Deep Learning Models for Detecting Neonatal Diseases Using Locally Sourced Data

Abstract

Neonatal diseases are the highest cause of death in Nigeria, second in Africa and number five globally. Despite the giant strides made by the World Health Organization (WHO), Global Burden of Diseases (GBD) and Center for Disease Control, the death of newborns has not reduced significantly, especially in Africa and other developing countries of the world. Most neonatal ailments are treatable and should not have constituted any form of threat to neonates’ lives but for the problem of error in diagnosis. Artificial intelligence (AI) based deep learning technology has been found to help solve several problems that are difficult for humans in almost all fields of life. That is why deep learning algorithms were employed for model development in this study. Data were collected from the medical records of 2750 previous neonatal patients from Federal Medical Center Owo, Ondo State and Ondo State Specialist Hospital, Akure Ondo State Nigeria. This research employed two deep learning algorithms: Artificial Neural Networks (ANN) and the recurrent neural networks-based Long Short-Term Memory (LSTM) to develop predictive models for the detection of neonatal diseases. The models were trained and evaluated. Artificial neural networks ANN recorded an accuracy of 88%, precision of 90%, recall of 89% and F1-score of 88%. The LSTM model was evaluated on the same metrics, and it achieved an accuracy of 89%, precision of 91%, recall of 93% and F1-score of 91%. The results clearly show that LSTM outperformed the ANN model, thus LSTM has a better capacity to detect neonatal diseases than ANN. Therefore, the LSTM model is recommended for detection of neonatal diseases in Nigeria.