Deep-Stacked Convolutional Neural Networks for Brain Abnormality Classification Based on MRI Images.

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Deep Learning

Background:

• Automated diagnosis systems are vital for efficient identification of brain abnormalities by radiologists.
• Convolutional Neural Networks (CNNs) offer automated feature extraction but face challenges like limited labeled data and class imbalance in medical imaging.
• Accurate diagnoses often require integrating expertise from multiple clinicians or algorithms.

Purpose of the Study:

• To develop a robust deep heterogeneous model, Deep-Stacked CNN, for multi-class brain disease classification.
• To address the limitations of single CNNs when training data is insufficient.
• To improve the robustness and accuracy of automated diagnostic systems for brain abnormalities.

Main Methods:

• Proposed a two-level learning process utilizing stacked generalization.
• Selected diverse, pre-trained CNNs fine-tuned via transfer learning as base classifiers at the first level.
• Employed a neural network as a meta-learner at the second level to combine outputs from base classifiers for final prediction.

Main Results:

• The Deep-Stacked CNN model achieved a high accuracy of 99.14% on an untouched dataset.
• Demonstrated superior performance compared to existing methods in brain disease classification.
• Required fewer parameters and computations while maintaining outstanding diagnostic performance.

Conclusions:

• Deep-Stacked CNN effectively harnesses the strengths of multiple CNN classifiers to overcome data scarcity and class imbalance issues.
• The proposed model offers a robust and efficient solution for automated multi-class brain disease diagnosis.
• This approach represents a significant advancement in deep learning applications for medical image analysis.