Selective Partitioned Regression for Accurate Kidney Health Monitoring.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Medical Diagnostics

Background:

• Rising incidence of advanced kidney disease necessitates early detection and monitoring.
• Current methods for kidney health assessment can be invasive or costly.
• Minimally invasive techniques are crucial for preventing severe kidney damage or failure.

Purpose of the Study:

• To develop and evaluate a cost-effective, machine learning-based system for kidney disease detection.
• To assess the efficacy of machine learning models in predicting creatinine concentration from colorimetric changes.
• To classify kidney disease severity (healthy, intermediate, critical) using smartphone technology.

Main Methods:

• Development of an iPhone application integrating a camera-based biosensor.
• Application of classical machine learning and deep learning techniques for creatinine prediction.
• Utilizing colorimetric changes on test strips to analyze creatinine levels.
• Evaluation of novel models, including selective partitioned regression (SPR), against state-of-the-art methods.
• Conducting ablation studies to optimize model performance.

Main Results:

• The selective partitioned regression (SPR) model demonstrated superior prediction performance compared to existing methods.
• SPR utilizes histogram of colors-based features and a gradient boosted trees estimator.
• The system accurately translates colorimetric reactions into kidney health predictions.
• Achieved better overall prediction performance through selective partitioned regression.

Conclusions:

• The proposed SPR model is effective for assessing kidney disease severity using low-cost lateral flow assay test strips and a smartphone app.
• This technology offers a promising, inexpensive tool for early kidney disease detection and monitoring.
• Further research is required to validate the model's performance across diverse clinical settings.