Interpretability of lung nodules segmentation
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Ali Rasekh 2ebb762ec2 fix: first change 7 months ago
interpretability_deep_lift first commit 7 months ago
other first commit 7 months ago
refactored first commit 7 months ago
Readme fix: first change 7 months ago

Readme

There is a folder called interpretability_deep_lift in the Tools folder. This folder has three main scripts. 


The first script which main_code has the implementation fo the model and training of that and some results of it. But next scripts mostly focus on interpretation of the model by different methods and teqniches.
The second one model script in this folder after traning the model will test different interpretability models on it and check the results. These different configs include deep lift, grad-cam, etc. It will check the effects of borders or other parts of the lesions as well.

Its worth noting that different channels and layers of the network will be considered as well.

The third script will load pre-train models and test different interpretability models with the same previous situations. It will calculate the infedility of model by deepLift and grad-cam as well and compare them with the calculated scores.


Besides there are two other scripts in the other folder which contains our implementation on some other areas. For example an implementation is done on skin dataset to check the results. However these datasets are not the main one which we focused on them.

Also there is another folder which is called preprocess. In this folder the preprocessing scripts that we used are saved. Our main dataset that we used for the project is LIDC. In order to use LIDC we needed to use pylidc library of python and this library should be installed on the system and environment that we are running it. The dataset path should be proper dataset and relative to the code that is going to run. After running the script it will load the dataset and process all images and save them in a format of 2D slices.

Our dataset has multiple labels both for classification and segmentation. For classification there are nine various leables. After preprocessing the segmentation masks and labels will be saved in a folder and all 2d slices of images will be saved in another folder. The class number of each nodule of lung will be saved in another excel file which shows other labels. Obviously these labels can be used for further ideas.