Chagas disease vectors identification using visible and near-infrared spectroscopy

• Published in: Chemometrics and intelligent laboratory systems Vol. 197; p. 103914
• Main Authors: Depickère, Stéphanie, Ravelo-García, Antonio G., Lardeux, Frédéric
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 15-02-2020; Elsevier
• Subjects: Chagas disease; Classification; Life Sciences; Machine learning; Quantitative Methods; Santé publique et épidémiologie; Species determination; Triatominae; VIS-NIRS; Chagas disease; Species determination; Triatominae; VIS-NIRS; Machine learning; Classification
• ISSN: 0169-7439; 1873-3239
• DOI: 10.1016/j.chemolab.2019.103914

Chagas disease, caused by the parasite Trypanosoma cruzi, is widespread in Latin America, where the disease remains one of the major public health problems. This condition is mostly transmitted by triatomines which are haematophagous insects all their life. With 154 species described in the world, the correct determination of the species involved in the transmission is crucial to develop efficient control strategies. This can be achieved by taxonomic keys (available only for adult stages, nymphal instars must be reared), or by molecular techniques. Both are time and/or money consuming, showing the needs of new identification tools, especially for nymphal instars which are the most frequently found on the field. Visible and near-infrared spectroscopy (VIS-NIR), used successfully these last years in various organisms’ determination, was applied on a sample of three species from Bolivia: Triatoma infestans, Triatoma sordida and Triatoma guasayana. The spectrum of the dorsal part of the head from nymphal instars and adult stages was taken for each specimen of each species. Different methods of pre-processing and selection of variables (wavelengths) were tested to find the best model of classification for the three species. Each model was evaluated by different indices: accuracy, specificity, and F1 score. The comparison of the performance of each model evidenced that the best results were obtained when using a short spectrum (400–2000 ​nm) without pre-processing. A total of 32 components were retained by tuning, and 933 wavelengths were kept by the backward feature selection algorithm. Applying it on a new sample of insects, this model showed a global accuracy of 97.2% (95.0–98.6). The F1 score was greater than 0.95, and the specificity greater than 0.94 for all the species. For the first time, a tool is available to quickly identify and with a high accuracy nymphal instars and adults of triatomines.
•Chagas disease vector identification based on spectroscopy.•Exploration of different pre-processing and feature selection methodologies.•Only one model for females, males and nymphal instars identification.•Accuracy of 97.2% (95.0–98.6) for three species, from which two morphologically similar.