From Flow to Feature Using a Proof-of-Concept Spectral-Driven Machine Learning Approach Using Smart Urinary and
Leonardo Poggi1,2, Anastasia Meckler3, Sebastian Künert3
1Institute of Diagnostic and Interventional Radiology and Neuroradiology, Essen University Hospital, Hufelandstraße 55, Essen, 45147, Germany, 43 20172377817.
JMIR Medical Informatics
|May 14, 2026
Summary
This study introduces a smart catheter system using spectral data and machine learning (ML) for real-time fluid analysis. The system accurately differentiates pathological from healthy fluids, improving diagnostics without manual preprocessing.
Area of Science:
- Biomedical engineering
- Medical diagnostics
- Machine learning applications
Background:
- Current catheter systems offer limited diagnostic value and are prone to errors.
- Existing machine learning (ML) methods require complex preprocessing, hindering real-time analysis.
Purpose of the Study:
- To develop and evaluate a fully automated, real-time diagnostic approach for smart urinary and drainage catheter systems.
- To differentiate pathological from healthy excreted fluids using spectral data and ML without manual preprocessing.
Main Methods:
- Analyzed 454 surgical drainage fluid and 401 urine samples using smart catheters with mini-spectrometer sensors.
- Collected spectral data were processed using random forest, partial least squares discriminant analysis regression, and convolutional neural network (CNN) models.
- ML models extracted features to differentiate pathological from healthy samples based on biomarkers.
Main Results:
- All three ML approaches showed promising results.
- CNN models achieved the highest accuracy, with Matthews correlation coefficient scores of 0.83 for hemoglobin and 0.81 for bilirubin.
- The models successfully differentiated pathological from healthy drainage and urine samples using spectral features.
Conclusions:
- Spectral-driven ML holds significant potential for smart catheter systems.
- This approach enables real-time, noninvasive fluid analysis for improved diagnostics and personalized patient care.
- Further research will focus on optimizing ML models for this application.
