Classification-of-types-of-Lung-Carcinoma-using-Convolution-Neural-Networks

Project Description

This is a python-based project classifying the type of Lung Carcinoma disease based on deep analysis of histopathological image samples through the application of four popular Convolution Neural Networks, namely GoogLeNet, VGG-16, ResNet-18 and DenseNet-121, one at a time, giving a different Training and Validation accuracy for each epoch.

Dataset description

'Lung and Colon Cancer Histopathological Images' dataset is a publicly available dataset, which contains 25,000 histopathological images with 5 classes. All images are 768 x 768 pixels in size and are in jpeg file format. The images were generated from an original sample of HIPAA compliant and validated sources, consisting of 750 total images of lung tissue (250 benign lung tissue, 250 lung adenocarcinomas, and 250 lung squamous cell carcinomas) and 500 total images of colon tissue (250 benign colon tissue and 250 colon adenocarcinomas) and augmented to 25,000 using the Augmentor package. The dataset is available at:
https://www.kaggle.com/andrewmvd/lung-and-colon-cancer-histopathological-images

Classes of Division

In this project, we have only used the histopathological image samples of human lungs, which have been classified into three categories, namely:

Lung Benign tissue
Lung Adenocarcinoma
Lung Squamous cell carcinoma

Convolution Neural Network models used

Four CNN models have been applied on the dataset, namely:

GoogLeNet
Visual Geometry Group (VGG-16)
ResNet-18
DenseNet-121

Train-Validation Learning Curve

Train-Validation Curve is a popular method to helps us confirm normal behavioural characteristics of model over increasing number of epochs

All the models have been trained over 20 epochs with batch_size of 60

```
GoogleNet
```
```
VGG-16
```
```
ResNet-18
```
```
DenseNet-121
```

Dependencies

Since the entire project is based on Python programming language, it is necessary to have Python installed in the system. It is recommended to use Python with version >=3.6. The Python packages which are in use in this project are matplotlib, numpy, pandas, torch and torchvision. All these dependencies can be installed just by the following command line argument

pip install requirements.txt

Code implementation

Data paths :

Current directory ----> Data_Folder
                            |
                            |
                            |               
                            ------------------>  train
                            |                      |
                            |             --------------------
                            |             |        |         |
                            |             V        V         V
                            |           lung_n  lung_aca  lung_scc
                            |
                            |
                            |              
                            ------------------>   val
                                                   |
                                          --------------------
                                          |        |         |
                                          V        V         V
                                        lung_n  lung_aca  lung_scc

Where the folders train and val contain the folders lung_n, lung_acaand lung_scc, which include the original histopathological images of respective type of lung carcinoma in .jpg/.png format.
Note: The folders lung_n, lung_acaand lung_scc contain the histopathological images of 'Benign', 'Adenocarcinoma' and 'Squamous cell carcinoma' samples respectively.

Training and Evaluating different CNN models :

-help

optional arguments:
  -h, --help            show this help message and exit
  -data DATA_FOLDER, --data_folder DATA_FOLDER
  -cnn CNN_TYPE, --CNN_type CNN_TYPE

Run the following for training and validation :
```
python main.py -data Data_Folder -cnn vgg16
```

Specific tokens :

GoogLeNet: 'googlenet'
VGG-16: 'vgg16'
ResNet-18: 'resnet18'
DenseNet-121: 'densenet121'

agnish13 / classification-of-types-of-lung-carcinoma-using-convolution-neural-networks Goto Github PK