Overview / Usage

Neonatal jaundice is a common condition that occurs in newborn infants in the first week of life. Today, the techniques used for detection include blood samples and other clinical testing using special equipment (be more specific). In Algeria, out of 2372 newborns hospitalized, 433 are born with neonatal jaundice.2.07% of these are very dangerous cases. As for the rest of the cases, it is either "physiological" or can be treated with "phototherapy". Lack of attention to this disease will destroy the brain of the newborn by causing “Kernicterus”. This project will assist the doctor’s early detection efforts, in order to predict whether the newborn jaundice is positive by looking at the newborn image. It is trained using an image dataset and it will do the classification depending on skin color. We are creating a model that can assist doctors’ early detection efforts.

Methodology / Approach

In this project we are building a deep learning model , The architecture of the neural_network is CNN layers with fully connected layers, to classify the newborn image by detecting the skin color, the training results are as follows:( The training is done with 50 epochs** )**

• The loss in training (epoch 50) is:0.0068
• The accuracy in training (epoch 10) is: 1
• The validation loss is: 0.78
• The validation accuracy is: 0.83

Technologies Used

Keras , tensorflow , intel OpenVINO, Intel Xeon CPU, Intel AI Analytics Toolkit (Intel OneAPI) , intel optimization Tensorflow , and intel distribution for python

Collaborators