Citation

Karim, Razuan (2023) Smart system for neonatal jaundice detection using image processing. PhD thesis, Multimedia University.

Full text not available from this repository.

Abstract

Neonatal jaundice is a common disease that affects newborn babies during their first week of life and requires medical treatment. The yellowing of the skin and the coloring of the eyes are also symptoms of hyperbilirubinemia, which is a condition in which there is an excessive amount of bilirubin in the blood. In order to diagnose jaundice, medical professionals need to take blood samples and carry out additional diagnostic procedures using specialized apparatus, which is an intrusive process. Visual assessment is currently unreliable, and the methods employed for detection, which require blood samples and include more precise procedures, are both expensive and time-consuming. This study aims to develop a non-invasive method for regularly detecting and monitoring jaundice in order to estimate the level of bilirubin, and it will also aid medical experts in establishing an accurate early diagnosis. For this purpose, a smartphone application can be used as an alternate method to detect and monitor neonatal jaundice. The Bangabandhu Memorial Hospital in Chittagong, Bangladesh, as well as Google, are the two primary sources for the datasets. The predictive analytical model for the classification of the levels of jaundice is planned based on the skin color of a total 800 neonates’ image that are included in the dataset. These images were divided by their yellow skin color, and 80% of the data from the dataset is used for training while the remaining 20% is used for testing with validation. In this research, a new screening technique for neonatal jaundice detection is proposed to exploit yellow discoloration in the skin. For the identification and classification of jaundice in this study, a deep learning technique with fine tuning was used. For training, the CNN model was utilized. To validate the correctness of the model, numerous crucial measures such as F1-score, precision, recall, specificity, confusion matrices and loss graphs were calculated. The achieved accuracies of F1-score are 0.99%, precision is 100%, recall is 0.99%, and specificity is 100%. For the purpose of displaying feature maps that represent the decomposition of an input image into distinct filters, intermediate activations are visualized. Various optimizers were utilized to minimize errors. Finally, the model was evaluated using the test dataset, and the results showed that the suggested CNN model had an accuracy level of 99.38%.

Item Type:	Thesis (PhD)
Additional Information:	Call No.: RJ276 .R39 2023
Uncontrolled Keywords:	Jaundice, Neonatal —Imaging
Subjects:	R Medicine > RJ Pediatrics
Divisions:	Faculty of Engineering (FOE)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	01 Oct 2025 08:22
Last Modified:	01 Oct 2025 08:22
URII:	http://shdl.mmu.edu.my/id/eprint/14653

Downloads

Edit (login required)