Environmental AI: Using machine learning to address mosquito-borne diseases

Description

Wovenware has developed a machine learning solution for the Puerto Rico Science, Technology & Research Trust, which automates the identification and classification of various species of mosquitoes in order to control the spread of diseases such as Zika and dengue fever across Puerto Rico and ultimately worldwide. The end goal is also to help researchers develop safe and effective insecticides to combat them.

Leslie De Jesus outlines how Wovenware applies AI to gain an understanding of why many mosquitoes have become immune to insecticides approved by the FDA. The challenges of undertaking such a project are great. Researchers have spread out across the island, capturing different mosquito species in traps, monitoring and testing them for viral presence and insecticide resistance, and labelling and classifying them. Manually capturing and classifying thousands of mosquitoes in different locales can take many months before any specific patterns can emerge.

With a deep learning solution, Wovenware has automated this time-consuming task, and through a private crowd of data specialists, it identifies and labels thousands of images of mosquitoes and datasets to train an algorithm to automatically identify and classify specific species. The AI-based solution will save the Puerto Rico Vector Control Unit (PRVCU) months of time that can be used to more quickly analyze the findings and identify the root cause of resistance to insecticides, as well as disease spread and prevention routes. The counting and classification of specific mosquitoes is being accomplished using CNNs, with a dataset that contains multiple mosquito annotations per image. Its experiments are incrementally adding complexity to the solution. As a baseline approach, it implements custom deep CNNs for gender and species classification using Keras, which is being trained end to end, independently, from scratch.

Following this approach, the model identifies singular features in order to overcome current challenges confronted from specimens mutilated by weather conditions and the retrieval of the mosquitoes from traps. Models are trained on four GPUs Octoputer powered by NVIDIA’s GeForce Titan XP Pascal. The speed of the server allows for more effective decision making in the training phase. A key outcome realized to date is an order-of-magnitude improvement in the time required for counting and classifying specific vectors. This project truly underscores the impact AI-based technology can have when bright minds are augmented by smart technology in helping to solve some of the world’s most pressing challenges.