Kim Kun Ho, Lee Jeong Oen, Du Juan, Sretavan David, Choo Hyuck
Department of Computer Science, California Institute of Technology, Pasadena, CA 91125 USA.
Department of Electrical Engineering and the Department of Medical Engineering, California Institute of Technology, Pasadena, CA 91125 USA.
IEEE Sens J. 2017;17(22):7394-7404. doi: 10.1109/JSEN.2017.2760140. Epub 2017 Oct 5.
Optimized glaucoma therapy requires frequent monitoring and timely lowering of elevated intraocular pressure (IOP). A recently developed microscale IOP-monitoring implant, when illuminated with broadband light, reflects a pressure-dependent optical spectrum that is captured and converted to measure IOP. However, its accuracy is limited by background noise and the difficulty of modeling non-linear shifts of the spectra with respect to pressure changes. Using an end-to-end calibration system to train an artificial neural network (ANN) for signal demodulation we improved the speed and accuracy of pressure measurements obtained with an optically probed IOP-monitoring implant and make it suitable for real-time in vivo IOP monitoring. The ANN converts captured optical spectra into corresponding IOP levels. We achieved an IOP-measurement accuracy of ±0.1 mmHg at a measurement rate of 100 Hz, which represents a ten-fold improvement from previously reported values. This technique allowed real-time tracking of artificially induced sub-1 s transient IOP elevations and minor fluctuations induced by the respiratory motion of the rabbits during in vivo monitoring. All in vivo sensor readings paralleled those obtained concurrently using a commercial tonometer and showed consistency within ±2 mmHg. Real-time processing is highly useful for IOP monitoring in clinical settings and home environments and improves the overall practicality of the optical IOP-monitoring approach.
优化青光眼治疗需要频繁监测并及时降低眼内压（IOP）升高的情况。最近开发的一种微型眼压监测植入物，在宽带光照射下，会反射出与压力相关的光谱，该光谱被捕获并转换以测量眼压。然而，其准确性受到背景噪声以及相对于压力变化对光谱非线性偏移进行建模的难度的限制。通过使用端到端校准系统来训练用于信号解调的人工神经网络（ANN），我们提高了通过光学探测的眼压监测植入物获得的压力测量的速度和准确性，并使其适用于实时体内眼压监测。该人工神经网络将捕获的光谱转换为相应的眼压水平。我们在100Hz的测量速率下实现了±0.1mmHg的眼压测量精度，这比先前报道的值提高了十倍。该技术允许在体内监测期间实时跟踪人为诱导的亚1秒短暂眼压升高以及兔子呼吸运动引起的微小波动。所有体内传感器读数与同时使用商用眼压计获得的读数平行，并且在±2mmHg范围内显示出一致性。实时处理对于临床环境和家庭环境中的眼压监测非常有用，并提高了光学眼压监测方法的整体实用性。
