A Unified YOLOv8 Approach for Point-of-Care Diagnostics of Salivary α-Amylase

Abstract

Salivary α-amylase (sAA) is a widely recognized biomarker for stress and autonomic nervous system activity. However, conventional enzymatic assays used to quantify sAA are limited by time-consuming, lab-based protocols. In this study, we present a portable, AI-driven point-of-care system for automated sAA classification via colorimetric image analysis. The system integrates SCHEDA, a custom-designed imaging device providing and ensuring standardized illumination, with a deep learning pipeline optimized for mobile deployment. Two classification strategies were compared: (1) a modular YOLOv4-CNN architecture and (2) a unified YOLOv8 segmentation-classification model. The models were trained on a dataset of 1024 images representing an eight-class classification problem corresponding to distinct sAA concentrations. The results show that red-channel input significantly enhances YOLOv4-CNN performance, achieving 93.5% accuracy compared to 88% with full RGB images. The YOLOv8 model further outperformed both approaches, reaching 96.5% accuracy while simplifying the pipeline and enabling real-time, on-device inference. The system was deployed and validated on a smartphone, demonstrating consistent results in live tests. This work highlights a robust, low-cost platform capable of delivering fast, reliable, and scalable salivary diagnostics for mobile health applications.

Keywords: YOLOv8; convolutional neural networks (CNNs); edge computing; image segmentation; machine learning; object detection; point-of-care testing; α-Amylase.

Figure 1

RGB intensity analysis of the colorimetric reaction. (a) Channel-wise intensity over time. (b) Red channel kinetics for each $\alpha$-amylase class. (c) RGB color space distribution of samples after 10 min divided to 8 classes.

Figure 2

Representative images of vial samples corresponding to different $\alpha$-amylase concentrations, illustrating the distinct color intensities observed at time = 10 min.

Figure 3

Images of the SCHEDA device. (a) Perspective view showing the main enclosure and removable drawers designed to hold reaction vials. (b) Top view with LED indicators. (c) Side view showing the USB-C charging port and power switch.

Figure 4

Example of dataset annotations. (a) Original image acquired by the device camera showing a vial positioned in the drawer, along with the image resolution. (b) Bounding box annotation used for YOLOv4-CNN training. (c) Pixel-wise segmentation mask used for YOLOv8 training.

Figure 5

Confusion matrices for the YOLOv4-CNN approach. Top: RGB input; Bottom: Red-channel-only input. Left panels show sample counts; right panels show percentage accuracy per class.

Figure 6

Confusion matrix for the YOLOv8 segmentation-classification model. Left: Absolute counts; Right: Per-class percentage accuracy.