Deep learning have led to impressive performances in a wide range of problems, including those within healthcare (Miotto et al., 2018) and in the processing of physiological signals (Faust et al., 2018) contexts. Often blamed for being essentially a “black box” methodology, substantial effort is now being made to provide explainable deep-learning models (Adadi and Berrada, 2018). In this Research Topic collection, the study by Pini et al. used the cardio-respiratory signals derived from a wearable chest belt to train a deep learning model with the ability to automatically detect sleep stages. A particular strength of the study is that the investigators included more than 1,000 subjects and evaluated a combination of temporal convolutional (Lea et al., 2017) and inception residual networks (Szegedy et al., 2017) to detect two (wake/sleep), three (wake/NREM/REM), or four (wake/light/deep/REM) sleep stages. Such, sophisticated methodologies clearly open the way for expanded generalizability studies that reconcile different sensors and signals to enable extraction of valid sleep staging irrespective of the specific physiological signal or signals being acquired by the wearable device.

A second study by Serrano Alarcón et al. focused on detecting apneic events during sleep by means of a convolutional neural network (CNN) architecture. In this instance, they used overnight blood oxygen saturation, heart rate, and thoracic and abdominal respiration from more than 4,000 adult subjects from Sleep Health Heart Study (SHHS) and Multi-Ethnic Study of Atherosclerosis (MESA) databases. In their study, the authors also applied explainable artificial intelligence (XAI) techniques such as Gradient-weighted Class Activation Mapping (Grad-CAM) (Selvaraju et al., 2017), which allowed them to visually identify those regions of the signals in which the CNN was focusing to derive its accurate predictions.

Other issues that remain persistently under exploration in sleep research are insomnia and depression. In this Research Topic, both are addressed at the same time by Xu et al.. The authors presented a very interesting study with 44 patients and 43 healthy controls to pave the way toward finding the mechanisms driving comorbidity of chronic insomnia and major depressive disorders. They showed new connections and associations between cerebral blood flow, brain function, and sleep and emotion regulation abnormalities, which may be behind of the pathogenesis of comorbidity in these diseases.

A new tool in the form of a simple yes/no questionnaire has been recently proposed for sleep apnea screening in adults (Duarte et al., 2020). The GOAL questionnaire (the initials standing for Gender, Obesity, Age, and Loud snoring) showed interesting screening performance when first proposed using a large Brazilian population (N = 7,377). In this Research Topic, a first study by Zheng et al. validated the GOAL instrument in a relatively large Asian population from China (N = 2,171). The authors reported similar results to those of the original study, as well as comparable screening ability than other well-known questionnaires, thus evidencing its robustness. Furthermore, a second study by Zhao et al. assessed the screening ability of GOAL questionnaire when combined with neck circumference or neck-to-height ratio by the use of logistic regression models. In their study, the authors enrolled 288 subjects and showed improved performance when compared with the original GOAL instrument alone.