Efficient Skip Connections-Based Residual Network (ESRNet) for Brain Tumor Classification.

Area of Science:

• Medical Diagnostics
• Artificial Intelligence
• Neuroscience

Background:

• Brain tumors present a significant diagnostic challenge due to their complexity and potential severity.
• Existing deep learning (DL) models for brain tumor classification (BTC) face limitations such as vanishing gradients and difficulty capturing intricate features.

Purpose of the Study:

• To propose an efficient skip connections-based residual network (ESRNet) to overcome the limitations of current DL models for BTC.
• To enhance feature learning and pattern recognition in brain tumor classification.

Main Methods:

• Developed ESRNet, leveraging residual networks (ResNet) with skip connections to ensure smooth gradient flow and mitigate vanishing gradients.
• Integrated multiple stages with increasing residual blocks, efficient downsampling, and batch normalization for robust performance.
• Utilized skip connections for identity mapping, preserving gradient flow and enabling effective information propagation.

Main Results:

• ESRNet demonstrated superior performance over other approaches in accuracy, sensitivity, specificity, F-score, and Kappa statistics.
• Achieved median performance metrics including 99.62% accuracy, 99.68% sensitivity, and 99.89% specificity.
• Minimum performance metrics, such as 99.34% accuracy and 99.79% specificity, highlight ESRNet's exceptional effectiveness.

Conclusions:

• The proposed ESRNet offers exceptional performance and efficiency for brain tumor classification.
• ESRNet has the potential to significantly advance clinical diagnosis and treatment planning for brain tumors.