Skip to content

Latest commit

 

History

History

ICIAR18_data_IRRCNN_model

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 

ICIAR 2018 BACH Challenge

Dataset

Dataset used is available here

The image dataset is composed of high-resolution (2048 x 1536 pixels), uncompressed, and annotated H&E stain images from the ICIAR 2018 BACH Challenge.

Each image is labeled with one of four classes: i) normal tissue, ii) benign lesion, iii) in situ carcinoma and iv) invasive carcinoma.

Patch Size used : 128 X 128

Image Format : RGB

Pre-Processing : 128 X 128 patches are cropped from the complete images without any overlap. As there is no seperate test dataset 20% of the extracted patches are kept aside for testing and the rest are used for training. Pixel values are normalized before training.

Pixel scale: 0.42 µm x 0.42 µm

Magnification : 200x

Sample Images :

Model Used :

The model used is the IRRCNN. It uses Residual connections in addition to the IRCNN model which adds the inputs at each step to the feature maps extracted by the IRCNN block after passing them through (1,1) convolutional filters to equate their dimensions .

Model architecture:

References :

Alom, Md Zahangir, Mahmudul Hasan, Chris Yakopcic, Tarek M. Taha, and Vijayan K. Asari. "Improved Inception-Residual Convolutional Neural Network for Object Recognition."arXiv preprint arXiv:1712.09888(2017).