Using Machine Learning to Objectively Determine Colorimetric Assay Results from Cell Phone Photos Taken Under Ambient Lighting

Abstract

Colorimetric assays are an important tool in point-of-care testing that offers several advantages such as rapid response times and inexpensive costs. A factor that currently limits their use is objective measures to determine results. Current solutions consist of creating a test reader that standardizes the conditions the strip is under before measuring. However, this increases the cost and decreases the portability of these assays. The focus of this study is to train a convolutional neural network (CNN) that can objectively determine results of colorimetric assays under varying conditions. To ensure the flexibility of the model to several types of colorimetric assays, three models are trained on the same CNN. The images these models are trained on consist of positive and negative images of ETG (99.87% positive classification, 99.96% negative classification), fentanyl (99.60% positive classification, 99.56% negative classification), and HPV antibody (99.86% positive classification, 100% negative classification) strips taken under different lighting and background conditions. A fourth model is trained on an image set composed of all three strip types with the lowest classification accuracy being 99.11%.