Worm Egg Counting using Machine learning

To evaluate the level of infestation of the soybean cyst nematode (SCN), Heterodera glycines, in the field, egg population densities are determined from soil samples. Sucrose centrifugation is a common technique for separating debris from the extracted SCN eggs. We have developed a procedure, however, that employs OptiPrep as a density gradient medium, with improved extraction and recovery of the eggs compared to the sucrose centrifugation technique. Also, we have built computerized methods to automate the identification and counting of the nematode eggs from the processed samples. One approach uses a high-resolution scanner to capture static images of the eggs and debris on filter papers and a deep learning network is trained to detect and count the eggs. The second approach utilizes a lensless imaging setup with off-the-shelf components and the egg samples flow through a microfluidic flowchip. Holographic videos are taken of the passing eggs and debris, which are then reconstructed and processed by a custom software program to calculate egg counts. To evaluate the performance of the software programs, SCN-infested soils were collected from two farms and the results were compared with manual counts.