A Small Intestinal Stromal Tumor Detection Method Based on an Attention Balance Feature Pyramid

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Small intestinal stromal tumors (SIST) are common gastrointestinal tumors.
• Current diagnosis relies on subjective and inefficient radiologist review of CT images.
• Automatic detection using computer vision faces challenges due to SIST's varied appearance and subtle differences from normal tissue.

Purpose of the Study:

• To develop an improved computer vision algorithm for accurate and efficient automatic detection of SIST in CT images.
• To address the feature imbalance issue in mainstream target detection models during feature fusion for SIST detection.

Main Methods:

• Proposed an Attention Balance Feature Pyramid (ABFP) algorithm for SIST detection.
• ABFP combines weighted multi-level feature maps, creating a balanced semantic feature map enhanced by spatial and channel attention modules.
• The algorithm enhances original feature information by integrating the enhanced balanced semantic feature map into the feature fusion stage.

Main Results:

• The ABFP algorithm significantly improved the detection performance of SIST detection models.
• The method effectively addressed the imbalance between deep and shallow features during network model processing.
• Experimental results demonstrated ABFP's compatibility with various models and feature fusion strategies.

Conclusions:

• The ABFP algorithm offers a versatile and effective solution for enhancing traditional target detection methods in medical imaging.
• This approach shows promise for improving the accuracy and efficiency of automatic SIST detection.
• The attention-based feature balancing mechanism is crucial for overcoming challenges in detecting subtle abnormalities like SIST.