Brain CT image classification based on mask RCNN and attention mechanism

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Machine learning and blockchain are increasingly used in healthcare for tasks like image analysis and diagnosis.
• Brain tumors are a significant global health concern with rising mortality rates.
• Accurate classification of brain CT images is crucial for timely diagnosis and treatment.

Purpose of the Study:

• To propose an enhanced Mask RCNN model for accurate classification of brain CT images.
• To improve the segmentation precision of tumor edges in medical images.
• To enhance the overall classification performance in brain tumor detection.

Main Methods:

• Utilized ResNet-10 as a backbone for local feature extraction in brain CT images.
• Incorporated deformable convolution within residual modules and attention mechanisms.
• Implemented parallel spatial and channel attention mechanisms with deformable convolution for global feature extraction.
• Improved the loss function to optimize target edge segmentation accuracy within the Mask RCNN framework.

Main Results:

• The proposed method effectively refines edge features in brain CT images.
• Demonstrated increased separation between the target tumor and background.
• Achieved improved classification performance on a public brain CT dataset.

Conclusions:

• The enhanced Mask RCNN model with attention mechanisms shows significant potential for improving brain tumor classification from CT scans.
• The integration of deformable convolution and improved loss functions contributes to more precise segmentation and better classification outcomes.
• This approach offers a promising tool for auxiliary diagnosis in neuro-oncology.