Konsep Inti
This study introduces a novel system for detecting human respiratory anomalies using infrared light-wave sensing and machine learning algorithms.
Abstrak
The key highlights and insights of this content are:
The study developed an infrared (IR) light-wave sensing system for non-contact respiratory monitoring. This system uses an IR light source, a photodetector, and a lock-in amplifier to capture breathing patterns.
To create labeled training data, a robot was used to simulate various normal and abnormal breathing patterns, including eupnea, apnea, tachypnea, bradypnea, hyperpnea, hypopnea, and Kussmaul's breathing. A separate "faulty data" class was also included to detect erroneous data caused by external disturbances.
Four handcrafted features were extracted from the collected data: peak-to-peak amplitude, breathing rate, effective spectral amplitude, and signal-to-noise ratio. These features were used to train three machine learning models - decision tree, random forest, and XGBoost - for the breathing anomaly detection task.
The models were evaluated using 10-fold cross-validation. The random forest model achieved the highest classification accuracy of 96.75% for data collected at a 0.5 m distance. Ensemble models like random forest and XGBoost generally outperformed the single decision tree model, especially when classifying data collected at multiple distances.
The system was able to effectively detect faulty data, with the faulty data class exhibiting the highest precision and recall scores. However, some misclassifications occurred between adjacent breathing classes with borderline values of breathing rate and depth.
The study also conducted binary and 3-class classifications to distinguish between normal, abnormal, and faulty breathing, achieving high accuracies of over 93% with the ensemble models.
Overall, this work establishes a promising framework for non-contact respiratory anomaly detection using infrared light-wave sensing and machine learning, with potential applications in healthcare monitoring and other domains.
Statistik
The peak-to-peak amplitude of the eupnea (normal breathing) data collected at 1.5 m distance was 0.022 V.
The breathing rate of the tachypnea (fast breathing) data was 30 breaths per minute.
Kutipan
"Abnormal breathing can indicate fatal health issues leading to further diagnosis and treatment."
"Automated breathing monitoring solutions generally employ contact-based or wearable sensors, which may be unsuitable for certain patients and can spread contagious diseases."