Irregular breathing classification from multiple patient datasets using neural networks.

Area of Science:

• Medical Physics
• Biomedical Engineering
• Radiation Oncology

Background:

• Irregular breathing patterns complicate accurate prediction of respiratory motion for precise radiation dose delivery.
• Previous research on irregular breathing patterns has been limited to monitoring extreme respiratory phases.

Purpose of the Study:

• To develop and validate a novel neural network-based approach for classifying irregular breathing patterns.
• To improve the quantitative modeling of irregular respiratory motion for radiation therapy.

Main Methods:

• Retrospective categorization of breathing data from 448 patients using feature metrics.
• Development of a neural network classifier that utilizes reconstruction error to model breathing patterns.
• Classification based on a 'regular ratio' to determine breathing pattern regularity.

Main Results:

• The proposed neural network classifier effectively categorized breathing patterns.
• The system demonstrated clinical merit in the statistical quantitative modeling of irregular breathing motion.
• Sensitivity, specificity, and ROC curve analysis validated the performance of the irregular breathing pattern detector.

Conclusions:

• The developed neural network approach provides a robust method for detecting and classifying irregular breathing patterns.
• This technique enhances the accuracy of respiratory motion prediction, crucial for precise radiation dose delivery.
• The study validates the proposed classifier's effectiveness across a large patient cohort.