search

nyukat / BIRADS_classifier

High-resolution breast cancer screening with multi-view deep convolutional neural networks
https://arxiv.org/abs/1703.07047
BSD 2-Clause "Simplified" License
148 stars 58 forks source link

Image normalization #15

Closed abrichr closed 5 years ago

abrichr commented 5 years ago

Hello,

Thank you for making this code available!

My question is regarding data preprocessing. I noticed that the example images you provide in the images directory have values in the range 0 - 255. Is this how you scaled the training images? Other than the whitening mentioned in the paper (subtracting the mean and dividing by the standard deviation), what other preprocessing did you apply during training and inference (e.g. adaptive histogram equalization)?

Thanks, -Richard

jpatrickpark commented 5 years ago

During training and inference, we directly standardized original 12-bit integer values without rescaling the images to 0-255 range. The images have values typically in the 0-4095 range. We did not apply any preprocessing that is not mentioned in the paper. Depending on the manufacturer, their proprietary preprocessing algorithms may have been applied in the images.

abrichr commented 5 years ago

Thank you for the quick response! Regarding the following:

The images have values typically in the 0-4095 range.

Can you please clarify: does this mean that some images had values outside this range? If so, do you know approximately how many, and in which ranges?

jpatrickpark commented 5 years ago

If I remember correctly, all valid mammograms we have were within the 0-4095 range.