Prediction of Apnoea and Non-apnoea Arousals from the Polysomnogram using a Neural Network Classifier Philip de Chazal, John Du, Nadi Sadr Charles Perkins Centre and School of Biomedical Engineering, The University of Sydney, Australia Abstract In this study we present a system for automated processing of signals from the polysomnogram (PSG) for the detection of apnoea and non-apnoea arousals. The PSG signals were divided into 15 second epochs and 59 time- and frequency-domain features were derived for each epoch. Features from adjacent 4 epochs were combined and processed with a bank of ten feed-forward neural networks each with a single hidden layer of 20 units. The system outputs a 200 Hz annotation signal containing probability estimates that each sample was associated with an apnoea or non-apnoea arousal, or no- arousal. Data from the Physionet Computing in Cardiology Challenge 2018 was used to develop and test the system. Performance of the system was assessed using three class and two class metrics. With the system classifying three classes, the volume under the receiver operator characteristic (ROC) surface was 0.74 with an optimal specificity of 0.67, a sensitivity of 0.77 for the apnoea arousals, and a sensitivity of 0.73 for the non- apnoea arousals. When the two arousal classes were combined into one arousal class, the area under the precision recall curve was 0.78, the area under the ROC curve was 0.93, with an optimal specificity and sensitivity of 0.85. 1. Introduction A wide range of negative health outcomes including neurocognitive disorders, mood and mental conditions, cardiovascular disease [1], hypertension and stroke [2] are associated with low quality sleep. Poor sleep is also an established contributor to workplace and road accidents [3]. While arousals are a normal feature of the sleep/wake cycle, an excessive number of arousals can lead to poor sleep quality [4-6]. Respiration interruptions during sleep are a common cause of arousals. These interruptions include obstructive apnoea and hypopnoea events, respiratory effort related arousals and other interruptions to breathing. Arousals can also be caused by snoring, muscle jerks, pain, and insomnia. The most common way to provide a detailed assessment of sleep is to record a polysomnogram (PSG) which provides range of signals from a sleeping patient [7]. Sleep technicians then manually assess the PSG. Part of their analysis includes scoring arousals which is a time- consuming manual task. Automated software that processes the PSG information and assists the technician in identifying arousals would clearly be of benefit and is the topic of this paper. The PhysioNet Computing in Cardiology Challenge 2018 [8] provided the framework for researchers to develop and test automatic algorithms for the detection of non-apnoea arousals from the PSG and we were one of the participating teams [9,10]. For the purposes of the Challenge apnoea-related arousals were ignored which limited the clinical application of the resulting algorithms. In this current study we address this application issue and extend the capability of our system to detect apnoea arousals. Our resulting system potentially has greater clinical application. 2. Input data The data used in this study dataset was provided by the 2018 Challenge organisers (https://physionet.org/ challenge) and is publicly available. It includes 994 overnight PSG study recordings and associated sleep, respiratory event and arousal annotations [8]. All recordings were acquired at the Massachusetts General Hospital (MGH) sleep laboratories. 2.1. Signals and expert scorings The signals include the electrocardiogram, electroencephalogram (EEG), chin electromyogram (EMG), electrooculogram (EOG), and pulse oximetry (SpO2). All signals were sampled at 200Hz. Standard sleep and respiratory events were scored by MGH staff. They also determined the time, duration and cause of all arousal events. Arousals were then grouped into apnoea related arousals and non-apnoea related arousals. Apnoea arousals included central, mixed and obstructive apnoeas and hypopnoeas. Non-apnoea related arousals included