rate. Their product is related to the visible evolution of the boundary front in time series of MRI, while their ratio is related to the actual extension of the tumour extension beyond its visible boundary. http://www.springerlink.com/content/03938m22x5n361x7/

Preliminary results on real cases show the potential benefit of coupling such in silico models with in vivo images at a macroscopic scale, in particular for a more personalized planning of radiotherapy margins. The recent development of in vivo confocal microscopy opens new possibilities for a calibration of such models at microscopic scales too.

Acknowledgements

This work was done in Asclepios project team at INRIA Sophia-Antipolis in collaboration with Harvard and MIT at Boston and CAL in Nice.

Declaration of competing interest: Nicholas Ayache is co-founder of and consultant with the Mauna Kea Technologies company which develops new in vivo confocal microscopy imaging techniques. These techniques will be evoked at the end of the talk as potential tools for calibration of the presented models at a microscopic scale.