Harshan Ravi

Graduation Semester and Year




Document Type


Degree Name

Master of Science in Biomedical Engineering



First Advisor

Khosrow Behbehani


Sleep apnea is a sleep disordered breathing resulting from limitation or cessation to breathing for 10 or more seconds. The prevalence of sleep apnea has increased exponentially in the past decade. The increase in epidemic in sleep apnea is associated with increase in the cases of obesity. It is estimated that 12-18 million of American adults suffer from sleep apnea, which is a sizable sector of the adult population. Sleep apnea is risk factor for hypertension, type II diabetes, and congestive heart failure. Sleep apnea may go undiagnosed for a long period of time after its onset due to complexity of diagnosing it. Nocturnal polysomnography is the standard method for diagnosing sleep apnea. It is often inaccessible and costly, hence making sleep apnea an under diagnosed disease. Further, widespread screen of the vulnerable sector of the population (ages 35 and above) is currently not feasible. To overcome these limitations many physiological markers are investigated as alternatives means of detecting sleep apnea. A number of studies during the past decade have investigated the possibility of detecting sleep apnea using features of the electrocardiogram (ECG).In this study, a support vector machine (SVM) based classifier was developed to detect obstructive sleep apnea (OSA) and normal breathing using features extracted from nocturnal ECG. NPSG was performed on 16 normal patients and 14 OSA patients. This approach combines both RPE and R-R interval to form a cluster. An optimum centroid is extracted from the cluster, and is used as an input to the SVM. The performance of the proposed algorithm in detecting respiratory event was tested by determining its ability to detect normal breathing, and OSA events in 15 minutes data epochs obtained from volunteer normal 16 subjects and 14 apnea patients. The SVM algorithm was designed and optimized using two heuristic and three numerical optimization techniques. For Manual optimization, a highest learning performance of, accuracy of 91.16%, sensitivity of 95.20% and specificity of 86.20% is achieved for training set and a highest testing performance of, accuracy of 75.98%, sensitivity of 81.20 %, and specificity of 69.87% is achieved for testing set. The computerized optimization resulted in slightly higher performance than the Manual optimization. The highest learning performance achieved for training set is, accuracy of 92.78%, sensitivity of 96.33% and specificity of 88.43% and a highest testing performance of, accuracy of 76.66%, sensitivity of 81.84 %, and specificity of 70.41% is achieved for testing set. The detection rates achieved using SVM is comparable to the results achieved with previous study using other form of classifiers [10].


Biomedical Engineering and Bioengineering | Engineering


Degree granted by The University of Texas at Arlington