Severity degree prediction of obstructive sleep apnea hypopnea syndrome using fuzzy expert system

OSAHS is an important health problem which results from repetitive obstructions in upper airways. In apnea, respiration completely arrests whereas airflow partly continues in hypopnea. Arrests or reductions in airflow lead to oxygen desaturation in blood, and this in turn, causes arousals. Insufficient sleep may lead to daytime sleepiness, fatigue and even traffic and occupational accidents. Also, it is detected that OSAHS is related to arterial hypertension, hypercoagulability, reduced cerebral perfusion, atherosclerosis, cardiac arrhythmia, coronary artery disease, congestive heart failure, ischemic stroke, axonal peripheral neuropathy and diabetes mellitus. Prevalence of the disease is 4% among middle-aged men, and 2% women in the same age category. It is predicted that these ratios increase for 65 age and older. OSAHS is diagnosed and graded by analyzing some physiological parameters such as brain and heart signals, oxygen saturation, airflow, respiratory movements which are obtained in sleep laboratories by means of PSG. Laboratory tests alone are not sufficient for diagnosis. Results must be interpreted by experienced staffs. However, manual scoring of the resulting recording entails too much effort and time to the medical specialists and as a consequence it implies a high economic cost. Also, in order to obtain accurate results, these minimal conditions have to be provided: Sound proof and climatized room that can be kept in complete darkness, a comfortable bed and the technical installations for recording the biological signals, avoiding the patient from the use of sleeping pills or narcotics several days before the study and from strenuous physical exercise on the day of the study, and not to drink coffee, black tea or alcoholic beverages. Insufficiency in one of these conditions may result in misleading results by reducing sleep quality. Moreover, OSAHS may be result from sleeping in supine position and it is shown that PSG may overestimate the severity of OSAHS in some patients with positional OSAHS. Another problem lies in the classification OSAHS patients according to their severity degree using crisp AHI intervals, because in contrast with one standard AHI intervals, there are 3 different hypopnea definitions. These differences in definition lead to misclassifications when the standard AHI intervals are not adjusted. Therefore, this indicates the need for more flexible AHI intervals rather than crisp ones. Expert systems are used for decision making in various fields such as agriculture, chemistry, space technology, geology and medicine. The basic idea behind ES is to transfer the vast body of task-specific knowledge from a human to a computer. This knowledge is then stored in the computer and users call upon the computer for specific advice as needed. The computer can make inferences and arrive at a specific conclusion. Then like a human consultant, it gives advices and explains, if necessary, the logic behind the advice. Zadeh's introduction the fuzzy sets enable to use both numerical and linguistic variables together, and provide expert systems with the ability of human-like reasoning. Therefore, expert systems can be solution for the problem uncertainty in medical data and its classification which arises from the complex nature of medicine. The aim of the study is to design a fuzzy expert system to predict the severity degree of OSAHS by using data which is predominantly collected through physical examination. Determining OSAHS severity is important because first, it is also an indicator for some of serious health problems such as heart diseases, hypertension and second, there is need for measures to evaluate treatment response. Pre-operation data of 24 patients who had robotic surgery for treatment of OSAHS are used. Data is divided into two: 14 of them for designing the fuzzy expert system and 10 patient data for testing the model. min SpO2, body mass index, Mallampati score, and neck circumference information are used as inputs of the system. 14 patients also had oxygen desaturation index (ODI) data and a second system which uses ODI instead of min SpO2 was designed. However, this system could not to be tested since the patients in test group did not have the ODI data. Both the inputs and output is fuzzified as the first step of the design. Then, fuzzy output is obtained via Mamdani inference method according to constructed rules which are based on domain expert views. Finally, the fuzzy output is defuzzified by centroid method in order to get crisp prediction for a given input. All these steps are performed using the FIS tool of Matlab version of R2013a. The output is severity prediction which is formed based on AASM 1999 AHI criteria. Then, this prediction compared with the actual AHI scores of the patients. Classification accuracy for the patients in the first group (design) is 0.86 and correlation between our prediction and AHI is 0.56. However, data of 4 patients seemed to be inconsistent with their AHI. This situation results from some other factors which affect OSAHS severity such as obstructions in soft palate, oropharynx and epiglottis. These patients are excluded since no data is available to evaluate these obstructions. After these exclusions, classification accuracy rises 100%, and correlation coefficient becomes 0.88. Second system could classify 12 patients accurately and correlation between its prediction results and AHI is 0.78. After same exclusions, accuracy becomes 100% and correlation reaches to 0.88. For the test result, classification accuracy is 90% and value of correlation coefficient is 0.53. Inconsistency problem here is also available for 1 patient. After leaving this patient out, accuracy and correlation coefficient become 100% and 0.78. The study shows that fuzzy logic can be successfully applied for OSAHS severity degree prediction problem and a possibility of simpler alternative to PSG and proposes fuzziness in standard AHI intervals as different point of view.