Hi, We just take a look at the training data set. There are only around 15 patients with breast cancer in the whole 58 subjects, which is really unbalance between normal controls and patients. Could we get more subjects with breast cancer? So that we can make our deep learning algorithms more accurate in prediction. Thanks!

Created by Chao Huang chaohuang
Mohammad: The data is from the general population and therefore reflects the percentage of positives found there. There is no disputing that there are several challenging aspects to this task. The details on the Docker submission are forthcoming. Please stay tuned!
Hi Bruce , correct me if I am wrong - I need to find features from the true positive samples , and according to the wiki almost 99% of the samples will be FP and TN . So we will have only 1% of data that will be TP + FN ; isn't this a little bit odd if we were to train a model using feature selection - and the amount of data from which we are supposed to train the data is less than 1%. Also can you elaborate a little bit of submission via docker container, since it was not mentioned anywhere in the wiki. Mohammad.
Hi Chao: You will definitely get more samples (both positives and negatives) but you will not be able to download them. You will package your algorithm as a Docker container and send it to the challenge submission queue will it will run/train on a large data set. The 58 subject sample is meant simply to give you a sense of the data (the file format, size and various mammography views). Does that help?

More subjects are expected in training data set page is loading…