Creating Computer Vision Models for Respiratory Status Detection.

Area of Science:

• Medical imaging
• Computer vision
• Artificial Intelligence in Medicine

Background:

• Accurate detection of ventilation objects (masks, tubes) on patients is crucial for medical monitoring.
• Computer vision offers potential for automating this detection process.
• Existing methods may lack real-time processing capabilities.

Purpose of the Study:

• To develop and compare computer vision models for detecting and recognizing ventilation objects on patient faces.
• To evaluate the performance of You Only Look Once (YOLO) and Transfer Learning (TL) models for this task.
• To assess the clinical relevance of automated object detection in medical settings.

Main Methods:

• Development of two distinct computer vision models: YOLO and TL.
• Training and testing models on datasets of patient images with ventilation objects.
• Quantitative performance evaluation using metrics such as accuracy and precision.
• Comparison of the detection and recognition capabilities of both models.

Main Results:

• Both YOLO and TL models achieved high performance in detecting ventilation objects.
• The Transfer Learning (TL) model exhibited a performance rate of 93%.
• The You Only Look Once (YOLO) model also achieved a performance rate of 93%.

Conclusions:

• Computer vision models, particularly TL, show significant promise for automated detection of medical equipment.
• Findings are relevant for healthcare providers and researchers in medical computer vision.
• The study highlights the potential of real-time video stream analysis for patient monitoring.