Related Experiment Video
Updated: Jul 8, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
DNN-PURE: A deep neural network approach to paper-based urea sensing
Souvik Biswas1, Arijit Pal2, Koel Chaudhury3
1School of Medical Science and Technology, Indian Institute of Technology Kharagpur, West Bengal, 721302, India; Department of Biomedical Engineering, National Institute of Technology Raipur, Chhattisgarh, 492010, India.
Biosensors & Bioelectronics
|July 3, 2026
Summary
A new paper-based sensor accurately quantifies urea in serum using impedance changes and deep learning. This low-cost diagnostic tool offers high sensitivity and improved resolution for point-of-care applications.
Area of Science:
- Biomedical Engineering
- Analytical Chemistry
- Sensor Technology
Background:
- Accurate urea quantification in serum is crucial for diagnosing various medical conditions.
- Existing methods for urea measurement can be costly, time-consuming, or require specialized laboratory equipment.
- Development of low-cost, rapid, and sensitive diagnostic tools for point-of-care urea monitoring is highly desirable.
Purpose of the Study:
- To develop a novel, low-cost, disposable paper-based enzymatic sensor for quantitative urea estimation in serum.
- To leverage a unique impedance-based detection mechanism for sensitive urea measurement.
- To integrate a deep neural network (DNN) model for enhanced accuracy and resolution in urea quantification.
Main Methods:
- Fabrication of a paper-based sensor by functionalizing polyaniline (PANI) on a chromatography paper substrate.
- Covalent immobilization of urease enzyme onto the PANI-modified substrate using glutaraldehyde cross-linking.
- Material characterization using SEM, Raman spectroscopy, and XPS; impedimetric analysis for sensor performance evaluation.
- Implementation and training of a DNN regression model on frequency-dependent impedance data for urea concentration prediction.
Main Results:
- The sensor demonstrated high sensitivity for urea detection within a clinically relevant range (0.01–1 mg/ml) with a low limit of detection (LOD) of 0.01 mg/ml.
- Material characterization confirmed successful PANI deposition and urease immobilization.
- The DNN model achieved over 99% accuracy and improved resolution to 0.02 mg/ml, significantly outperforming manual calibration (94.71%).
- Validation using Bland-Altman analysis and spiked serum samples confirmed the sensor's robustness and efficacy in biological matrices.
Conclusions:
- The developed paper-based enzymatic sensor offers a sensitive, accurate, and low-cost method for urea quantification in serum.
- The integration of a DNN model significantly enhances the sensor's precision, resolution, and automation capabilities.
- This sensing platform shows great promise for real-time, point-of-care diagnostics and various biomedical applications.
