Convolutional Neural Networks-Based Approach to Detect Neonatal Respiratory System Anomalies with Limited Thermal Image

Abstract

Respiratory system diseases in neonates are thought-about major causes of neonatal morbidity and mortality, particularly in developing countries Early diagnosis and management of these diseases is very important. Thermal imaging stands out as a harmless non-ionizing method, and monitoring of temperature changes or thermal symmetry is used as a diagnostic tool in medicine. This study aims to detect respiratory abnormalities of neonates by artificial intelligence using limited thermal image. Convolutional neural network (CNN) models, although a powerful classification tool, require a balanced and large amount of data. The conditions that require the attention of infants in neonatal intensive care units make medical imaging difficult. It may not always be possible to have much data in the neonatal thermal image database, as in some real-world problems. To overcome this, an effective deep learning model and various data enhancement techniques were used and their effects on the classification results were observed. Neonates with respiratory abnormalities were evaluated in one class, with cardiovascular diseases and abdominal abnormalities were evaluated in the other class. As a result, when the number of images is increased by 4 times with data augmentation, it was determined that the classification accuracy increased from 84.5% to 90.9%.

The past, current, and future of neonatal intensive care units with artificial intelligence: a systematic review

“…DL algorithms have been used to classify neonatal diseases from thermal images 189 – 192 . These studies analyzed neonatal thermograms to determine the health status of infants and achieved good AUC scores 189 – 192 . However, these studies didn’t include any clinical information (Table 6 ).…”

Neurosonographic Classification in Premature Infants Receiving Omega-3 Supplementation Using Convolutional Neural Networks

“…CNNs are crucial in medicine because they can identify complex patterns in images and data, significantly improving the accuracy and speed of diagnosis. Their ability to automate and enhance the analysis of medical images contributes to the identification of subtle changes that might go unnoticed with traditional methods, thus enabling timely and effective patient treatment [12][13][14][15][16]32].…”

“…Furthermore, S. Ervural and M. Ceylan used a CNN to assess temperature and thermal symmetry, which is crucial for monitoring health status and predicting potential risks in newborns [14]. Additionally, S.F.…”