A Machine Learning-based pH Color Recognition for Monitoring Chronic Kidney Disease

Nearly 15% of the adult American population are affected by Chronic Kidney Disease (CKD) according to the Center for Disease Control and Prevention (CDC). According to the National Institutes of Health (NIH), there are five stages of CKD, with each stage reflecting a different level of kidney function. In its early stages, CKD can often be asymptomatic. As a result, less than 10% of individuals are aware of having stage 1-3 CKD. Therefore, CKD is described as a "silent killer" for them. One common method of detecting and monitoring CKD is through blood or urine tests that measure biomarkers such as blood urea nitrogen (BUN), creatinine, or protein levels. However, these tests require patients to visit clinics for sample collection, and wait for results and feedback from their doctors, which causes much inconvenience and stress to the patient. A less burdensome alternative for diagnosing CKD is to measure BUN levels through a urine or saliva test using pH test strips. Although pH test strips are widely available, accurate determination of a patient’s pH level through analysis of strip images remains challenging. This empirical pilot study aims to explore the possibility and effectiveness of using supervised machine learning models, such as KNN, SVM, and Neural Network, to predict pH values by recognizing the RGB profile data of pH test strips. The model settings are thoroughly explained, and the accuracy of the machine learning-based pH color recognition demonstrates promise in detecting CKD. Future research may focus on optimizing the dataset and machine learning models.