FIU researchers build AI mosquito-egg counter, with human checks still needed
AeEggCounter’s field counts tracked specialist tallies closely despite weaker detection performance. The tool is available through a free Android and iOS app.
Listen to this story
The audio brief
Story brief
3 key pointsFIU’s AeEggCounter is now being used in CDC-funded field studies in Little Havana and East Hialeah, extending the tool beyond classroom and dataset evaluation. The evidence supports it as an aid to surveillance: on 76 unseen field samples, its total counts correlated strongly with specialists’ tallies, but FIU reported weaker detection performance and researchers say humans must confirm results, particularly in sensitive settings. The free ACTT-Study app packages the tool for Android and iOS, while egg-by-egg reliability in field conditions remains a key limitation to watch.
- 01
Researchers trained the YOLO-based detector using 4,398 manually labeled eggs across 168 smartphone images.
- 02
In initial testing, 88.3% of objects flagged as eggs were correct, and the model detected 91.7% of eggs present.
- 03
The field comparison used 76 previously unseen samples photographed with a Samsung Galaxy S24 Ultra or an iPhone 16 Pro Max.
Counting mosquito eggs no longer has to begin and end at a microscope. An FIU-led team has developed AeEggCounter, which detects and counts eggs in smartphone photos. In a November 16 research account, the university described promising initial results and a tougher field test: counts tracked specialists closely, but detection performance declined.
The same egg traps, a different way to count
The surveillance process starts with ovitraps: small containers filled with water that attract female Aedes mosquitoes looking for somewhere to lay eggs. Brown paper strips collect the eggs. Specialists traditionally examine those strips under a microscope, identifying and counting the tiny objects in a process FIU describes as labor-intensive and prone to error.
AeEggCounter applies an object-detection algorithm called YOLO, short for “You Only Look Once.” It identifies and locates multiple objects in an image in a single pass. Here, its targets are individual mosquito eggs, including those surrounded by small pieces of debris in laboratory and field samples.
That recognition depended on manual preparation. Researchers identified and labeled 4,398 eggs across 168 smartphone images to train the model. The work was led by Andre Costa-da-Silva, an assistant professor of biology and researcher in Florida International University’s Biomolecular Sciences Institute.
Spotting eggs and matching counts are different tests
The researchers reported two measures from the initial test. Of the objects the model flagged as eggs, 88.3% were correctly identified. Looking instead at all the eggs actually present in the images, it detected 91.7%. Those figures answer different questions: whether a flagged object is an egg, and whether an egg gets noticed.
The next evaluation moved to 76 previously unseen field samples. Trained specialists supplied the human counts against which the automated results were compared. The testing dataset excluded samples used for training, so this stage assessed the tool on material it had not already encountered during that process.
The NLM Dataset Catalog adds detail about that comparison. Specialists screened and counted the samples before researchers photographed the paper linings in panoramic mode. They used a Samsung Galaxy S24 Ultra for 44 samples and an iPhone 16 Pro Max for 32. The dataset supports reproducibility and independent evaluation of egg-detection algorithms.
Performance declined on those field samples, FIU said, although the overall egg counts showed a strong correlation with specialists’ counts. That is a narrower finding than matching people egg by egg. The field result concerns how closely the totals tracked human tallies; the initial percentages describe detection of individual eggs.
From evaluation to field studies
The findings were published in the Journal of the American Mosquito Control Association. Alongside FIU, the research team included the University of South Carolina, Boston University, Baylor College of Medicine and Texas Children’s Hospital. AeEggCounter is available through ACTT-Study, a free smartphone app for Android and iOS.
The app has also reached FIU’s classroom. Biology student Melissa Gonzalez described setting out egg-collection cups at home, then returning to class to count the eggs both manually and with the app. That exercise put the automated result alongside the traditional count rather than treating it as a substitute.
Beyond the classroom, co-investigator Matthew DeGennaro, director of FIU’s Biomolecular Sciences Institute, is using the app in CDC-funded field studies in Little Havana and East Hialeah. The counting tool is already part of field research, not just an evaluated model.
Sources
- datasetcatalog.nlm.nih.govAeEggCounter development: Unmarked and marked images from the testing dataset
- news.fiu.eduUnder surveillance: AI takes aim at mosquito eggs
Reader comments
Newest comments first. Replies stay oldest first.