Deep Learning-Assisted Segmentation and Classification of Brain Tumor Types on Magnetic Resonance and Surgical...

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Medical Imaging

Background:

• Neurosurgical procedures require precise tumor boundary delineation and classification.
• Current methods may lack the accuracy needed for optimal surgical planning and execution.

Purpose of the Study:

• To enhance neurosurgical procedures using deep learning for accurate tumor boundary delineation and classification.
• To provide surgeons with advanced diagnostic tools for improved surgical outcomes and patient care.

Main Methods:

• The study employed the Mask R-convolutional neural network (CNN) architecture with Resnet101 and Resnet50 backbones.
• Models were trained using data from surgical microscope videos and preoperative magnetic resonance images.
• Performance was evaluated using metrics including accuracy, precision, recall, Dice coefficient (DICE), and Jaccard index.

Main Results:

• The Mask R-CNN Resnet 101 architecture achieved 96% precision, 93% recall, 91% DICE, and 84% Jaccard index.
• The Mask R-CNN Resnet 50 architecture yielded 94% precision, 89% recall, 89% DICE, and 82% Jaccard index.
• The model demonstrated 98% accuracy in pathology estimation with a DICE score range of 94%-95%.

Conclusions:

• Deep learning integration represents a transformative advancement in neurosurgery.
• The developed algorithm shows significant promise for distinguishing between different tumor types.
• Further research emphasizing diverse datasets is crucial for refining deep learning models in neurosurgery.