Implementing machine learning algorithms to detect and screen obstructive sleep apnoea episodes using a headband

点击翻译按钮获取中文摘要

Can machine learning algorithms be used to detect obstructive sleep apnoea with high accuracy and precision? How do machine learning algorithms compare to the traditional method (polysomnography) of diagnosing obstructive sleep apnoea? Implement machine learning algorithms to detect obstructive sleep apnoea episodes in patients who are suspected of obstructive sleep apnoea. It is costly and labour-intensive for a clinician to manually go through a patient's sleep data to detect obstructive sleep apnoea. Especially because there is more than one aspect that needs to be looked at. For example, sound data (snoring) is not enough on its own to be able to detect obstructive sleep apnoea. Also, the frequency of certain events is important. Therefore, the clinician needs to go through a lot of data to be able to diagnose the patient with obstructive sleep apnoea. It is likely that the human error will result from this laborious activity. Therefore, it can be difficult to make decisions and diagnose the patients if important events have been missed during the analysis. Sensor noise can also mask certain events and signals making it hard to make decisions. The general aim is to reduce the burden on clinical staff that manually sleep score patients. So, learning algorithms will be used to automatically highlight the obstructive sleep apnoea episodes. It is a mass screening tool rather than a diagnostic test. A sleep score will be generated, so a clinician can decide whether a PSG is required. The machine learning tool will use data for multiple sensors. This is to prevent the collected data being redundant if one of the sensors fail or a connection is lost etc. The algorithm will take into account multiple nights of data which will have an advantage over the expensive PSG. The band that is to be used includes multiple sensors that are essential for the detection of obstructive sleep apnoea. The band can be taken home, and it is not necessary for the patient to be at a clinic or a hospital. Therefore, the extracted data is more reliable because patients being at hospital do not feel comfortable and they might not sleep as they would at home. Different machine learning algorithms have been researched and reviewed as to which is more accurate and suitable for detection of obstructive sleep apnoea. However, there has been little implementation into an actual device.