Session: Artificial Intelligence in Mosquito and Vector Control/Behavior/Biology/Genetics
158 - Deployed use and performance of IDX for AI vector identification in Maryland’s mosquito surveillance program
Wednesday, March 6, 2024
4:50pm – 5:00pm
Location: D3
Abstract: Convolutional neural networks (CNNs) for image recognition, a deep learning method, have emerged as a promising modality with the capability to visually differentiate between mosquito species. However, use of this technology on wild field-collected specimens in an operational vector surveillance program has been limited. Here we present the first performance metrics of the IDX, Vectech’s system for AI mosquito identification, as part of Maryland’s mosquito control program in Anne Arundel County, MD. Specimens were collected on a weekly basis from twelve CDC gravid trap collection sites throughout Anne Arundel County over five weeks. Specimens were first identified and separated by species and sex using conventional methods by a professional entomologist inspecting morphology under a microscope. A technician then imaged the specimens using Vectech’s IDX system. By comparing entomologist identification to the algorithm output by IDX, we are able to calculate the accuracy of the system across species. Over the study period, 2,591 specimens were collected and imaged representing 14 species, 10 of which were available in the identification algorithm on the device during the study period. The micro average accuracy was 94.9%. Of these 10 species, 7 species consisted of less than 30 samples. The macro average accuracy when including these species was 79%, while the macro average when excluding these species was 93%. Roughly three-fourths of the specimens collected were Culex pipiens sl,with algorithm identification accuracy of 96.4%.These advancements demonstrate the current capabilities of the IDX in the Mid-Atlantic region and its potential to support vector surveillance programs that may not have access to expert taxonomists or may have limited capacity to evaluate significant quantities of specimens collected at peak season.
