Multi-class brain tumor MRI segmentation and classification using deep learning and machine learning approaches.

Area of Science:

• Medical Imaging Analysis
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Brain tumor classification via Magnetic Resonance Imaging (MRI) is vital for diagnosis and treatment.
• Differentiating tumor types (malignant vs. benign) is challenging, necessitating advanced computational methods.

Purpose of the Study:

• To evaluate deep learning (DL) and machine learning (ML) for brain tumor classification using MRI data.
• To explore the efficacy of computational approaches in improving diagnostic accuracy.

Main Methods:

• A dataset of 1200 brain tumor MRI images (DICOM) was processed and segmented using edge refined binary histogram segmentation (ER-BHS).
• Hybrid features were extracted and optimized to 11 key features using a correlation-based method.
• Multiple DL and ML classifiers were assessed via 10-fold cross-validation.

Main Results:

• The random committee (RC) classifier achieved the highest accuracy at 98.61% on the optimized dataset.
• Both DL and ML methods demonstrated effectiveness in automating brain tumor classification.
• The study highlights the performance of optimized feature sets in classification tasks.

Conclusions:

• DL and ML approaches show significant potential for enhancing medical image analysis.
• These computational techniques can improve diagnostic accuracy in brain tumor classification.
• The findings suggest a potential revolution in clinical workflows through AI-powered diagnostics.