Artificial intelligence in the differentiation of tropical infections: A step ahead
Smita Shenoy, Girish Thunga, S. Nair, K. Vijayanarayana, M. Varma, M. Rashid · International Journal of Infectious Diseases · 2020
Background: Tropical infectious diseases became an unavoidable major concern of today's world with a similarly presenting symptomatology and laboratory evaluation which makes difficult the early detection and distinct separation of infection. There is a need for the development of a tool that makes optimal use of distinct symptomatology amongst these infections and distinguishable laboratory parameters. So, we aimed to develop a decision-making tool for differentiating tropical infections such as Dengue, Malaria, Leptospirosis and Scrub typhus in a tertiary care hospital. Methods and materials: The study was conducted in Kasturba Hospital, Manipal after obtaining the ethical approval. A 9-item validated questionnaire was used to assess the need of physicians for the development of the decision-making tool and a case report form was designed accordingly. A total of 800 cases were collected with 200 cases each under individual disease category (Dengue, Malaria, Leptospirosis and Scrub typhus). The data set was divided into two portions for performing multinomial logistic regression analysis i.e., 170 x 4 cases training set and 30 x 4 cases validation set. Multinomial regression analysis was performed on the dataset followed by machine learning models are applied on the same dataset. In case of machine learning models, both multi-classification and binary classification models are applied accordingly. Results: The multinomial logistic regression analysis where dengue is taken as the reference category revealed a 60.7%, 62.5% and 66% of predictability for dengue, malaria and leptospirosis respectively which were relatively good. Whereas, scrub typhus showed only 39.5% of predictability score. Similarly, the multi-classification machine learning model showed 55–60% of predictability. On the other hand, binary classification machine learning algorithms showed a higher predictability score where 79–84% was observed with one Vs rest of disease and 69–88% was observed with one Vs one disease category, respectively. Conclusion: As of our knowledge, this is the first study on the application of machine learning in disease detection. Our study strongly recommends that machine learning techniques can be a very good tool in predicting tropical diseases and aid in early detection and better patient care.