Brain Tumor Detection and Classification on MR Images by a Deep Wavelet Auto-Encoder Model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Brain tumor diagnosis is complex due to intricate brain structures.
• Machine learning, particularly deep learning, shows promise in medical image analysis for tumor detection.
• Existing methods require improvement in accuracy and efficiency.

Purpose of the Study:

• To propose a novel Deep Wavelet Autoencoder (DWAE) model for automated brain tumor detection and classification.
• To enhance the quality of MRI slices for better tumor visualization.
• To accurately differentiate between normal and abnormal brain MR images.

Main Methods:

• A Deep Wavelet Autoencoder (DWAE) model was developed for binary classification of MRI slices (tumor vs. no tumor).
• Preprocessing involved high-pass filtering for heterogeneity, median filtering for merging, and edge highlighting for image enhancement.
• The model utilized a seed growing method and a two-layer network with 200 and 400 hidden units, followed by a softmax layer for classification.
• The model was trained and tested on 2500 MR brain images from datasets including BRATS and ISLES.

Main Results:

• The DWAE model achieved a high accuracy of 99.3% in detecting brain tumors.
• The model demonstrated a low loss validation of 0.1.
• Low False Positive Rate (FPR) and False Negative Rate (FNR) values were recorded, indicating high reliability.
• The model efficiently analyzed pixel patterns for accurate tumor classification.

Conclusions:

• The proposed Deep Wavelet Autoencoder (DWAE) model offers a highly accurate and efficient solution for automatic brain tumor detection.
• The DWAE model's ability to analyze pixel patterns and classify tumors demonstrates its potential to assist clinicians in diagnosis.
• This research facilitates the automatic detection of brain tumors, potentially improving patient outcomes.