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

Abstract

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.

Pathological Test Type and Chemical Detection Using Deep Neural Networks: A Case Study Using ELISA and LFA Assays

“…Although the dataset utilised by Hoque Tania et al (2020) and Mutlu et al (2017) were much smaller in size, as compared to the dataset utilised in this paper, the colourimetric classification accuracy attained by the traditional approaches (Table 6) shows a higher prospect if a combination of DL and TML is utilised as proposed in our framework (Figure 21: Proposed framework for assay type detection and corresponding colourimetric classification). 2017Based on the aforementioned discussion, this paper suggests (Figure 21: Proposed framework for assay type detection and corresponding colourimetric classification) to utilise pre-trained model-based DL algorithms for the assay type and to process the outcome using traditional machine techniques to produce the colourimetric decision as proposed in (Shabut et al, 2018;Hoque Tania et al, 2020). In this way, the image-based intelligent colourimetric test would provide more automation while processing the pathological test or colourimetric chemical detection while maintaining high accuracy and reliability, faster computation, and retaining more autonomy to users.…”

“…Therefore, based on the aforementioned discussion, subsequent to the assay type detection, it is more logical to use a simpler machine learning model (Figure 21: Proposed framework for assay type detection and corresponding colourimetric classification) for the colourimetric classification, instead of building more deep layers which would require more processing capacity, memory size, larger dataset and more dependency on the cloud-based approach. In the literature, Hoque Tania et al (2020) and Mutlu et al (2017) utilised the same case study of LFA to perform colourimetric classification as this paper. Both of these reported studies vary in compatibility with the ASSURED criteria.…”

“…Both of these reported studies vary in compatibility with the ASSURED criteria. Taking a different approach to feature extraction, both Hoque Tania et al (2020) and Mutlu et al (2017) settled for utilising Least-squares Support-Vector Machine (LS-SVM) to classify the pH levels, achieving 100% classification accuracy. The important considerations to perceive the comparative performance between the DL-based models in this paper and TMLs in the literature is shown in Table 6.…”

“…to observe the model's ability to learn from more data. The dataset was downsized up to the dataset of Hoque Tania et al (2020), consisting of 65 images for each of the 8 chosen pH levels. The classification accuracy for Hoque Tania et al (2020) dataset, using 70% of them for training, was 86.86%.…”

“…The dataset was downsized up to the dataset of Hoque Tania et al (2020), consisting of 65 images for each of the 8 chosen pH levels. The classification accuracy for Hoque Tania et al (2020) dataset, using 70% of them for training, was 86.86%. The result in Figure 18 shows, the AlexNet was able to learn more from new data, as the accuracy increased up to ~7% when more data were added.…”