OocystMeter, a machine-learning algorithm to count and measure Plasmodium oocysts, reveals a clustering pattern in the Anopheles midgut

Abstract We present OocystMeter, a machine learning-based software developed to automate the segmentation of malaria oocysts from images of mosquito midguts stained with mercurochrome. Existing bioimage analysis tools, including machine learning-based ones, often struggle with the unique staining patterns, complex midgut backgrounds, and variable morphology of oocysts, making the determination of oocyst size and numbers cumbersome. To overcome these challenges, we curated a high-quality dataset comprised of 11,178 Plasmodium falciparum oocysts in Anopheles gambiae midguts annotated by expert parasitologists. Using this dataset, we fine-tuned a Mask R-CNN object detection model to achieve segmentation accuracy comparable to human parasitologists (Spearman’s correlation of 0.998 for oocyst counts and 0.978 for size measurements). Combining tool-assisted expert annotations with spatial analysis, we identified a non-random oocyst clustering pattern, detectable at shorter spatial scales in infections with 4–74 oocysts/midgut. The analyses did not identify a consistent anatomical region or axis that explains this pattern. Our workflow accelerates image-based oocyst counting and size measurement, standardizes image analysis, and provides spatial coordinates for parasitology studies. Manual dissection, staining, mounting, and image acquisition by human scientists remain necessary. OocystMeter is freely available at https://github.com/duopeng/oocystMeter , and as a web-based tool at http://oocystmeter.org/ , offering a valuable resource for researchers investigating the oocyst stage of malaria development.