OBJECTIVE: We present a non-invasive wearable device for fertility monitoring and propose an effective and flexible statistical learning algorithm to detect and predict ovulation using data captured by this device. METHODS: The system consists of an earpiece, which measures the ear canal temperature every 5 min during night sleep hours, and a base station that transmits data to a smartphone application for analysis. We establish a data-cleaning protocol for data preprocessing and then fit a Hidden Markov Model (HMM) with two hidden states of high and low temperature to identify the more probable state of each time point via the predicted probabilities. Finally, a post-processing procedure is developed to incorporate biorhythm information to form a time-course biphasic profile for each subject. RESULTS: The performance of the proposed algorithms applied to data collected by the device are compared with traditional methods in terms of match rate with self-reported ovulation days confirmed with an ovulation test kit. Empirical study results from a group of 34 users yielded significant improvements over the traditional methods in terms of detection accuracy (with sensitivity 92.31%) and prediction power (23.07-31.55% higher). CONCLUSION: We demonstrated the feasibility for reliable ovulation detection and prediction with high-frequency temperature data collected by a non-invasive wearable device. SIGNIFICANCE: Traditional fertility monitoring methods are often either inaccurate or inconvenient. The wearable device and learning algorithm presented in this paper provide a user friendly and reliable platform for tracking ovulation, which may have a broad impact on both fertility research and real-world family planning.
OBJECTIVE: We present a non-invasive wearable device for fertility monitoring and propose an effective and flexible statistical learning algorithm to detect and predict ovulation using data captured by this device. METHODS: The system consists of an earpiece, which measures the ear canal temperature every 5 min during night sleep hours, and a base station that transmits data to a smartphone application for analysis. We establish a data-cleaning protocol for data preprocessing and then fit a Hidden Markov Model (HMM) with two hidden states of high and low temperature to identify the more probable state of each time point via the predicted probabilities. Finally, a post-processing procedure is developed to incorporate biorhythm information to form a time-course biphasic profile for each subject. RESULTS: The performance of the proposed algorithms applied to data collected by the device are compared with traditional methods in terms of match rate with self-reported ovulation days confirmed with an ovulation test kit. Empirical study results from a group of 34 users yielded significant improvements over the traditional methods in terms of detection accuracy (with sensitivity 92.31%) and prediction power (23.07-31.55% higher). CONCLUSION: We demonstrated the feasibility for reliable ovulation detection and prediction with high-frequency temperature data collected by a non-invasive wearable device. SIGNIFICANCE: Traditional fertility monitoring methods are often either inaccurate or inconvenient. The wearable device and learning algorithm presented in this paper provide a user friendly and reliable platform for tracking ovulation, which may have a broad impact on both fertility research and real-world family planning.