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 :
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:
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).