Explainable Deep Learning-Assisted Self-Calibrating Colorimetric Patches for In Situ Sweat Analysis

Area of Science:

• Biomedical Engineering
• Analytical Chemistry
• Artificial Intelligence

Background:

• Sweat contains diverse biomarkers reflecting physiological states.
• Noninvasive biosensing offers a promising alternative to traditional methods.
• Accurate and efficient detection of sweat biomarkers is crucial for health monitoring.

Purpose of the Study:

• To develop a wearable, self-calibrating colorimetric biosensing platform assisted by explainable deep learning (DL).
• To precisely detect and quantify biomarker concentrations in sweat.
• To enhance the interpretability and reliability of noninvasive biosensing technologies.

Main Methods:

• Integration of colorimetric sensing, adsorbing-swelling hydrogel, and explainable DL algorithms.
• Development of an enzyme/indicator-immobilized colorimetric patch.
• Utilizing a dataset of 5625 colorimetric images for algorithm assessment, including convolutional neural networks (CNNs).

Main Results:

• The CNN model achieved 100% accuracy in classifying and quantifying Zn2+, glucose, and Ca2+ in sweat.
• DL-assisted colorimetric analysis showed 91.7-97.2% agreement with UV-Vis spectroscopy for actual sweat samples.
• Class activation mapping (CAM) visualized CNN operations, validating the biosensing technology's design.

Conclusions:

• The developed platform provides an efficient and accurate method for noninvasive sweat biomarker analysis.
• Explainable DL enhances the understanding and reliability of the biosensing system.
• This technology offers potential for health monitoring, disease prevention, and clinical diagnosis.