Study and analysis of different segmentation methods for brain tumor MRI application.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Medical Resonance Imaging (MRI) is crucial for brain tumor diagnosis, requiring effective image segmentation.
• Image segmentation simplifies complex MRI data, aiding in tumor identification, location, and size determination.
• Various segmentation algorithms exist, each with distinct performance characteristics.

Purpose of the Study:

• To compare the performance of multiple image segmentation algorithms for brain tumor diagnosis using MRI.
• To evaluate algorithms including Otsu's, watershed, level set, K-means, Discrete Wavelet Transform (DWT), and Convolutional Neural Network (CNN).
• To analyze algorithm performance based on accuracy, precision, recall, F-measure, and response time.

Main Methods:

• Simulation of segmentation algorithms in MATLAB using the BRATS-2018 dataset.
• Performance evaluation using metrics such as accuracy, precision, recall, and F-measure.
• Measurement of response time for each algorithm, particularly for CNN.

Main Results:

• Convolutional Neural Network (CNN) achieved the highest accuracy at 91.39%.
• Other methods showed varying accuracies: Otsu's (71.42%), watershed (78.26%), level set (80.45%), K-means (84.34%), and DWT (86.95%).
• CNN demonstrated a response time of 2.519 seconds in the MATLAB simulation.

Conclusions:

• CNN significantly outperforms traditional segmentation methods for brain tumor diagnosis from MRI.
• The findings guide researchers in selecting optimal algorithms for embedded systems and developing advanced machine learning models.
• The study emphasizes the potential of CNN and deep learning for developing efficient brain tumor detection hardware.