Open-access ultrasonic diaphragm dataset and an automatic diaphragm measurement using deep learning network

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Respiratory Physiology

Background:

• Diaphragm function assessment is vital for managing diaphragmatic dysfunction.
• Current manual measurement methods are prone to human error.
• Objective and automated assessment tools are needed.

Purpose of the Study:

• To develop and evaluate an automatic diaphragm measurement system using a segmentation neural network.
• To compare the performance of the automatic system against manual clinical assessments.
• To enhance the accuracy and efficiency of diaphragm parameter measurement.

Main Methods:

• Development of a novel deep learning segmentation network (MDRU-Net) for diaphragm thickness and excursion.
• Utilized a new ultrasound diaphragm dataset with B-mode and M-mode images/videos.
• Implemented an automated measurement system and plan.

Main Results:

• The automatic system achieved an average error of 8.12% for diaphragm thickening fraction.
• The automatic system achieved an average relative error of 4.3% for diaphragm excursion.
• Demonstrated minor discrepancies compared to manual assessments, with enhanced clinical detection potential.

Conclusions:

• A diaphragm ultrasound dataset and an automatic segmentation algorithm (based on U-Net) were developed.
• The automatic measurement scheme demonstrated high accuracy and efficiency.
• The system eliminates subjective influence, advancing automated diaphragm ultrasound assessment.