| Currently the only method for quantitative sleep analysis is a procedure known as polysomnography, where the subject spends a night in a clinic with numerous sensing modalities attached. This evaluative procedure has shortcomings in that it disturbs the sleep of the subject and that the interpretation of the procedure is time intensive and somewhat subjective. The focus of this research is on the mitigation of these problems through the development and validation of a bandage, called the SleepiBand, with near-infrared spectroscopy capabilities and dimensions of 2.8 cm x 1.7 cm x 0.6 cm. Using this optical sensor, measurements of heart rate, respiratory rate, cerebral oxygenation, and relative concentration changes of doxyhemoglobin and oxyhemoglobin were obtained and streamed to a peripheral device using BlueTooth low energy. These measurements were then analyzed in the frequency domain and separated into 100 different frequency buckets having a width of 1.25 Hz over the range of 0 Hz to 125 Hz. These discrete frequency based features were the input to a k nearest neighbor algorithm that optimized its number of neighbors, weight of neighbors, distance metric, and other attributes based on the training set of data. Four patients were recorded overnight with traditional polysomnography and the SleepiBand and the resulting data was concatenated into one data set. Upon using 15% of randomly selected points from the data set, the algorithm could reconstruct the diagnostic output with 71.8% accuracy. Combining additional sensing modalities, such as EEG, improved the total accuracy. |
