Automatic diagnosis of tuberculosis disease based on Plasmonic ELISA and color-based image classification

Abstract

Abstract-Tuberculosis (TB) remains one of the most devastating infectious diseases and its treatment efficiency is majorly influenced by the stage at which infection with the TB bacterium is diagnosed. The available methods for TB diagnosis are either time consuming, costly or not efficient. This study employs a signal generation mechanism for biosensing, known as Plasmonic ELISA, and computational intelligence to facilitate automatic diagnosis of TB. Plasmonic ELISA enables the detection of a few molecules of analyte by the incorporation of smart nanomaterials for better sensitivity of the developed detection system. The computational system uses k-means clustering and thresholding for image segmentation. This paper presents the results of the classification performance of the Plasmonic ELISA imaging data by using various types of classifiers. The five-fold cross-validation results show high accuracy rate (>97%) in classifying TB images using the entire data set. Future work will focus on developing an intelligent mobile-enabled expert system to diagnose TB in real-time. The intelligent system will be clinically validated and tested in collaboration with healthcare providers in Malaysia.

Assay Type Detection Using Advanced Machine Learning Algorithms

“…Such an interactive overlay could ease the necessity to use computational intelligence to separate the region of interest. Alternatively, one could use computer vision and machine learning techniques such as [17], [41].…”

“…One is the Gold Nanoparticles (AuNP) based plasmonic ELISA for TB-antigen specific antibody detection, referred to as the TB-test in this paper. The description of the sample preparation for the TB-test is provided in [9], [41], [42]. However, development of the biosensor is not the focus of this work.…”

“…However, the intra-class classification would require the model to search for only the colors even when the rest of the geometric features are similar, and occurrence of those colors are in the same location. Therefore, it is more rational to use a simpler machine learning model for the colorimetric classification (such as [17], [41]), instead of building more deep layers which would require more processing capacity, memory size, larger dataset and more dependency on the cloud-based approach [11], [21].…”

Clustering and Classification of a Qualitative Colorimetric Test

“…The aim of this work is to improve the detection accuracy than [5] for a larger dataset (Table I). The plasmonic ELISA based tuberculosis wet-chemical experiment was conducted in University Putra Malaysia.…”

“…The detail bio-chemical based experiment of plasmonic ELISA is described in [4]. In our preliminary study [5], we have provided the colorimetric detection utilizing unsupervised learning for image processing and supervised machine learning techniques for classification. Using 71 samples, we achieved 97.2% accuracy with 5-fold cross validation (CV) via Random Forest (RF).…”