Brain Tumour Segmentation based on Extremely Randomized Forest with high-level features.

Area of Science:

• Neuro-oncology
• Medical image analysis
• Machine learning for healthcare

Background:

• Gliomas are common and aggressive brain tumors requiring accurate segmentation for effective management.
• Tumor segmentation is challenging due to variability in size, location, and complex delineation even with advanced Magnetic Resonance Imaging (MRI).

Purpose of the Study:

• To develop a fully automatic and discriminative method for glioma segmentation.
• To improve the accuracy and efficiency of brain tumor delineation for clinical applications.

Main Methods:

• Utilized appearance- and context-based features for segmentation.
• Employed an Extremely Randomized Forest (Extra-Trees) classifier.
• Incorporated non-linear image transformations for feature extraction.

Main Results:

• Achieved a Dice score of 0.83 for complete tumor segmentation.
• Obtained Dice scores of 0.78 for the tumor core and 0.73 for the enhanced regions.
• Demonstrated competitive performance against other methods on the BraTS 2013 dataset.

Conclusions:

• The proposed automatic method offers a robust approach for glioma segmentation.
• The method shows potential for enhancing surgical planning, treatment monitoring, and patient follow-up.
• Results indicate the effectiveness of appearance- and context-based features in automated brain tumor segmentation.