ArtiFISHal Intelligence: merging technology and ecology to understand infectious disease dynamics
Lead Research Organisation:
CARDIFF UNIVERSITY
Department Name: School of Biosciences
Abstract
Wildlife diseases play a role in species extinctions, spillover of pathogens to humans (zoonoses), and can threaten food security (e.g., wild fisheries). Systematic monitoring of infectious diseases in wildlife, however, is extremely sparse, exceptions in the UK include key zoonotic pathogens, e.g., avian influenza in birds and Mycobacterium bovis (bovine TB) in badgers, Meles meles. For the majority of wildlife, disease goes undetected, yet understanding where, when and why wildlife disease occurs is essential for safeguarding biodiversity, agricultural resources as well as animal and public health. Artificial Intelligence (AI), however, could offer a new tool by which to fill the knowledge gap. Many images are shared publicly on social media with no particular scientific intent, but those images can be a rich source of important ecological data. The presence of wildlife disease is one such example. Many diseases have detectable, visible symptoms, for example, skin lesions and ulcers. Images submitted to social media are numerous and are often taken across broad geographical ranges and time periods. Indeed, social media with its user network of >2 billion worldwide reflects the observed reality (in image format) of a quarter of the human population. Social media sites 'Flickr' and 'Facebook' have amassed an estimated 100 billion images alone, with over 100 million new images uploaded every day. To harness this huge potential for information we will collate a large database of images from social media and use computer vision, a form of artificial intelligence (AI) to detect disease within those images.
We will use wild fish as our test bed because the angling community (an estimated 2.9 million anglers), have a strong culture of photographing their catches and posting them online, which provides high monitoring potential. Wild fisheries are an important group of species that represent a significant income generator, worth over £1billion annually in the UK alone in recreation and job creation. Poor fish health therefore represents an angling, economic and conservation threat.
Our pilot work has shown that computer vision can identify symptoms of common diseases of fish from images with high efficacy. Because many images are taken with GPS-enabled devices we can extract time and date from those images, allowing us to gain insight into where and when hotspots of disease occur. Once we have data on where and when disease occurs, we can then assess why disease occurs. Understanding why disease occurs is essential to management. We will use statistical models to assess which factors underlie disease presence on fish (temperature, water flow, pollution), allowing us to facilitate better control of diseases. Ultimately this research approach could be widened to a range of diseases in different wildlife host species at a global level. Such data would provide important insight into infectious disease dynamics and potentially provide early warnings for zoonotic spillover potential to humans.
We will use wild fish as our test bed because the angling community (an estimated 2.9 million anglers), have a strong culture of photographing their catches and posting them online, which provides high monitoring potential. Wild fisheries are an important group of species that represent a significant income generator, worth over £1billion annually in the UK alone in recreation and job creation. Poor fish health therefore represents an angling, economic and conservation threat.
Our pilot work has shown that computer vision can identify symptoms of common diseases of fish from images with high efficacy. Because many images are taken with GPS-enabled devices we can extract time and date from those images, allowing us to gain insight into where and when hotspots of disease occur. Once we have data on where and when disease occurs, we can then assess why disease occurs. Understanding why disease occurs is essential to management. We will use statistical models to assess which factors underlie disease presence on fish (temperature, water flow, pollution), allowing us to facilitate better control of diseases. Ultimately this research approach could be widened to a range of diseases in different wildlife host species at a global level. Such data would provide important insight into infectious disease dynamics and potentially provide early warnings for zoonotic spillover potential to humans.
Organisations
Publications
Jones C
(2024)
AI-based discovery of habitats from museum collections
in Trends in Ecology & Evolution
Title | Science outreach film on project aims |
Description | Recorded by the Water Institute at Cardiff University a short film was made communicating the aims and outcomes of the research project. |
Type Of Art | Film/Video/Animation |
Year Produced | 2023 |
Impact | We recieved feedback and input from the public (anglers) and an increase in questions about our work. |
Description | We have successfully created a pipeline for extracting images from photo-sharing websites and using an image classifier we can correctly identify 95% of images with disease. This represents a new mode of disease surveillance that can be used to rapidly screen 1000's of images. |
Exploitation Route | Anglers and fisheries could use the created pipeline to screen both historic and future images of fish for infection. |
Sectors | Agriculture Food and Drink Healthcare |
Description | Anglers have engaged with us to submit images of fish taken so that we can assess if disease is present. We have engaged with a local angling group in detail, identifying that 8% of the fish caught show signs of disease. |
First Year Of Impact | 2024 |
Sector | Environment |
Impact Types | Societal |
Title | Computer vision pipeline for identifying the presence of disease in images of fish (salmonids) |
Description | We have developed a pipeline for extracting images from photo-sharing websites and using convoluted neural networks to assess the presence of absence of disease from a still image. The efficiency of disease detection has been refined to 95%. |
Type Of Material | Computer model/algorithm |
Year Produced | 2023 |
Provided To Others? | No |
Impact | The outcome is that we can rapidly screen images for the presence of disease where previously no tool was available. |
Description | Engaging with Angling Trust and Anglers |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | Local |
Primary Audience | Professional Practitioners |
Results and Impact | 20 anglers and representatives from the Angling Trust and Rivers Trust were engaged with at an informal workshop at the Water Research Institute at Cardiff University. The project was introduced to anglers and led to production of a webpage for dissemination to anglers. A blog and weblink was shared by the Angling Trust in their UK Newsletter. |
Year(s) Of Engagement Activity | 2023 |