Smartphone based clinical decision support system for early detection of neonatal jaundice using artificial intelligence with temporal convolutional network
Machine learning study demonstrates improved noninvasive jaundice classification in neonates using smartphone images, indicating strong diagnostic support utility.
Key Points
To develop a smartphone-based computer vision decision support system that optimizes deep learning models for accurate, noninvasive early detection of neonatal jaundice.
Designed the AEJDN-SCVGJO framework incorporating an adaptive median filter (AMF) to remove image noise and enhance smartphone photograph quality.
Extracted image features using a squeeze-and-excitation DenseNet (SE-DenseNet) architecture.
Classified jaundice status using a temporal convolutional network (TCN) tuned via the Golden Jackal Optimizer (GJO) algorithm.
Comparative experimental evaluations revealed that the AEJDN-SCVGJO system outperformed baseline image classification approaches.
Hyperparameter optimization through the Golden Jackal Optimizer effectively improved diagnostic feature selection and classification accuracy.