Welcome to the snail host detector!

This is an API to be used by the public to detect snail species that act as an intermediate host for two common neglected tropical diseases of great public and veterinary health importance: schistosomiasis and liver fluke disease.

Introduction

This web-API is a deep learning object detector that is designed for the public to upload snail images and detect if the images contain one of the intermediate host (IH) snails of the neglected tropical disease (NTD), schistosomiasis or fasciolasis. These diseases are spread by snail species that belong to the genus Biomphalaria or Bulinus, and Radix respectively. This makes the detection of IH snails automatic and more efficient than manual identification and, ultimately, may help in faster detection of disease hotspots.

Biomphalaria spp.

Radix spp.

Currently, the detection model is trained based on ~2500 images (up to 50,000 snails) collected by citizen scientists, from 2020 March to 2021 October, from southern Lake Albert in Uganda in the frame of the ATRAP project. (detailed description). In this area, only intestinal schistosomiasis occurs, with only few Bulinus truncatus species. Therefore, this web-API focuses on Biomphalaria species only. Since all three Biomphalaria species endemic to this area transmit intestinal schistosomiasis, we identify up to genus level.

The model is based on YOLOv4 algorithm written by Bochkovskiy, Wang and Liao (2020) and it is possible to detect 4 main groups of snails, Biomphalaria spp. (AP50*= 97.93%), Radix spp. (AP50= 98.98%), Gyraulus spp. (AP50= 87.94%) (morphologically similar to Biomphalaria but not transmitting schistosomiasis) and Pool species (non-IH snails, grouped by 4 classes of labels (Pool, Pool_triangle, Pool_spiral, Pool_oval)).

* AP (Average precision) is a standard metric to evaluate the accuracy of a object detection model.

Note that the model is trained on field images, it is possible that the model could not accurately detect snails from other settings

Thus, we also hope to collect more images related to IH snails across the world to improve detection performance. As freshwater snails exhibit cryptic interspecific variation and high intraspecific trait plasticity, we also aim to build up an image database from both public and academic support to develop detection models tailored to the regional snail community.


What function do we provide here?

In this API, we provide the Detection and counting of snails

Take me to detect snails

Using the Detection function, the general public/ students/ scientists can detect the presence of IH snails in the images they upload, and count them according to genus. You can i) adjust the configuration of the detection model (such as intersection of union and confidence threshold) and ii) upload images containing snails to be detected. Once the detection is finished, the API will show the snail counts according to the group and the detected images. We also provide a iii) download function where you can download the ZIP file containing all the information (images, detailed results in json format) related to the detection, including bounding box coordinates, pixel size and image blurriness. You can also iv) upload the result json file to read the results through the API if you are not familiar with the json format.

Detection examples

Here are some examples of detections:

Notably, the performance towards detecting multiple genera within one image is not as good as towards one genus in one image. We are currently collecting more data and aim to improve its performance.

This API is financed by the ATRAP project of the Development Cooperation program of the Royal Museum for Central Africa with support of the Directorate-General Development Cooperation and Humanitarian Aid. The ATRAP project is coordinated by the Royal Museum of Central Africa (Belgium) with the collaboration with the University of Kinshasa (Democratic Republic of Congo), Mabarara University of Science and Technology (Uganda) and KU Leuven (Belgium).

logo logo logo logo