Diagnostic Classification Of Lung Nodules Using 3D Neural Networks

Network Architecture for the ISBI_2018 paper : DIAGNOSTIC CLASSIFICATION OF LUNG NODULES USING 3D NEURAL NETWORKS To view the paper on Archive click the following https://arxiv.org/abs/1803.07192 Visit My website - .... to be updated shortly for intuition, hints etc

Built With/Things Needed to implement experiments

Python - Python-2
Keras - Deep Learning Framework used
Numpy - Numpy
Sklearn - Scipy/Sklearn/Scikit-learn
CUDA - CUDA-8
CUDNN - CUDNN-5 You have to register to get access to CUDNN
LIDC-IDRI - LIDC-IDRI-dataset download
Itk-Snap - To, view the CT images

The code in this repository provides only the stand alone code for this architecture. You may implement it as is, or convert it into modular structure
if you so wish. The dataset of OASIS can obtained from the link above and the preprocessiong steps involved are mentioned in the paper. 
You have to provide the inputs.

Some guidelines to run the network (Please follow the paper for exact details)

The following lines provide a very simple example of how you may train the network provided in this repository and obtain predictions.

We are not providing the complete implementation (Dataset/Preprocessing/Calculation of ROC/ACC/ etc etc) but just the core network architecture.

You may include validation set and callback function as demonstrated, however please note we never used those. In the following lines the network is referred as "finalmodel" You may apply early stopping if desired, but for our experiments we didnt.

The following line demonstrated how to initialize the early stopping, Set patience to whichever epochs you wish for the network to continue once performance starts to decline. The network stops once patience+1 epochs have been reached without any improvement from the last stored best result. Counter is reset if a better performance is acheived at any stage of the countdown.

xyz=keras.callbacks.EarlyStopping(monitor='val_loss', patience=4, verbose=1, mode='auto')

The inputs are X1_train,X2_train. Use Keras's model.fit function as follows to train

finalmodel.fit([X1_train,X2_train], [y_train,y_train,y_train,y_train,y_train,y_train,y_train,y_train,y_train],  
                batch_size=8, 
                nb_epoch=150,
                validation_data=([X1_validate,X2_validate],[y_validate,y_validate,y_validate,y_validate,y_validate,y_validate,y_validate,y_validate,y_validate]), 
                shuffle=True,
                 callbacks=[xyz], 
                class_weight=class_weightt)
#Once again please note we never used any calls backs or validation set in our experiments.
#I used a batch size of 8.
#Compared to batch size of 2-16, 8 gave the best result, higher batch size couldnt be supported on our GPU's.
#Please read the paper for more details (Link to be provided soon.)

You can obtain predicitions of the different outputs using the keras's model.predict function ####Set the index variable from 0-8 to obtain corresponding outputs with 8 being the final output

predict_x=finalmodel.predict([X1_test,X2_test],batch_size=8)[index]

Contribution

All the code were written by the authors of this paper. Please contact (raun- rd31879@uga.edu) for questions and queries and more details if you are having trouble with the complete implimentation.

Please cite our paper if this code or the idea of multioutput or if this study of pulmonary nodules helps your work

Our paper -

@inproceedings{dey2018diagnostic, title={Diagnostic classification of lung nodules using 3D neural networks}, author={Dey, Raunak and Lu, Zhongjie and Hong, Yi}, booktitle={Biomedical Imaging (ISBI 2018), 2018 IEEE 15th International Symposium on}, pages={774--778}, year={2018}, organization={IEEE} }