search

nyukat / breast_cancer_classifier

Deep Neural Networks Improve Radiologists' Performance in Breast Cancer Screening
https://ieeexplore.ieee.org/document/8861376
GNU Affero General Public License v3.0
840 stars 269 forks source link

More Dataset Information #19

Closed Tonthatj closed 5 years ago

Tonthatj commented 5 years ago

Hi All,

I am currently testing out your suite of models and was hoping to learn more about the dataset on which the models were trained. I have read your paper describing the data in detail, but I could not find the answer to my question in it. I was curious as to what is the statistical distribution of the original dicom images in regards to the Relative X-Ray Exposure as well as the Exposure Index/Sensitivity values found in the dicom tags?

kjgeras commented 5 years ago

Hi Tonthatj,

We didn't extract these from the dicom files. How do you think this could be useful?

Tonthatj commented 5 years ago

I have noticed after reviewing my own data set that your model performs better or worse on different subsets based on different dicom attributes. For example any mammogram with a relative xray exposure > 10000 we noticed a significant drop in the models auc, roughly 30%. The same could be said if the magnification was different from 1. The model also performed variably for the different sensitivity levels/exposure indexes. Would it be possible for you to generate the statistics for the various dicom tags?

kjgeras commented 5 years ago

Very interesting though I think it is to be expected to a large degree. Please send me an email, I will see what we can do.