Abstract: TH-PO770
Patient-Friendly Kidney Function Monitoring
Session Information
- Bioengineering
October 25, 2018 | Location: Exhibit Hall, San Diego Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Bioengineering
- 300 Bioengineering
Author
- Elsayed, Ragwa, San Jose State University, San Francisco, California, United States
Background
Chronic Kidney Disease was ranked the 13th leading cause of Death in 2013. Measuring the creatinine level in blood became an important biomarker to evaluate the kidney performance by filtering the wastes. The objective is to develop an easy low-cost monitoring method of the level of creatinine in blood without visiting a medical facility. It is an alert for individuals to check their kidney functions and communicate with a nephrologist when they suspect a kidney failure in the early stages.
Methods
The project is based on using paper microfluidics technology. A paper strip with alkaline picrate solution produces a colorimetric response upon applying a creatinine sample. 65 different creatinine concentrations were tested to detect the colorimetry change and images were captured using a smart-phone camera inside a light-box. The color intensity feature was extracted using different technqiues. The features extracted were used to train the several machine learning models for testing.
Results
The results show that the best machine learning models that gives the best accurate results and the lowest root mean square error is Nearest Neighbor regression model using the histogram of colors extraction technique.
Conclusion
We proved the feasibility and the reliability of our proposed technique. We are providing individuals with an affordable, reliable, and remote technology to allow rapid and frequent testing of kidney health. Individuals will be able to monitor their kidney functions without visiting a medical facility.
So, we can imagine how this method would help patients to get early treatment, adjust medication dose, or monitor the performance of the transplanted kidney. Our approach is using a smart-phone for detection due to the accessibility of a smart-phone to any individual.
Root-mean square error for creatinine concentration prediction 0-4 (mg/dl)
| Linear Regression | Logistic Regression | Nearest Neighbor Regression | Support Vector Machine | |
| REG pixels | 0.51312543 | 0.8106177963 | 0.3936857306 | 0.4423723639 |
| Histogram of Gradients | 0.766976729 | 0.9150882978 | 0.7523593551 | 0.7856010536 |
| Histogram of Colors | 0.2710061214 | 0.380095781 | 0.2365255844 | 0.265266147 |
Root-mean square error for creatinine prediction concentration 0-4 (mg/dl).