Related Experiment Video
Updated: Aug 23, 2025
A Detailed Protocol for Perspiration Monitoring Using a Novel, Small, Wireless Device
Published on: November 24, 2016
Explainable Deep-Learning-Assisted Sweat Assessment via a Programmable Colorimetric Chip.
Zhihao Liu1,2, Jiang Li1,2, Jianliang Li1,2
1College of Chemistry and Materials Science, Guangdong Provincial Key Laboratory of Functional Supramolecular Coordination Materials and Applications, Guangdong Engineering & Technology Research Centre of Graphene-like Materials and Products, Jinan University, Guangzhou 510632, China.
This study introduces a programmable colorimetric chip with explainable deep learning (DL) for sweat analysis. The system accurately classifies and quantifies glucose, pH, and lactate, offering transparent insights into DL algorithms for health monitoring.
Area of Science:
- Biomedical Engineering
- Analytical Chemistry
- Artificial Intelligence
Background:
- Biofluid analysis faces challenges in speed, throughput, and accuracy due to complexity and individual differences.
- Deep learning (DL) shows promise in image analysis for identification but often functions as a "black box" with opaque mechanisms.
- Current methods for sweat analysis lack facile, high-throughput, and explainable diagnostic tools.
Purpose of the Study:
- To develop a programmable colorimetric chip integrated with explainable deep learning for accurate sweat sample analysis.
- To address the transparency limitations of current DL algorithms in clinical diagnostics.
- To enable facile, high-throughput, and accurate classification and quantification of analytes in human sweat.
Main Methods:
- Designed programmable colorimetric chips using gel capsules with various indicators.
- Collected a dataset of 4600 colorimetric response images for algorithm training and assessment.
- Evaluated two DL algorithms, including a convolutional neural network (CNN), and seven machine learning (ML) algorithms.
- Utilized Class Activation Mapping (CAM) to visualize and interpret the CNN's decision-making process.
Main Results:
- The CNN DL algorithm achieved 100% accuracy in classifying and quantifying glucose, pH, and lactate in sweat.
- DL-assisted colorimetric approach testing on actual sweat samples showed 91.0-99.7% agreement with laboratory measurements.
- Class activation mapping (CAM) successfully visualized the CNN's inner workings, validating the colorimetric chip design.
Conclusions:
- The explainable DL-assisted programmable colorimetric chip offers an "end-to-end" strategy for transparent DL analysis of sweat.
- This approach facilitates software design optimization and provides facile indicators for clinical monitoring and disease prevention.
- The developed system contributes to new scientific discoveries and improved health assessment strategies.
