Non-Contact Thermal and Acoustic Sensors with Embedded Artificial Intelligence for Point-of-Care Diagnostics

Abstract

This work involves exploring non-invasive sensor technologies for data collection and preprocessing, specifically focusing on novel thermal calibration methods and assessing low-cost infrared radiation sensors for facial temperature analysis. Additionally, it investigates innovative approaches to analyzing acoustic signals for quantifying coughing episodes. The research integrates diverse data capture technologies to analyze them collectively, considering their temporal evolution and physical attributes, aiming to extract statistically significant relationships among various variables for valuable insights. The study delineates two distinct aspects: cough detection employing a microphone and a neural network, and thermal sensors employing a calibration curve to refine their output values, reducing errors within a specified temperature range. Regarding control units, the initial implementation with an ESP32 transitioned to a Raspberry Pi model 3B+ due to neural network integration issues. A comprehensive testing is conducted for both fever and cough detection, ensuring robustness and accuracy in each scenario. The subsequent work involves practical experimentation and interoperability tests, validating the proof of concept for each system component. Furthermore, this work assesses the technical specifications of the prototype developed in the preceding tasks. Real-time testing is performed for each symptom to evaluate the system's effectiveness. This research contributes to the advancement of non-invasive sensor technologies, with implications for healthcare applications such as remote health monitoring and early disease detection.

Keywords: acoustic; low-cost hardware; neural networks; remote; sensors; thermal.

Figure 1

Proposed system diagram.

Figure 2

Diagram of the thermal system.

Figure 3

Captured thermographic images with (a) sensor AMG8833 and (b) sensor MLX90640.

Figure 4

Validation test diagram.

Figure 5

Example of cough spectrogram.

Figure 6

Confusion matrix for the chosen model.

Figure 7

Flow diagram of cough detection system.

Figure 8

Example of a 2 s duration waveform.

Figure 9

Examples of audio signals for the new classes: (a) music, (b) conversation, and (c) ambient noise. Examples of spectrograms of audio signals for the new classes: (a) music, (b) conversation, and (c) ambient noise.

Figure 10

Complete system with validation tool. The different parts of the system are (1) bed mounting; (2) control unit; (3) Peltiers for calibration; (4) reference system for tests.