ashishsardana / jaundice-detection

Neonatal Jaundice Detection using Convolutional Neural Networks

Make-A-Thon 2017

Neonatal Jaundice Detection

A way to alert a mother in home about early stage of Neonatal Jaundice in order to get an early treatment to prevent newborn mortality/morbidity.

Solution:

The most reliable and non-invasive way to detect jaundice is to check the skin tone of a person.

Adopting the same technique for a new born baby, identifying the color of baby's face will help in detecting jaundice.

• CNN to classify an image for jaundice/non-jaundice. Dataset used for this were images collected from google.
• Uniform Image - Resizing the images to 28x28 px.
• Face Detection - To use the baby's face only for identifying the color of the skin. (Using Haar-Cascade and OpenCV)
• Skin Detection - To detect just the skin of the face and ignoring eyes, hair and other unnecessary features.
• Obtaining Color - Extracting color only from the skin region. (Using ColorThief)
• Classification - Comparing the color value with different shades of yellow.

1. Click the picture of the baby's face using Raspberry Pi's camera.
2. The image will be processed and the corresponding LED (Jaundice/Non-Jaundice) glows.

jaundice-detection's Issues

Where can i find dataset?