BREATHMONITOR: SLEEP APNEA MOBILE DETECTOR Anatolii Petrenko System Design Department, National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Kyiv, Ukraine, petrenko@cad.kiev.ua Abstract —This paper describes the on-line AI system for diagnosis and monitoring sleep apnea at home, based on the processing of human respiratory signals from an accelerometer and pressure transducer composition by using Deep machine learning and alternative louder analytics Keywords— monitoring, respiratory diseases, Deep learning, accelerometer, sleep apnea, CNN 1. INTRODUCTION The need for diagnosis and observation of different respiratory diseases or their exacerbations has contributed to the development of different direct and indirect methods of respiration measuring: the use of a spirometer in which the patient should breathe; determination of oxygen concentration in blood, which is impossible in real time; analysis of the tracheal sounds, which does not really fit the patient's calm state, since the sound changes very little. At the same time, the recent development of IoT technologies and Deep learning using a convolutional neural network (CNN) have created an opportunity to combine them and build a system of continuous breathing monitoring. Such a system can be used to detect or predict and prevent exacerbations of dangerous conditions in various daily human activities (such as walking, sleeping, and other physical activity). In particular, sleep apnea is a common disease that affects both children and adults. It is characterized by periods of cessation of breathing (apnea) and periods of decline in respiration (hypopnea). Both types of events have similar pathophysiology and are generally considered equal in their impact on patients. The most common form of sleep apnea, called obstructive sleep apnea, is due to partial or complete collapse of the upper respiratory tract. Obstructive apnea is caused by mechanical stresses on the throat, central sleep apnea is the inability of the brain to send a signal to the diaphragm. There are several methods of quantifying the severity of respiratory distress, such as measuring the amount of sleep apnea and hypopnea per hour (ie, sleep apnea index AHI), the severity of oxygen starvation during sleep (oximetry, SpO2), or the degree of daytime sleepiness [1]). When AHI > = 5% then 24% of men and 9% of women aged 30 to 60 years are suspected in sleep apnea. Diagnosis of sleep apnea can use many signals from different sensors (polysomnography, polysomnography) when during night examination in the clinic data on respiratory flow, respiratory motion, SpO2, posture, electroencephalography (EEG), electromyography (EMG), electrocullography (EGG), and electrocardiography (ECG) are measured. Since such a procedure is very expensive (average device cost $ 2,625) and is not possible at the home, only one or two signals (ECG, SpO2 oximetry, sound snoring spectrum, etc.) are used to perform "portable" sleep apnea diagnosis in the home setting. Of course, home diagnosis is inferior to laboratory accuracy in diagnosing the disease (80-84% vs. 90-94%) [2]. There is now a great demand for real-time monitoring of the state of the patient's respiratory system in the home environment. For patients, such systems allow home-based measurement of the disease, with the physician, relatives (and / or ambulance) alerted automatically if the patient's vital signs are close to a dangerous limit. For the doctor it is possible to remotely monitor the patient's condition, promptly change the plan of his treatment, maintain contact with the patient, as well as the opportunity to consult with colleagues and specialists in the mode of television sessions with confidential transfer of patient data. In May 2018, an on-line system based on the use of a single-channel ECG signal, which can operate for 46 hours with an accuracy of 88%, was proposed [3]. It identifies the RR intervals and RS amplitudes of the