Segmentation and grading of brain tumors on apparent diffusion coefficient images using self-organizing maps

Area of Science:

• * Medical imaging analysis
• * Artificial intelligence in oncology
• * Computational neuroscience

Background:

• * Accurate brain tumor segmentation and grading are crucial for effective treatment planning.
• * Existing methods may lack the precision required for differentiating tumor components like necrosis and edema.
• * Advanced computational techniques offer potential for improved diagnostic accuracy.

Purpose of the Study:

• * To develop an accurate computer-assisted method for brain tumor segmentation and grading using apparent diffusion coefficient (ADC) images.
• * To integrate unsupervised artificial neural networks (ANN) and hierarchical multiresolution wavelet transforms for enhanced image analysis.
• * To evaluate the performance of the proposed method against manual segmentation.

Main Methods:

• * ADC images were decomposed using multiresolution wavelets and reconstructed to create filtered images.
• * Wavelet filtered images, along with FLAIR and T2 weighted images, served as features for a self-organizing map (SOM) neural network.
• * A novel segmentation algorithm based on Best Matching Unit (BMU) hits on SOM maps was developed.

Main Results:

• * The SOM effectively differentiated patterns of tumor, edema, necrosis, cerebrospinal fluid (CSF), and normal tissue on ADC images.
• * The method successfully identified high or low-grade tumors, edema, necrosis, CSF, and normal tissue.
• * Validation against manual segmentation yielded high sensitivity (0.86) and specificity (0.93).

Conclusions:

• * The proposed computer-assisted method demonstrates high accuracy in segmenting and grading brain tumors.
• * The combination of wavelets and SOM provides a robust tool for analyzing complex brain tissue patterns.
• * This technique holds promise for improving the diagnostic workflow in neuro-oncology.