Evolution of Deep Learning Algorithms for MRI-Based Brain Tumor Image Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Extracting brain tumor textures from MRIs is difficult for neuroradiologists.
• High-grade gliomas necessitate rapid diagnosis and intervention due to their aggressive nature.
• Manual segmentation methods are insufficient for these complex cases.

Purpose of the Study:

• To conduct a comprehensive review of MRI-based brain tumor image segmentation using CNNs.
• To explain the typical CNN process chain for brain MRI segmentation.
• To compare various CNN architectures for their effectiveness in brain tumor analysis.

Main Methods:

• Surveying existing literature on CNNs for brain tumor segmentation.
• Detailing the structure and advantages of CNN architectures.
• Classifying and comparing CNN models based on complexity and performance metrics.

Main Results:

• CNNs offer superior feature extraction quality and generalizability compared to other machine learning algorithms.
• Various CNN architectures have been developed and classified for brain tumor segmentation.
• Performance comparison of CNNs is presented using specific metrics and datasets.

Conclusions:

• Deep learning, especially CNNs, presents a powerful computer-assisted approach for brain tumor segmentation in MRIs.
• CNNs enhance the precision and efficiency of diagnosing and treating brain tumors.
• This review provides a valuable resource for understanding and advancing CNN applications in neuroimaging.