Multi-channel-DeepLiverNet2.0

Liver project by Redha Ali

This is the official implementation of our proposed DeepLiverNet2.0:

Tensorflow implementation of Swin Transformer model. Based on Official Tensorflow implementation.

Requirements

• tensorflow >= 2.6.0
• scikit-image==0.17.2
• scipy==1.7.1
• numpy
• matplotlib==3.5.1
• os==2.1.4
• MATLAB 2021a or later or use environment.yml

Pretrained Swin Transformer Checkpoints

ImageNet-1K and ImageNet-22K Pretrained Checkpoints

name	pretrain	resolution	acc@1	#params	model
swin_base_224	ImageNet-22K	224x224	85.2	88M	github

Data Preparation

To extract the 11 sileces please run the script "Create_11_Slices_dataset.m" file

Deep Features Extraction

To extract the features from the 11 sliecse using pre-trained Swin transfomer please run the script "run_Swin_Fea_Ext.py"

Ten Fold Cross Validation Classification

TensorFlow: To train and test the Multi-channel-DeepLiverNet2.0 please use the script "train.py" and "test.py" MATLAB: To train the Multi-channel-DeepLiverNet2.0 please run the script "run_train_10k_Fold_CV_classification.m"

License

The Multi-channel-DeepLiverNet2.0 model is released under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International Public License (CC BY-NC-ND 4.0).

Copyright © 2024 Cincinnati Children's Hospital Medical Center