search

vinbigdata-medical / midl2020-cnnlstm-ich

CNN-LSTM for intracranial hemorrhage detection
Apache License 2.0
38 stars 7 forks source link

train on CQ500 #4

Open AI678 opened 3 years ago

AI678 commented 3 years ago

Thanks for sharing your great work. The paper claimed that cnn-lstm model was trained on Kaggle dataset and further tested on the CQ500 dataset. Can the code provided in this repository be used to train a cnn-lstm model using train data from CQ500 ?

nhannguyen2709 commented 3 years ago

Hi @AI678, You can train with any brain-ct classification dataset following the same data-preprocessing + modeling pipeline from this repo e.g. 3-windowed RGB-like images + CNN-LSTM architecture. CQ500 was originally created by the Qure.ai team (http://headctstudy.qure.ai/), which can be downloaded, contained only 500 studies. The larger training set (Qure25k) is not publicly available.

500 studies is a small number so I advise you against training models solely on the CQ500 dataset.

sureshsivakumar commented 3 years ago

@nhannguyen2709 can you please share the Hardware requirement and the package requirements to execute this project for learning purpose ?