Session: SYMP 1-8: Wearables

Paper Number: 140275

140275 - Accurate Sweat Sensing With Multifunctional Wearable Patch and Machine Learning 

In healthcare, blood pH and glucose concentration are vital indicators, especially for chronic conditions like diabetes. However, blood sampling is invasive and unaffordable for many. On the other hand, non-invasive and continuous monitoring of pH value and glucose concentration face major challenges, like inaccuracy. In this talk, we will present a promising solution based on a colorimetric wearable patch that converts pH values and glucose concentrations in sweat into color transitions. These battery-free and affordable colorimetric biosensors are made of cotton textile and can be combined with smartphone-based data collection and machine learning analysis. The sensing indicator for sweat pH is composed of a mixture of methyl orange and bromocresol green and those indicators are spin-coated onto the cotton substrate. Through the reduction of cotton textile surface modification and a new synthesis approach, the performance of pH sensor is significantly improved with clear color transition from yellow to dark green as the pH increase. This enhancement is further evidenced by the increase in the prediction accuracy of machine learning models. Additionally, two types of high-contrast glucose sensors, which utilize 3,3',5,5'-tetramethylbenzidine (TMB) and potassium iodide (KI) as chromogenic agent respectively, are fabricated using a controlled deposition method. The TMB-based glucose sensor transitions from light blue to dark blue, whereas the KI-based glucose sensor shifts from beige to brown with the increase of glucose concentrations. The distinct color transitions of TMB-based and KI-based glucose sensor demonstrate the effectiveness of our data analysis for various types of textile colorimetric sensor with a completely different color spectrum. Standard solutions with varying pH values (pH of 4-10) and glucose concentrations (0.03 mM to 1 mM) are used to assess the effectiveness of the pH sensors and glucose sensors. To achieve accurate and convenient data acquisition, we utilize a smartphone to capture the images of the sensors’ colored patches after exposure to standard solution. The captured images are then pre-processed in Python and analyzed by three machine learning algorithms: Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Convolutional Neural Network (CNN). Through employing image segmentation strategy, our classification models can learn richer information and capture more representative features from those captured images. The classification accuracy of our machine learning models improves significantly, achieving an accuracy of 90%. These findings not only provide effective methods of sweat sensor synthesis but also demonstrate the significance and practicality of various machine learning algorithms for colorimetric analysis of textile sweat sensors.

Presenting Author: Lijun Zhou University of Washington

Presenting Author Biography: Lijun Zhou got his Bachelor's degree at Xi'an Jiaotong University in China, in 2022.
Now he is a second-year PhD student at the University of Washington under the supervision of Professor Mohammad Malakooti. Now he is working on multifunctional composites inverse design with machine learning.

Authors:

Lijun Zhou
Mohammad Malakooti