Machine Learning-Based Quantification of Lateral Flow Assay Using Smartphone-Captured Images

Abstract

Lateral flow assay has been extensively used for at-home testing and point-of-care diagnostics in rural areas. Despite its advantages as convenient and low-cost testing, it suffers from poor quantification capacity where only yes/no or positive/negative diagnostics are achieved. In this study, machine learning and deep learning models were developed to quantify the analyte load from smartphone-captured images of the lateral flow assay test. The comparative analysis identified that random forest and convolutional neural network (CNN) models performed well in classifying the lateral flow assay results compared to other well-established machine learning models. When trained on small-size images, random forest models excelled CNN models in image classification. Contrarily, CNN models outperformed random forest models in classifying noisy images.

Keywords: CNN; deep learning; lateral flow assay; machine learning.

Figure 1

Smartphone-captured image data collection process utilized a ring light positioned overhead a white photography box containing an LFA to limit the impact of variable lighting conditions.

Figure 2

(a) The smartphone images of lateral flow assay test conducted with varying loads of SARS-CoV-2 N proteins. (b) Magnified test window. (c) Sample images after augmentation.

Figure 3

The test accuracies, which represent the rate of correct model predictions, of the SVC, KNN, decision tree, random forest, and CNN models. The asterisk indicates a statistically significant difference (p < 0.05).

Figure 4

The training and validation losses and accuracies of the CNN model. The loss and accuracy represent the error of the model and the rate of correct model predictions, respectively.

Figure 5

The test accuracies of the CNN models trained on the lateral flow assay images with different sizes. The asterisk indicates a statistically significant difference (p < 0.05).

Figure 6

The test accuracy (breakdown by image size) of a single random forest model and CNN model trained on the lateral flow assay images with all sizes from 16 × 16 to 128 × 128. The asterisk indicates a statistically significant difference (p < 0.05).

Figure 7

The confusion matrices of the CNN models trained on the lateral flow assay images converted into RGB, HSV, YcrCb, CIE XYZ, and CIELab colorspaces.

Figure 8

The accuracies of the random forest and CNN models trained on the images with different noise levels and the photos of the images. The asterisk indicates a statistically significant difference (p < 0.05).