A QUALITATIVE LITERATURE REVIEW OF MACHINE LEARNING TECHNIQUES USED FOR DIAGNOSIS OF NEONATAL DISEASES IN HEALTHCARE

Abstract

Background:There has been signi\uf001cant growth in the use of Arti\uf001cial Intelligence (AI) for healthcare in the last decade.\nAim: To identify effective AI techniques for the prediction & diagnosis of neonatal diseases and preventive measures & treatment plan for them.\nNeonates are newborn babies less than a month old.\nMethods:Research papers published in databases like IEEE Xplore, Medline, PUBMED and Elsevier were searched to \uf001nd publications reporting\nthe application of AI for the prediction and prevention of neonatal diseases. The overall search strategy was to retrieve articles that included terms\nthat were related to “NICU”, “Arti\uf001cial Intelligence”, “Neonatal diseases” and “Healthcare”.\nResults: Hundreds of papers were identi\uf001ed in initial search, out of which 13 publications met the evaluation criteria of related terms inclusion, AI\nfor Neonatal Diseases in particular. These papers described application of AI techniques in neonatal healthcare for disease detection and were\nsummarized for \uf001nal analysis. Most of the papers are focused on using supervised machine learning techniques for the prediction of diseases.\nVarious other approaches in AI techniques used in neonatal disease diagnosis have been tested for related \uf001ndings, factors, methods, to address and\ndocument performance metrics. The comparative analysis of ML model evaluation parameters like AUC (Area under Curve), Speci\uf001city,\nSensitivity, True Positive and False-negative Rates was done to develop the scope for improving performance of AI/MLtechniques.\nConclusion: The systematic study and review of different AI techniques such as supervised machine learning; arti\uf001cial neural networks, data\nmining techniques used for neonatal disease diagnosis highlighted their role in disease prediction, management, and treatment plan. More studies\nare needed to improve the use of AI for timely prediction of neonatal diseases like respiratory distress syndrome, sepsis for increasing the survival\nchances in preterm or normal neonates. The supervised learning models like Support Vector Machines(SVM), Decision Trees, K nearest neighbors\nare found to be effective for neonatal disease detection and will be applied in future research.