InsightSleepNet: the interpretable and uncertainty-aware deep learning network for sleep staging using continuous Photoplethysmography.

BACKGROUND: This study was conducted to address the existing drawbacks of inconvenience and high costs associated with sleep monitoring. In this research, we performed sleep staging using continuous photoplethysmography (PPG) signals for sleep monitoring with wearable devices. Furthermore, our aim was to develop a more efficient sleep monitoring method by considering both the interpretability and uncertainty of the model's prediction results, with the goal of providing support to medical professionals in their decision-making process. METHOD: The developed 4-class sleep staging model based on continuous PPG data incorporates several key components: a local attention module, an InceptionTime module, a time-distributed dense layer, a temporal convolutional network (TCN), and a 1D convolutional network (CNN). This model prioritizes both interpretability and uncertainty estimation in its prediction results. The local attention module is introduced to provide insights into the impact of each epoch within the continuous PPG data. It achieves this by leveraging the TCN structure. To quantify the uncertainty of prediction results and facilitate selective predictions, an energy score estimation is employed. By enhancing both the performance and interpretability of the model and taking into consideration the reliability of its predictions, we developed the InsightSleepNet for accurate sleep staging. RESULT: InsightSleepNet was evaluated using three distinct datasets: MESA, CFS, and CAP. Initially, we assessed the model's classification performance both before and after applying an energy score threshold. We observed a significant improvement in the model's performance with the implementation of the energy score threshold. On the MESA dataset, prior to applying the energy score threshold, the accuracy was 84.2% with a Cohen's kappa of 0.742 and weighted F1 score of 0.842. After implementing the energy score threshold, the accuracy increased to a range of 84.8-86.1%, Cohen's kappa values ranged from 0.75 to 0.78 and weighted F1 scores ranged from 0.848 to 0.861. In the case of the CFS dataset, we also noted enhanced performance. Before the application of the energy score threshold, the accuracy stood at 80.6% with a Cohen's kappa of 0.72 and weighted F1 score of 0.808. After thresholding, the accuracy improved to a range of 81.9-85.6%, Cohen's kappa values ranged from 0.74 to 0.79 and weighted F1 scores ranged from 0.821 to 0.857. Similarly, on the CAP dataset, the initial accuracy was 80.6%, accompanied by a Cohen's kappa of 0.73 and weighted F1 score was 0.805. Following the application of the threshold, the accuracy increased to a range of 81.4-84.3%, Cohen's kappa values ranged from 0.74 to 0.79 and weighted F1 scores ranged from 0.813 to 0.842. Additionally, by interpreting the model's predictions, we obtained results indicating a correlation between the peak of the PPG signal and sleep stage classification. CONCLUSION: InsightSleepNet is a 4-class sleep staging model that utilizes continuous PPG data, serves the purpose of continuous sleep monitoring with wearable devices. Beyond its primary function, it might facilitate in-depth sleep analysis by medical professionals and empower them with interpretability for intervention-based predictions. This capability can also support well-informed clinical decision-making, providing valuable insights and serving as a reliable second opinion in medical settings.

SelANet: decision-assisting selective sleep apnea detection based on confidence score.

BACKGROUND: One of the most common sleep disorders is sleep apnea syndrome. To diagnose sleep apnea syndrome, polysomnography is typically used, but it has limitations in terms of labor, cost, and time. Therefore, studies have been conducted to develop automated detection algorithms using limited biological signals that can be more easily diagnosed. However, the lack of information from limited signals can result in uncertainty from artificial intelligence judgments. Therefore, we performed selective prediction by using estimated respiratory signals from electrocardiogram and oxygen saturation signals based on confidence scores to classify only those sleep apnea occurrence samples with high confidence. In addition, for samples with high uncertainty, this algorithm rejected them, providing a second opinion to the clinician. METHOD: Our developed model utilized polysomnography data from 994 subjects obtained from Massachusetts General Hospital. We performed feature extraction from the latent vector using the autoencoder. Then, one dimensional convolutional neural network-long short-term memory (1D CNN-LSTM) was designed and trained to measure confidence scores for input, with an additional selection function. We set a confidence score threshold called the target coverage and performed optimization only on samples with confidence scores higher than the target coverage. As a result, we demonstrated that the empirical coverage trained in the model converged to the target coverage. RESULT: To confirm whether the model has been optimized according to the objectives, the coverage violation was used to measure the difference between the target coverage and the empirical coverage. As a result, the value of coverage violation was found to be an average of 0.067. Based on the model, we evaluated the classification performance of sleep apnea and confirmed that it achieved 90.26% accuracy, 91.29% sensitivity, and 89.21% specificity. This represents an improvement of approximately 7.03% in all metrics compared to the performance achieved without using a selective prediction. CONCLUSION: This algorithm based on selective prediction utilizes confidence measurement method to minimize the problem caused by limited biological information. Based on this approach, this algorithm is applicable to wearable devices despite low signal quality and can be used as a simple detection method that determine the need for polysomnography or complement it.

Selective Prediction With Long Short-term Memory Using Unit-Wise Batch Standardization for Time Series Health Data Sets: Algorithm Development and Validation.

BACKGROUND: In any health care system, both the classification of data and the confidence level of such classifications are important. Therefore, a selective prediction model is required to classify time series health data according to confidence levels of prediction. OBJECTIVE: This study aims to develop a method using long short-term memory (LSTM) models with a reject option for time series health data classification. METHODS: An existing selective prediction method was adopted to implement an option for rejecting a classification output in LSTM models. However, a conventional selection function approach to LSTM does not achieve acceptable performance during learning stages. To tackle this problem, we proposed a unit-wise batch standardization that attempts to normalize each hidden unit in LSTM to apply the structural characteristics of LSTM models that concern the selection function. RESULTS: The ability of our method to approximate the target confidence level was compared by coverage violations for 2 time series of health data sets consisting of human activity and arrhythmia. For both data sets, our approach yielded lower average coverage violations (0.98% and 1.79% for each data set) than those of the conventional approach. In addition, the classification performance when using the reject option was compared with that of other normalization methods. Our method demonstrated superior performance for selective risk (12.63% and 17.82% for each data set), false-positive rates (2.09% and 5.8% for each data set), and false-negative rates (10.58% and 17.24% for each data set). CONCLUSIONS: Our normalization approach can help make selective predictions for time series health data. We expect this technique to enhance the confidence of users in classification systems and improve collaborative efforts between humans and artificial intelligence in the medical field through the use of classification that considers confidence.