A hybrid explainable model based on advanced machine learning and deep learning models for classifying brain tumors...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Brain tumors pose a significant global health challenge, necessitating early detection and accurate classification for effective treatment.
• Current diagnostic methods require improvement in speed and accuracy for diverse brain tumor types.

Purpose of the Study:

• To develop and evaluate a novel, computationally efficient framework for accurate classification of four brain tumor types (glioma, meningioma, no tumor, pituitary) using MRI images.
• To enhance feature visibility and extraction for improved diagnostic performance.

Main Methods:

• Utilized Contrast-Limited Adaptive Histogram Equalization (CLAHE) to improve MRI image quality.
• Employed a lightweight Parallel Depthwise Separable Convolutional Neural Network (PDSCNN) for feature extraction.
• Developed a hybrid Ridge Regression Extreme Learning Machine (RRELM) for enhanced classification.
• Validated performance using five-fold cross-validation and compared against state-of-the-art models.

Main Results:

• Achieved high average precision (99.35%), recall (99.30%), and accuracy (99.22%) in classifying four brain tumor types.
• The proposed PDSCNN-RRELM framework demonstrated superior performance compared to the Pseudoinverse Extreme Learning Machine (PELM) and other models.
• Ridge regression integration significantly improved ELM classification performance, model parameters, and layer sizes.

Conclusions:

• The novel PDSCNN-RRELM framework offers a highly accurate and efficient solution for brain tumor classification from MRI data.
• The method enhances diagnostic confidence through improved accuracy and interpretability via SHAP analysis.
• This approach holds promise for improving clinical decision-making in neuro-oncology.