Automated detection of ‘Ca. Liberibacter asiaticus’ infection in citrus using immune tissue prints and machine learning
Jonathan Shao: USDA ARS Molecular Plant Pathology Lab
<div>Huanglongbing, associated with infection by ‘<em>Ca</em>. Liberibacter asiaticus’, has caused catastrophic losses to the Florida citrus industry, and is widespread in urban areas of southern California. Infection of trees is followed by erratic distribution of the pathogen and a years-long latent period before symptoms appear but during which time the pathogen can be spread by the psyllid vector. qPCR is used to confirm visual symptoms before regulatory actions are taken, and thus the development of early detection methods is crucial. Immune tissue printing with a rabbit polyclonal antibody that recognizes epitopes of the major outer membrane protein of ‘<em>Ca</em>. Liberibacter asiaticus’ has been developed. The assay can be scaled to process large numbers of samples. However, scoring the tissue prints as positive or negative for the pathogen requires an expert to view images of the tissue print to determine if individual phloem cells are stained indicating infection. We have begun the development of an automated system using the Tensorflow software to create small Convoluted Neural Networks for image recognition and scoring of the tissue prints. Preliminary results are promising. We have trained our system using known positive and known negative samples obtained from graft-inoculated and qPCR-verified trees. Our results will enable rapid, accurate and unbiased scoring of images for the presence of the pathogen and may facilitate the early removal of infected trees.</div>
View Presentation |
