Journal of Medical Diagnostic Methods
Open Access

Abstract

Improved Mask RCNN Network for Brain Tumor MRI Image Instance Segmentation Analysis

Ming Him Foun*

In the pursuit of advancing the precision of brain tumor MRI image detection and segmentation for the purpose of fulfilling the requirements of automated medical analysis, this research introduces an enhanced Mask R-CNN method specifically tailored for high-precision brain tumor instance segmentation. The augmentation involves the incorporation of the Convolutional Block Attention Module (CBAM) hybrid attention mechanism, aimed at improving the model’s feature extraction capabilities and adaptively reinforcing its responsiveness to critical features. This enhancement facilitates a more precise capture of key tumor information. Furthermore, the integration of the Bi-directional Feature Pyramid Network (BiFPN) feature fusion technology ensures the model’s ability to accurately segment brain tumors of diverse sizes and shapes, thereby enhancing its capacity to identify and segment multi-scale targets. Through a series of rigorous experimental validations, the proposed model demonstrates notable improvements. The precision of the model attains 90.79%, marking a 0.67% enhancement compared to the original model. Similarly, the recall achieves 91.44%, indicating a 0.79% improvement, while the mean Average Precision (mAP) reaches 95.12%, reflecting a substantial increase of 1.88%. Beyond achieving accurate segmentation of brain tumor MRI images, the proposed method excels in precisely calculating the tumor’s area and diameter. Consequently, these findings furnish valuable reference data for medical research and diagnosis, underscoring the potential clinical significance of the developed methodology.

Published Date: 2024-08-02; Received Date: 2024-06-29