Delimiting cryptic morphological variation among human malaria vector species using convolutional neural networks.

Area of Science:

• Entomology
• Computer Science
• Machine Learning

Background:

• Visual identification of mosquito species is crucial for disease surveillance but is challenged by cryptic morphological variation within species complexes like Anopheles gambiae.
• Accurate identification is essential for effective mosquito-borne disease management and control strategies.

Purpose of the Study:

• To assess the feasibility of using Convolutional Neural Networks (CNNs) for automatic classification of mosquito sex, genus, species, and strains from 2D whole-body images.
• To demonstrate CNNs' capability in distinguishing morphologically similar cryptic species and different strains of the same species.

Main Methods:

• A library of 1,709 mosquito images from 16 colonies across five geographic regions was curated.
• Image processing, data augmentation, and CNN training/validation methodologies were developed and applied.
• CNNs were trained to classify mosquito sex, genus, species, and strains using the image dataset.

Main Results:

• The best CNN configuration achieved high prediction accuracies: 96.96% for species identification and 98.48% for sex identification.
• CNNs successfully differentiated between cryptic species, two strains of a single species, and specimens stored using different methods.
• Visualizations of CNN feature space aided in interpreting classification predictions.

Conclusions:

• CNNs offer a feasible and accurate approach for automated mosquito species and sex identification, even with cryptic morphological variation.
• This deep learning methodology holds significant potential for enhancing malaria mosquito surveillance and disease management efforts.
• The study provides a robust framework for applying AI in entomological studies and public health surveillance.