Brain tumour segmentation

1.-Labeling

Diferent clases presents in the dataset.

One of the international contest where we have participated is the BRAT contest, the aim is to use multispectra resonance datasets carefully labeled by trained physician.

Every label corresponds to a diferent tumour stage, from beningn tumours to growing agresive, or necresed region and more.

One of the major contributions of our group was to provide enhanced methods for 3D convulational networs, this kind of networks performs very well when comparing the multiple layeres from the 3D image dataset, to progress in the tumour clasification, using the multisepectral filters to better detect lession.

2 Multispectral image showing diferent textures

Data clasification requieres the extra information providede by diferent RMS modalities.

Why this tumour segmentation is important?

The publication shows various patters, first of all is that in a very competitive enviromnent, most of the models still perform with clinical accuracy, the reason for this is related about the way pixel classification is evaluated. In this environment, we were able to construct better models by providing esnambling of all participants, showing for these case a performance close to 99.9% whihc basically means that at this stage, whith this image modality this can be considered as a solved problem.

Tumour segmentation helps to several clinical issues associated to diferent clinical aspects of this kind of analysis.

• Volume determination of the lession
• Disease evolution
• Malingant and evolution patterns of the lession
• Precise location for quirurgical or radiotherapy tasks
• Volume determination of the lession
• Precise dose evaluation in radiotherapy, and time reduction of the cost of the segmentation tasks.

The location and relative position of lessions founds can determinate what is the most precisse treatment to apply, and what are the risks associated to every issue.

Main organizators

• Center for Biomed- ical Image Computing and Analytics (CBICA), University of Pennsylvania (UPenn), PA, USA,
• University of Alabama at Birmingham, AL, USA,
• Heidelberg University, Germany,
• University Hospital of Bern, Switzerland, 5) University of Debrecen, Hungary,
• Henry Ford Hospital, MI, USA, 7) University of California, CA, USA,
• MD Anderson Cancer Center, TX, USA, 9) Emory University, GA, USA, 10) Mayo Clinic, MN, USA,
• Thomas Jefferson University, PA, USA,
• Duke University School of Medicine, NC, USA,
• Saint Joseph Hospital and Medical Center, AZ, USA,
• Case Western Reserve University, OH, USA,
• University of North Carolina, NC, USA,
• Fondazione IRCCS Instituto Neuroligico C. Besta, Italy,
• MD Anderson Cancer Center, TX, USA,
• Washington University School of Medicine in St. Louis, MO, USA, and
• Tata Memo- rial Center, Mumbai, India. Note that data from institutions 6-16 are provided through The Cancer Imaging Archive (TCIA - http://www.cancerimagingarchive.net/), supported by the Cancer Imaging Program (CIP) of the National Cancer Institute (NCI) of the National Institutes of Health (NIH).

Publications & Scientific outputs