Application of LightGBM hybrid model based on TPE algorithm optimization in sleep apnea detection.

Area of Science:

• Cardiology
• Sleep Medicine
• Artificial Intelligence

Background:

• Sleep apnea syndrome (SAS) is a significant health concern requiring early detection to improve patient outcomes and reduce healthcare costs.
• Polysomnography (PSG), the current gold standard, is costly, time-consuming, and disruptive.
• Electrocardiogram (ECG) analysis presents a promising, simpler, and less invasive alternative for diagnosing sleep apnea.

Purpose of the Study:

• To develop and validate an efficient LightGBM hybrid model for sleep apnea detection using ECG signals.
• To address data challenges like abnormal data and sample imbalance using improved algorithms.
• To optimize model parameters for enhanced diagnostic accuracy.

Main Methods:

• An improved Isolated Forest algorithm was employed for data preprocessing, including outlier removal and addressing class imbalance.
• The Tree-structured Parzen Estimator (TPE) algorithm was utilized to optimize the parameters of the LightGBM model.
• A fusion model, TPE_OptGBM, was developed and evaluated on a public sleep apnea ECG database.

Main Results:

• The TPE_OptGBM model demonstrated high diagnostic performance with an accuracy of 95.08%.
• The model achieved a precision of 94.80%, a recall of 97.51%, and an F1 score of 96.14%.
• These results surpassed those of current mainstream diagnostic models.

Conclusions:

• The proposed TPE_OptGBM model offers a highly accurate and efficient method for sleep apnea detection via ECG analysis.
• This AI-driven approach has the potential to serve as a valuable diagnostic aid for clinicians.
• Implementing this model can lead to improved patient experiences and potentially reduced healthcare burdens associated with sleep apnea diagnosis.