An Infusion Containers Detection Method Based on YOLOv4 with Enhanced Image Feature Fusion.

Area of Science:

• Computer Vision
• Medical Imaging
• Artificial Intelligence

Background:

• Current infusion container detection methods struggle in complex clinical environments.
• High accuracy and efficiency are crucial for reducing medical staff workload.

Purpose of the Study:

• To propose a novel, enhanced You Only Look Once version 4 (YOLOv4) method for accurate infusion container detection.
• To improve detection performance in challenging clinical settings.

Main Methods:

• Incorporated a coordinate attention module to enhance spatial and directional perception.
• Replaced the spatial pyramid pooling (SPP) module with a cross stage partial-spatial pyramid pooling (CSP-SPP) module for feature reuse.
• Integrated an adaptively spatial feature fusion (ASFF) module with the path aggregation network (PANet) for multi-scale feature fusion.
• Utilized Epsilon Intersection over Union (EIoU) as the loss function to address anchor aspect ratio issues.

Main Results:

• The proposed method demonstrated significant improvements in recall.
• Enhanced timeliness in detection was observed.
• Mean average precision (mAP) was notably increased compared to conventional methods.

Conclusions:

• The novel YOLOv4-based method offers superior performance for infusion container detection.
• This advancement contributes to reducing medical staff workload and improving clinical efficiency.