Enhanced MRI brain tumor detection using deep learning in conjunction with explainable AI SHAP based diverse and...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Computational Neuroscience

Background:

• Conventional diagnostic methods for brain tumors have limitations including low resolution, radiation exposure, and poor contrast.
• Magnetic Resonance Imaging (MRI) offers high-resolution, non-invasive visualization crucial for accurate tumor characterization.
• Advanced feature representation and machine learning algorithms are needed to enhance diagnostic capabilities.

Purpose of the Study:

• To investigate the synergistic performance of multiple feature representation schemes and learning algorithms for MRI-based brain tumor identification.
• To evaluate the efficacy of local Binary Patterns (LBP) in conjunction with classifiers like Support Vector Classifier (SVC) and Convolutional Neural Networks (CNN).
• To assess the generalization power of proposed models using both small and large benchmark datasets.

Main Methods:

• Exploration of feature representation schemes: local Binary Patterns (LBP), Gabor filters, Discrete Wavelet Transform, Fast Fourier Transform, and Gray-Level Run Length Matrix.
• Application of learning algorithms: k-nearest Neighbor, Random Forest, Support Vector Classifier (SVC), probabilistic neural network (PNN), and Convolutional Neural Networks (CNN).
• Statistical analysis (chi-square, p-value) and SHAP analysis were employed to validate feature importance and classification impact.

Main Results:

• Local Binary Patterns (LBP) combined with Support Vector Classifier (SVC) and Convolutional Neural Networks (CNN) demonstrated high specificity and accuracy in initial tests.
• On a small dataset, SVC achieved 98.06% accuracy and CNN achieved 97.8% accuracy.
• On a large benchmark dataset, CNN yielded the highest accuracy at 98.9%, followed by SVC at 96.7%, indicating strong generalization.

Conclusions:

• The combination of MRI-based feature extraction, particularly LBP, with advanced algorithms like CNN offers a highly accurate and automated approach to brain tumor diagnosis.
• Convolutional Neural Networks (CNN) exhibit superiority in medical imaging due to their ability to learn intricate spatial patterns and generalize effectively.
• This integrated approach enhances the accuracy, speed, and consistency of brain tumor detection, potentially improving patient outcomes and healthcare efficiency.