Pebrine diagnosis using quantitative phase imaging and machine learning.

Abstract

Pebrine is the most dreaded infectious disease of the silkworm and has devastated sericulture in Europe during the 18th century. Thereafter, if it is detected, the crop is burned to prevent further dissemination. The conventional microscopic examination of moth's body fluid is erroneous and it exacerbates on Metarhizium anisopliae (MA) contaminated test samples. This is due to the resemblance of pebrine and MA spores in the microscopic examination. Therefore, this study aims to demonstrate an efficient pebrine detection technique. In the proposed method, a motorised brightfield microscope is custom-made to acquire focused and defocused images of test spores. These images are used to produce quantitative phase images of the spores by the transport of intensity equation method. The phase images' histogram of oriented gradients feature is used by a machine learning classifier to categorise the spores. This system classified 92 pebrine and 185 MA spores with an accuracy of 97% at 0.04 seconds/spore. The duration taken for image acquisition is 2.5minutes per sample (10 fields of view covering an area of 302×260 μm2 ). The proposed method shows reliable results in pebrine diagnosis and would be an efficient alternative for current approaches.