A Robust Brain Tumor Detector Using BiLSTM and Mayfly Optimization and Multi-Level Thresholding.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Traditional brain tumor detection methods are time-consuming and may miss small tumors.
• Accurate and early detection of brain tumors is crucial for effective treatment.
• Developing automated, efficient diagnostic tools is a key area in medical research.

Purpose of the Study:

• To propose an effective and automated approach for brain tumor detection and classification using deep learning.
• To improve the accuracy and efficiency of identifying brain tumors, particularly smaller ones.
• To compare the proposed method's performance against existing machine learning and deep learning techniques.

Main Methods:

• Brain tumor segmentation using the mayfly optimization algorithm and multilevel Kapur's thresholding on MRI scans.
• Feature extraction via Histogram of Oriented Gradients (HOG) and ResNet-V2.
• Tumor classification into pituitary, glioma, and meningioma using a bidirectional long short-term memory (BiLSTM) network.

Main Results:

• The proposed method achieved high accuracy, precision, recall, F1 score, and AUC on Figshare and Harvard datasets.
• Demonstrated superior performance compared to existing deep learning (DL) and machine learning (ML) methods.
• Successfully segmented and classified brain tumors with enhanced detection capabilities for small lesions.

Conclusions:

• The combined segmentation and feature fusion approach offers a significant improvement in brain tumor detection.
• This AI-driven methodology shows potential for clinical application, especially in identifying subtle or small tumors.
• Further clinical validation is recommended before widespread adoption in healthcare settings.