An Interpretable and Efficient Sleep Staging Algorithm: DetectsleepNet

Sleep quality directly impacts human health and quality of life, so accurate sleep staging is essential for assessing sleep quality. However, most traditional methods are inefficient and time-consuming due to segmenting different sleep cycles by manual labeling. In contrast, automated sleep staging technology not only directly assesses sleep quality but also helps sleep specialists analyze sleep status, significantly improving efficiency and reducing the cost of sleep monitoring, especially for continuous sleep monitoring. Most of the existing models, however, are deficient in computational efficiency, lightweight design, and model interpretability. In this paper, we propose a neural network architecture based on the prior knowledge of sleep experts. Specifically, 1) Propose an end-to-end model named DetectsleepNet that uses single-channel EEG signals without additional data processing, which has achieved an impressive 80.9% accuracy on the SHHS dataset and an outstanding 88.0% accuracy on the Physio2018 dataset. 2) Constructure an efficient lightweight sleep staging model named DetectsleepNet-tiny based on DetectsleepNet, which has just 6% of the parameter numbers of existing models, but its accuracy exceeds 99% of state-of-the-art models, 3) Introducing a specific inference header to assess the attention given to a specific EEG segment in each sleep frame, enhancing the transparency in the decisions of models. Our model comprises fewer parameters compared to existing ones and ulteriorly explores the interpretability of the model to facilitate its application in healthcare. The code is available at https://github.com/komdec/DetectSleepNet.git.

SI-FID: Only One Objective Indicator for Evaluating Stitched Images

Image quality evaluation accurately is vital in developing image stitching algorithms as it directly reflects the algorithms progress. However, commonly used objective indicators always produce inconsistent and even conflicting results with subjective indicators. To enhance the consistency between objective and subjective evaluations, this paper introduces a novel indicator the Frechet Distance for Stitched Images (SI-FID). To be specific, our training network employs the contrastive learning architecture overall. We employ data augmentation approaches that serve as noise to distort images in the training set. Both the initial and distorted training sets are then input into the pre-training model for fine-tuning. We then evaluate the altered FID after introducing interference to the test set and examine if the noise can improve the consistency between objective and subjective evaluation results. The rank correlation coefficient is utilized to measure the consistency. SI-FID is an altered FID that generates the highest rank correlation coefficient under the effect of a certain noise. The experimental results demonstrate that the rank correlation coefficient obtained by SI-FID is at least 25% higher than other objective indicators, which means achieving evaluation results closer to human subjective evaluation.