Microscopic brain tumor detection and classification using 3D CNN and feature selection architecture.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Brain tumors represent a significant global health challenge, causing numerous deaths annually.
• Accurate and timely diagnosis is crucial for effective treatment and improved patient survival rates.
• Current diagnostic methods require improvement, highlighting the need for automated techniques in brain tumor grading.

Purpose of the Study:

• To develop and validate a deep learning-based method for microscopic brain tumor detection and classification.
• To enhance the precision of brain tumor grading using advanced computational techniques.
• To address the limitations of existing methods by proposing a novel automated approach.

Main Methods:

• A 3D convolutional neural network (CNN) was designed for initial brain tumor extraction.
• Pre-trained CNN models were employed for feature extraction from the segmented tumors.
• Correlation-based feature selection and a feed-forward neural network were utilized for final classification.

Main Results:

• The proposed deep learning method achieved high accuracy rates on multiple BraTS datasets (2015, 2017, 2018): 98.32%, 96.97%, and 92.67%, respectively.
• Feature selection using the correlation-based method effectively identified the most relevant features for classification.
• The overall performance demonstrated the efficacy of the proposed automated technique in brain tumor analysis.

Conclusions:

• The developed deep learning model offers a precise and automated solution for brain tumor detection and classification.
• The method shows comparable accuracy to existing techniques, suggesting its potential for clinical application.
• This approach can aid radiologists and oncologists in making faster and more accurate diagnoses, potentially improving patient outcomes.