Joint Classification and Prediction CNN Framework for Automatic Sleep Stage Classification (1805.06546v3)

Abstract: Correctly identifying sleep stages is important in diagnosing and treating sleep disorders. This work proposes a joint classification-and-prediction framework based on CNNs for automatic sleep staging, and, subsequently, introduces a simple yet efficient CNN architecture to power the framework. Given a single input epoch, the novel framework jointly determines its label (classification) and its neighboring epochs' labels (prediction) in the contextual output. While the proposed framework is orthogonal to the widely adopted classification schemes, which take one or multiple epochs as contextual inputs and produce a single classification decision on the target epoch, we demonstrate its advantages in several ways. First, it leverages the dependency among consecutive sleep epochs while surpassing the problems experienced with the common classification schemes. Second, even with a single model, the framework has the capacity to produce multiple decisions, which are essential in obtaining a good performance as in ensemble-of-models methods, with very little induced computational overhead. Probabilistic aggregation techniques are then proposed to leverage the availability of multiple decisions. We conducted experiments on two public datasets: Sleep-EDF Expanded with 20 subjects, and Montreal Archive of Sleep Studies dataset with 200 subjects. The proposed framework yields an overall classification accuracy of 82.3% and 83.6%, respectively. We also show that the proposed framework not only is superior to the baselines based on the common classification schemes but also outperforms existing deep-learning approaches. To our knowledge, this is the first work going beyond the standard single-output classification to consider multitask neural networks for automatic sleep staging. This framework provides avenues for further studies of different neural-network architectures for automatic sleep staging.

• The paper introduces a joint classification and prediction CNN that simultaneously classifies an input epoch and predicts adjacent sleep stages.
• It employs a simplified CNN architecture with a multi-task softmax layer and time-frequency image representation for robust feature learning.
• Performance on Sleep-EDF and MASS datasets achieved accuracies of 82.3% and 83.6%, respectively, outperforming traditional methods.

Overview of Joint Classification and Prediction CNN Framework for Automatic Sleep Stage Classification

This paper presents a novel approach for automatic sleep stage classification leveraging a joint classification and prediction framework using convolutional neural networks (CNNs). The paper acknowledges the critical role of accurately identifying sleep stages for diagnosing and treating sleep disorders. It introduces a CNN framework designed for simultaneous classification of an input epoch and prediction of the labels of neighboring epochs, thus exhibiting significant advances over conventional methods relying solely on contextual input.

Key Contributions and Findings

1. Joint Classification and Prediction: This framework redefines the sleep staging problem as a joint classification and prediction task, moving beyond the traditional single-output models. This approach capitalizes on the dependency between consecutive sleep stages and efficiently handles the alternating sleep patterns found in polysomnography (PSG) data.
2. CNN Architecture: The proposed framework is powered by a simplified yet effective CNN architecture, distinctively designed to manage the joint task with a multi-task softmax layer. The CNN uses a time-frequency image representation, which enables robust feature learning without heavily relying on hand-crafted data features.
3. Probabilistic Aggregation Methods: The paper introduces probabilistic aggregation methods, namely additive and multiplicative voting, which synthesize multiple decisions yielded by the model to achieve improved classification accuracy.
4. Performance Evaluation: The methodology is rigorously validated on two public datasets, Sleep-EDF and the Montreal Archive of Sleep Studies (MASS), demonstrating classification accuracies of 82.3% and 83.6%, respectively. The results indicate superior performance when contrasted with existing models, marking an enhancement over further elaborated deep-learning driven modalities.

Implications and Future Directions

The proposed joint classification and prediction framework holds promising practical implications for enhancing automatic sleep stage classification. The ability to accurately predict sleep stages with a high degree of confidence using a single model can optimize the diagnostic processes in wearable and telemedicine applications, promoting more accessible long-term sleep monitoring.

Theoretically, this framework opens a pathway for exploring various neural network architectures within sleep stage classification. Future developments could involve integrating recurrent neural network structures to model long-term dependencies more explicitly or deploying the framework in other temporally dependent classification tasks outside of sleep analysis.

By addressing both classification and prediction tasks within a unified framework, the paper paves the way for a more comprehensive modeling approach, ensuring more accurate sleep staging while reducing computational complexity. As the field advances, the potential integration of multi-modal physiological signals could further enhance the robustness and applicability of this framework.

In conclusion, the joint classification and prediction CNN framework stands as a significant contribution to automatic sleep stage classification, offering both improved performance and novel insights into the application of deep learning in temporal classification tasks.