Conv-TabNet: an efficient adaptive color correction network for smartphone-based urine component analysis.

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Image Processing

Background:

• Smartphone cameras offer potential for ubiquitous, quantitative urine parameter detection.
• Variations in camera sensors and ambient lighting conditions introduce color inaccuracies in urine test strip images.
• Accurate color measurement is crucial for reliable interpretation of urine test strip results.

Purpose of the Study:

• To develop and validate a color correction method for smartphone-acquired urine test strip images.
• To address the challenge of variable color capture due to different cameras and environments.
• To enable accurate, on-demand quantitative analysis of urine parameters using mobile devices.

Main Methods:

• A color correction model was developed utilizing urine test strip color information, ambient light conditions, and camera parameters.
• Conv-TabNet architecture was employed to focus on individual feature parameters for precise color correction of test strip color blocks.
• Experiments were conducted using four different mobile phones under eight distinct light sources.

Main Results:

• The proposed color correction method achieved a low mean absolute error of 2.8±1.8.
• The CIEDE2000 color difference was measured at 1.5±1.5, indicating high accuracy.
• Visual evaluation confirmed that the corrected colors closely matched standard reference colors.

Conclusions:

• The developed color correction technique significantly enhances the reliability of smartphone-based urine test strip analysis.
• This method overcomes environmental and camera-specific color variations, enabling consistent quantitative detection.
• The technology facilitates convenient and accurate urine parameter measurement anytime and anywhere.