Intelligent image-based colourimetric tests using machine learning framework for lateral flow assays

Tania, Marzia Hoque, Lwin, Khin T., Shabut, Antesar M., Najlah, Mohammad, Chin, Jeannette ORCID: and Hossain, M. A. (2020) Intelligent image-based colourimetric tests using machine learning framework for lateral flow assays. Expert Systems with Applications, 139. ISSN 0957-4174

[thumbnail of Accepted_Manuscript]
PDF (Accepted_Manuscript) - Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (8MB) | Preview


This paper aims to deliberately examine the scope of an intelligent colourimetric test that fulfils ASSURED criteria (Affordable, Sensitive, Specific, User-friendly, Rapid and robust, Equipment-free, and Deliverable) and demonstrate the claim as well. This paper presents an investigation into an intelligent image-based system to perform automatic paper-based colourimetric tests in real-time to provide a proof-of-concept for a dry-chemical based or microfluidic, stable and semi-quantitative assay using a larger dataset with diverse conditions. The universal pH indicator papers were utilised as a case study. Unlike the works done in the literature, this work performs multiclass colourimetric tests using histogram based image processing and machine learning algorithm without any user intervention. The proposed image processing framework is based on colour channel separation, global thresholding, morphological operation and object detection. We have also deployed a server based convolutional neural network framework for image classification using inductive transfer learning on a mobile platform. The results obtained by both traditional machine learning and pre-trained model-based deep learning were critically analysed with the set evaluation criteria (ASSURED criteria). The features were optimised using univariate analysis and exploratory data analysis to improve the performance. The image processing algorithm showed >98% accuracy while the classification accuracy by Least Squares Support Vector Machine (LS- SVM) was 100%. On the other hand, the deep learning technique provided >86% accuracy, which could be further improved with a large amount of data. The k-fold cross validated LS- SVM based final system, examined on different datasets, confirmed the robustness and reliability of the presented approach, which was further validated using statistical analysis. The understaffed and resource limited healthcare system can benefit from such an easy-to-use technology to support remote aid workers, assist in elderly care and promote personalised healthcare by eliminating the subjectivity of interpretation.

Item Type: Article
Uncontrolled Keywords: image processing,histogram thresholding,feature selection,machine learning,deep learning,colourimetric tests
Faculty \ School: Faculty of Science > School of Computing Sciences
UEA Research Groups: Faculty of Science > Research Groups > Cyber Security Privacy and Trust Laboratory
Related URLs:
Depositing User: LivePure Connector
Date Deposited: 30 Jul 2019 11:30
Last Modified: 14 Mar 2023 08:33
DOI: 10.1016/j.eswa.2019.112843


Downloads per month over past year

Actions (login required)

View Item View Item