Related Experiment Video
Updated: Jan 8, 2026
09:16
Supervised Machine Learning for Semi-Quantification of Extracellular DNA in Glomerulonephritis
Published on: June 18, 2020
7.3K
Enhancing Glomerular Hematuria Identification in Automated Urinalysis Using a Light Gradient Boosting Machine-Based
Rongrong Wang1, Jia Xu1, Jing Jin1
1Department of Clinical Laboratory, State Key Laboratory of Complex Severe and Rare Diseases, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
Annals of Laboratory Medicine
|December 17, 2025
Summary
A new LightGBM model significantly improves the detection of glomerular hematuria (GH) in urine analysis. This machine learning approach enhances accuracy and sensitivity compared to automated systems, aiding in earlier disease diagnosis.
Area of Science:
- Nephrology
- Medical Diagnostics
- Machine Learning in Healthcare
Background:
- Automated urinalysis systems like UF-5000 struggle to accurately identify glomerular hematuria (GH).
- Misclassification of GH can delay the diagnosis of underlying kidney diseases.
- Developing improved methods for GH detection is crucial for timely and accurate patient care.
Purpose of the Study:
- To develop and validate a machine learning model for enhanced detection of glomerular hematuria (GH) using automated urinalysis data.
- To compare the performance of the developed model against existing automated systems and other machine learning algorithms.
- To identify key urinary parameters predictive of GH using explainable AI techniques.
Main Methods:
- A dataset of 5,444 urine samples with positive occult blood results was analyzed.
- Samples were manually classified into non-glomerular hematuria (NGH), GH, mixed hematuria, and non-hematuria groups.
- Sixty-five parameters from UF-5000 and UC-3500 analyzers were used to train and compare LightGBM, XGBoost, Random Forest, and Logistic Regression models.
- The LightGBM model was validated using 10-fold cross-validation and an independent test set.
Main Results:
- The LightGBM model achieved significantly higher GH recognition accuracy (44% validation, 37% testing) compared to the UF-5000 system (16% validation, 11% testing).
- Sensitivity for GH detection increased from 0.3 (UF-5000) to 0.7 (LightGBM).
- Key predictive parameters identified by SHapley Additive exPlanations included red blood cell count, forward scatter peak, and urinary protein level.
Conclusions:
- An interpretable LightGBM-based model substantially improves automated identification of glomerular hematuria (GH).
- This model demonstrates significant potential as a tool for accurate hematuria source classification.
- The findings suggest a promising advancement in urinalysis for diagnosing kidney conditions.
