Machine learning-assisted image label-free smartphone platform for rapid segmentation and robust multi-urinalysis

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Machine Learning Applications

Background:

• Early detection of chronic kidney disease (CKD) and urological disorders is crucial for effective management.
• Traditional urinalysis methods often require specialized equipment and trained personnel, limiting accessibility.
• Existing automated methods may lack robustness or generalizability in diverse settings.

Purpose of the Study:

• To develop an image-label-free, multi-dipstick identification method for early CKD and urological disorder detection.
• To create a cost-effective and portable urinalysis system utilizing smartphone technology.
• To enhance the accuracy and generalizability of machine learning models for automated urinalysis.

Main Methods:

• Utilized machine learning algorithms trained on human urine data to identify reaction pads on multi-dipsticks.
• Developed algorithms for primary colour extraction and urine colour correction on reaction pads.
• Employed random forest algorithms in the HSV colour space for optimal performance in smartphone-assisted urinalysis.

Main Results:

• Successfully identified reaction pads on 187 multi-dipsticks (11 pads each).
• Achieved performance rivaling professional medical equipment under regular indoor lighting.
• Demonstrated effective management of false positives and negatives with remarkable accuracy.

Conclusions:

• The developed smartphone-based urinalysis method offers a robust, cost-effective, and portable solution for early CKD and urological disorder detection.
• Novel urine colour correction and ISO parameter considerations significantly improve machine learning model performance.
• This technology holds significant potential for community screening and home monitoring, especially in low-resource settings.