Karlen Walter, Lim Joanne, Ansermino J Mark, Dumont Guy A, Scheffer Cornie
Annu Int Conf IEEE Eng Med Biol Soc. 2013;2013:7480-3. doi: 10.1109/EMBC.2013.6611288.
In mobile health applications, non-expert users often perform the required medical measurements without supervision. Therefore, it is important that the mobile device guides them through the correct measurement process and automatically detects potential errors that could impact the readings. Camera oximetry provides a non-invasive measurement of heart rate and blood oxygen saturation using the camera of a mobile phone. We describe a novel method to automatically detect the correct finger placement on the camera lens for camera oximetry. Incorrect placement can cause optical shunt and if ignored, lead to low quality oximetry readings. The presented algorithm uses the spectral properties of the pixels to discriminate between correct and incorrect placements. Experimental results demonstrate high mean accuracy (99.06%), sensitivity (98.06%) and specificity (99.30%) with low variability. By sub-sampling pixels, the computational cost of classifying a frame has been reduced by more than three orders of magnitude. The algorithm has been integrated in a newly developed application called OxiCam where it provides real-time user feedback.
在移动健康应用中，非专业用户常常在无人监督的情况下进行所需的医学测量。因此，移动设备引导他们完成正确的测量过程并自动检测可能影响读数的潜在误差非常重要。摄像头血氧测定法利用手机摄像头对心率和血氧饱和度进行非侵入式测量。我们描述了一种新颖的方法，用于自动检测摄像头血氧测定法中手指在摄像头镜头上的正确放置位置。放置不正确会导致光学分流，如果被忽视，会导致低质量的血氧测定读数。所提出的算法利用像素的光谱特性来区分正确和不正确的放置位置。实验结果表明，该算法具有较高的平均准确率（99.06%）、灵敏度（98.06%）和特异性（99.30%），且变异性较低。通过对像素进行子采样，对一帧进行分类的计算成本降低了三个多数量级。该算法已集成到一个名为OxiCam的新开发应用程序中，在那里它提供实时用户反馈。
