Have anybody did experiment about the pCR predication on the I-SPY 1 or I-SPY 1 breast cancer dataset?

I was trying to do some experiment about the pCR prediction using the I-SPY 1 or I-SPY 2 DCE MRI dataset. I was using a 3D densenet and T2 post-contrast, but the performance is terrible (less than 70%). Anybody did such experiment on this dataset before? I wonder anything special I should do to achieve a satisfactory result?

Consider using the fast gradient echo data, pre and post contrast as multichannel 3d data

Thank you so much for your help. I tried what you suggest, but still no luck.

Any of a large number of things could be contributing to the poor performance. You are pursuing a wonderful research topic, and as research, experimentation and knowledge of AI is needed. Seek out a local AI expert to join you on your adventure.

Or if you are using MONAI, you should post to their slack channel for support.