A Systematic Approach for MRI Brain Tumor Localization and Segmentation Using Deep Learning and Active Contouring.

Area of Science:

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

Background:

• Accurate tumor boundary annotation and segmentation are critical for effective tumor extraction.
• Traditional edge detection methods often struggle with the complexities of medical image segmentation.
• Deep learning offers promising solutions for enhancing medical image analysis tasks.

Purpose of the Study:

• To develop and evaluate a robust deep learning architecture for precise tumor boundary segmentation.
• To address the limitations of conventional methods in medical image segmentation.
• To improve the accuracy and reliability of tumor extraction processes.

Main Methods:

• A threefold deep learning architecture combining a deep convolutional neural network (CNN) for classification and a region-based convolutional neural network (R-CNN) for localization.
• Utilizing the Chan-Vese segmentation algorithm, an active contour model, for precise tumor boundary contouring.
• Employing a suite of performance metrics including Dice Score, Rand Index (RI), and Mean Absolute Error (MAE) for comprehensive evaluation.

Main Results:

• The proposed deep learning architecture achieved an average Dice Score of 0.92 for glioma and meningioma segmentation.
• Excellent performance was also demonstrated with an average RI of 0.9936, VOI of 0.0301, GCE of 0.004, BDE of 2.099, PSNR of 77.076, and MAE of 52.946.
• The results indicate high reliability and accuracy compared to expert demarcations (gold standard).

Conclusions:

• The developed threefold deep learning architecture provides a highly reliable and accurate solution for tumor segmentation.
• The Chan-Vese algorithm effectively contours tumor boundaries, overcoming limitations of gradient-based methods.
• This approach holds significant potential for advancing tumor extraction and analysis in clinical practice.