(Reuters Health) - Artificial intelligence may help clinicians tell the difference between different tropical diseases, a new study suggests.
Researchers used data from 800 patients to develop a machine learning tool that may aid in early discrimination of four tropical infections: malaria, scrub typhus, leptospirosis and dengue, according to the report published in PLOS Neglected Tropical Diseases.
"Early diagnosis of tropical infections helps in further improving patient care," said the research team from Manipal, India. "Our study is the first of its kind where both machine learning and statistical techniques were applied to develop a model in tropical infectious diseases which need to be studied further in implementation level."
The authors did not respond to requests for comment.
To determine whether AI could help clinicians diagnose specific tropical diseases, the researchers designed a two-phase study: first they designed a nine-item self-administered questionnaire, and then they developed a prediction tool with a decision tree involving a multinomial regression analysis and machine learning algorithm.
The researchers recruited 800 patients, with 200 per disease, and used data from these patients to home in on 13 variables that might help in differentiating between tropical infections.
Based on the clinical presentation of the four tropical infections, the researchers performed a multinomial regression analysis to identify the significant factors in identifying diseases. They identified nine important variables which were categorized on the basis of arbitrary cut offs "as myalgia (present or absent), arthralgia (present or absent), abdominal pain (present or absent), urine output (normal or decreased), total bilirubin (0-1.6; 1.6-3.2; 3.2mg/dl and above), sodium level (100-140ml; 140ml and above), albumin level (0-3.4; 3.5mg/dl and above), RBC (1,500,000-3,500,000; 3,600,000-4,500,000; 4,600,000-5,500,000; 5,600,000 cells/cumm and above), lymphocytes (10-20; 21-40; 40 cells/cumm and above), hematocrit (25-35; 35-45; 45-55%), platelets (5000-50,000; 50,000-100,000; 100,000-150,000; 150,000-450,000; 450,000 cells/cumm and above) and erythrocyte sedimentation rate (0-22; 22-30; 30 mm/hr and above)."
The researchers used the WEKA machine learning tool to test binary (one disease at a time) and multi-class (all four diseases). In the multi-class case the researchers found an average accuracy of 55-60%. The binary approach, in which one disease was compared to all the rest, led to an estimated accuracy of 79-84%.
AI for diagnosing tropical diseases has a ways to go, said Dr. Peter Hotez, dean of the National School of Tropical Medicine and a professor in the departments of pediatrics, molecular virology and microbiology at the Baylor College of Medicine in Houston.
"I think AI is still at an early stage for all aspects of clinical diagnosis, including tropical diseases," Dr. Hotez said. "However the technology is improving and increasingly can be considered as a useful adjunct. This might be especially true given the dearth in training opportunities for tropical diseases, something we are working to address at our very unique BCM National School of Tropical Medicine."
SOURCE: PLOS Neglected Tropical Diseases, online June 30. 2022.
Reuters Health Information © 2022
