An optimized deep neural network with explainable artificial intelligence framework for brain tumour classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate brain tumour classification is crucial for effective patient care and treatment planning.
• Magnetic Resonance Imaging (MRI) is a primary tool for visualizing brain tumours.
• Deep learning models offer potential for automated and accurate image analysis.

Purpose of the Study:

• To develop and evaluate an optimized ResNet framework for improved brain tumour classification using MRI.
• To tune key hyperparameters of ResNet models (ResNet-18 and ResNet-50) for enhanced performance.
• To validate the model's effectiveness on distinct brain tumour datasets using comprehensive metrics.

Main Methods:

• Implementation of ResNet-18 and ResNet-50 architectures for brain tumour classification.
• Optimization of ResNet hyperparameters (momentum, learning rate, epochs, validation frequency) using the Henry gas solubility optimization (HGSO) algorithm.
• Evaluation of the optimized models on two MRI databases (4 and 3 tumour classes) using accuracy, sensitivity, specificity, precision, and F-score.
• Utilization of Gradient-weighted Class Activation Mapping (GRAD-CAM) for model interpretability.

Main Results:

• The optimized ResNet-50 framework achieved a highest classification accuracy of 0.9825 on Database1.
• The HGSO algorithm effectively tuned ResNet hyperparameters, leading to superior classification performance.
• GRAD-CAM visualizations confirmed the model's focus on relevant tumour features, ensuring reliable predictions.

Conclusions:

• The proposed HGSO-optimized ResNet framework significantly enhances MRI brain tumour classification accuracy.
• This deep learning optimization strategy offers a promising approach for improving diagnostic precision in neuro-oncology.
• The interpretability provided by GRAD-CAM builds confidence in the model's clinical applicability.