AI Shows Promise in Swine Medicine: Comparing AI and Human Evaluators in Detecting Lung Lesions

Researcher compares AI, human evaluators in swine medicine

In their recently published study, the team, led by Dr. Robert Valeris-Chacin, an assistant professor at VERO in the Texas A&M College of Veterinary Medicine & Biomedical Sciences' (VMBS) Department of Veterinary Pathobiology, assessed the capabilities of an AI to detect lesions in pig lungs, which can be a sign of pneumonia-causing bacteria. The team found that while the AI is not yet as accurate as a veterinarian evaluator, it has behaviors that are very similar to a person. Particularly in European food animal production, it is common for vaccine manufacturers to send veterinarians to the processing plants to monitor the success rates of their vaccines, such as those that prevent respiratory disease. "Veterinarian evaluators provide important technical assistance in food production," Valeris-Chacin said. "But it requires a highly trained individual to detect lungs with bacterial pneumonia. One of our three goals was to test the accuracy of an AI to see if it can increase the efficiency and accuracy of this process." Their other two goals included measuring the agreement and consistency of expert evaluators and comparing them to the AI, understanding that some conditions of the study would be different from real life, where veterinarians in the field can also touch the lungs to aid in the detection of pneumonia. "In our study, we asked our experts to evaluate a series of hundreds of images, but we repeated some images to see if the experts would score them the same way each time," Valeris-Chacin said. "What we learned is that human evaluators were very consistent as individuals -- compared to each other, the evaluators disagreed somewhat often, but the same evaluator was very likely to score repeat images the same way. "What's exciting is that the AI also had perfect consistency, even though multiple people were involved in its training," he said. "The company behind this AI wanted to create an AI that would mimic the way human evaluators score the lungs, and the AI is very promising in this regard."

Researcher compares AI with human evaluators in swine medicine

A Texas A&M Veterinary Education, Research, & Outreach (VERO) program-led research team is studying whether artificial intelligence (AI) could play a supportive role in the evaluation of respiratory disease in pigs. In their recently published study, the team, led by Dr. Robert Valeris-Chacin, an assistant professor at VERO in the Texas A&M College of Veterinary Medicine & Biomedical Sciences' (VMBS) Department of Veterinary Pathobiology, assessed the capabilities of an AI to detect lesions in pig lungs, which can be a sign of pneumonia-causing bacteria. The team found that while the AI is not yet as accurate as a veterinary evaluator, it has behaviors that are very similar to a person's. The work is published in the journal Veterinary Research. Particularly in European food animal production, it is common for vaccine manufacturers to send veterinarians to the processing plants to monitor the success rates of their vaccines, such as those that prevent respiratory disease. "Veterinarian evaluators provide important technical assistance in food production," Valeris-Chacin said. "But it requires a highly trained individual to detect lungs with bacterial pneumonia. One of our three goals was to test the accuracy of an AI to see if it can increase the efficiency and accuracy of this process." Their other two goals included measuring the agreement and consistency of expert evaluators and comparing them to the AI, understanding that some conditions of the study would be different from real life, where veterinarians in the field can also touch the lungs to aid in the detection of pneumonia. "In our study, we asked our experts to evaluate a series of hundreds of images, but we repeated some images to see if the experts would score them the same way each time," Valeris-Chacin said. "What we learned is that human evaluators were very consistent as individuals -- compared to each other, the evaluators disagreed somewhat often, but the same evaluator was very likely to score repeat images the same way. "What's exciting is that the AI also had perfect consistency, even though multiple people were involved in its training," he said. "The company behind this AI wanted to create an AI that would mimic the way human evaluators score the lungs, and the AI is very promising in this regard."