CV Science

Segmentation of brain tumors is an important task for treatment planning and therapy evaluation. This task could also lead to new applications, including data compression, robust registration, and effective content based image retrieval in large medical databases. Accurate delineation of tumor can also be helpful for general modeling of pathological brains and the construction of pathological brain atlases. Nevertheless, precise delineation of brain Tumor in MRI is a challenging problem that depends on many factors. Indeed, there is a large class of tumor types which vary greatly in size and position, have a variety of shape and appearance properties, have intensities overlapping with normal brain tissue, may deform and defect the surrounding structures giving an abnormal geometry also for healthy tissue. Moreover, MR images segmentation widely depends on the specific application and image modality. These images contain sometimes various amounts of noise and/or artifacts due to patient's motion and soft tissue boundaries are sometimes not well defined.

Our purpose is to develop new methods for the automatic detection, characterization and quantification of brain tumors in three-dimensional magnetic resonance images. So, these methods might become an efficient tool for clinic applications.

Brain tumor diagnosis by Deformable Model and multi-modality MR images

The existence of several MR acquisition protocols provides different information on the brain. Each image usually highlights a particular region of the tumor. In visualizing brain tumors, a second T1-weighted image is often acquired after the injection of a 'contrast agent'. The presence of this type of 'enhancing' area can indicate the presence of a tumor. Although the presence of this 'enhancement' can be a strong indicator of tumor location, there exist a large variety of types of brain tumors, and their appearance in MR images can vary considerably.

Deformable models are popular methods that are widely used for a wide range of applications and have proved to be a successful segmentation technique. They have become used for medical image processing and especially for brain segmentation, implicitly in the form of a level set function or explicitly as a snake function.

In general, precise and reproducible segmentation of brain tumors are still a challenging and difficult task which is far from being solved, even if much effort has been spent in the medical imaging community.

As with most other works, we focus on the development of a generic algorithm that could help in the automation of medical image analysis tasks. Our work takes place in this growing area and we are mainly motivated by the deformable model approaches. Our purpose is to propose an unsupervised method that incorporates additional information to better disambiguate the tumor from the surrounding deformed brain tissue. 

This part of the project is the continuation of our previous team works in this field (Bourouis et al. IJIG 2010, Bourouis et al. ICIAR  2008).