Department of Artificial Intelligence in Biomedical EngineeringFriedrich-Alexander-Universität Erlangen-Nürnberg 91052 Erlangen Germany.
Department of Gynecology and ObstetricsUniversitätsklinikum Erlangen 91054 Erlangen Germany.
IEEE J Transl Eng Health Med. 2022 May 30;10:2800109. doi: 10.1109/JTEHM.2022.3179147. eCollection 2022.
Clinical urine tests are a key component of prenatal care. As of now, urine test strips are evaluated through a time consuming, often error-prone and operator-dependent visual color comparison of test strips and reference cards by medical staff.
This work presents an automated pipeline for urinalysis with urine test strips using smartphone camera images in home environments, combining several image processing and color combination techniques. Our approach is applicable to off-the-shelf test strips in home conditions with no additional hardware required. For development and evaluation of our pipeline we collected image data from two sources: i) A user study (26 participants, 150 images) and ii) a lab study (135 images).
We trained a region-based convolutional neural network that is able to detect the urine test strip location and orientation in images with a wide variety of light conditions, backgrounds and perspectives with an accuracy of 85.5%. The reference card can be robustly detected through a feature matching approach in 98.6% of the images. Color comparison by Hue channel (0.81 F1-Score), Matching factor (0.80 F1-Score) and Euclidean distance (0.70 F1-Score) were evaluated to determine the urinalysis results.
We show that an automated smartphone-based colorimetric analysis of urine test strips in a home environment is feasible. It facilitates examinations and provides the possibility to shift care into an at-home environment.
The findings demonstrate that routine urine examinations can be transferred into the home environment using a smartphone. Simultaneously, human error is avoided, accuracy is increased and medical staff is relieved.
临床尿液检测是产前护理的一个关键组成部分。到目前为止，尿液检测试纸是由医务人员通过耗时、易错且依赖操作者的视觉比色法对检测试纸和参考卡进行评估的。
这项工作提出了一种在家庭环境中使用智能手机摄像头图像对尿液检测试纸进行自动分析的流水线，结合了多种图像处理和颜色组合技术。我们的方法适用于家庭环境中的现成检测试纸，不需要额外的硬件。为了开发和评估我们的流水线，我们从两个来源收集了图像数据：i）一项用户研究（26 名参与者，150 张图像）和 ii）一项实验室研究（135 张图像）。
我们训练了一个基于区域的卷积神经网络，该网络能够在各种光照条件、背景和视角下准确检测图像中的尿液检测试纸位置和方向，准确率为 85.5%。参考卡可以通过特征匹配方法在 98.6%的图像中可靠地检测到。通过 Hue 通道（0.81 F1 分数）、匹配因子（0.80 F1 分数）和欧几里得距离（0.70 F1 分数）进行颜色比较，以确定尿液分析结果。
我们表明，在家庭环境中使用智能手机对尿液检测试纸进行自动化比色分析是可行的。它方便了检查，并提供了将护理转移到家庭环境中的可能性。
研究结果表明，常规尿液检查可以使用智能手机转移到家庭环境中。同时，避免了人为错误，提高了准确性，减轻了医务人员的负担。
