Advancing multi-categorization and segmentation in brain tumors using novel efficient deep learning approaches | JoVE Visualize

Area of Science:

Medical Imaging and Diagnostics
Artificial Intelligence in Healthcare
Computational Biology

Background:

Brain tumors, abnormal cell growths, require early detection for improved patient prognosis.
Current magnetic resonance imaging (MRI) analysis methods for brain tumors are time-consuming and lack accuracy due to tumor variability.
Existing classification approaches struggle with the diverse sizes, shapes, and locations of brain tumors.

Purpose of the Study:

To develop an efficient deep learning-based approach for accurate brain tumor classification and segmentation.
To categorize brain tumors into types: pituitary, glioma, meningioma, and identify the absence of tumors.
To enhance the speed and precision of brain tumor analysis compared to existing methods.

Main Methods:

Introduced ERSACA-Net (extension residual structure and adaptive channel attention mechanism) for brain tumor classification.
Utilized Enhanced Res2Net for extracting critical features (shape, texture, color) from brain MRI scans.
Employed the Binary Chaotic Transient Search Optimization (BCTSO) Algorithm for feature selection and complexity reduction.
Developed a novel LWIFCM_CSA approach (Local-information weighted intuitionistic Fuzzy C-means clustering algorithm and Chameleon Swarm Algorithm) for segmentation.
Addressed class imbalance issues using Conditional Tabular Generative Adversarial Network (CTGAN).

Main Results:

The proposed ERSACA-Net demonstrated superior accuracy in classifying brain tumor types (pituitary, glioma, meningioma, no tumor).
The LWIFCM_CSA approach achieved improved segmentation of affected brain areas.
Achieved a significantly faster processing time, averaging 0.11 seconds, outperforming previous methods.
The ensemble approach showed stable improvements in accuracy, indicating robust performance for trustworthy brain tumor classification.

Conclusions:

The developed deep learning model (ERSACA-Net) and segmentation approach (LWIFCM_CSA) offer a more precise and efficient solution for brain tumor diagnosis.
The enhanced feature extraction and optimization techniques contribute to higher classification accuracy and reduced computational complexity.
This research paves the way for more reliable and faster clinical applications in brain tumor detection and analysis.