Diabetic-Retinopathy-Detection (Deep Learning and Practice homework 3)

This project is to classify people's degree of diabetes through pictures of retinopathy, which is related to kaggle competition. All the models in this project were built by pytorch.

In addition, please refer to the following report link for detailed report and description of the experimental results.

Hardware

In this project, a total of two different CPUs and GPUs are used for model training.

The first GPU used is NVIDIA GeForce GTX TITAN X, and the second is NVIDIA GeForce RTX 2080 Ti.

	Operating System	CPU	GPU
First machine	Windows 10	Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz	NVIDIA GeForce GTX TITAN X
Second machine	Windows 10	Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz	NVIDIA GeForce RTX 2080 Ti

Requirement

In this work, you can use the following two option to build the environment.

First option (recommend)

$ conda env create -f environment.yml

Second option

$ conda create --name Summer python=3.8 -y
$ conda activate Summer
$ conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch
$ conda install numpy
$ conda install matplotlib -y
$ conda install scikit-learn -y
$ conda install pandas -y
$ pip install torchsummary
$ pip install pkbar
$ pip install tqdm

Model Architecture

In this project, I used the architecture of ResNet 18 and ResNet 50 to classify images.

Due to the skip connection and shortcut connection methods in the ResNet architecture, gradient vanishing and exploding gradient problems are less likely to occur, and a model has lower parameters to train.

skip connection

shortcut connection

Data Description

In this project, the training and testing data were provided by Kaggle competition-Diabetic Retinopathy Detection.

There are a total of 28100 images in the training data and a total of 7026 images in the testing data. In addition, there are a total of 5 categories in the dataset, corresponding to the severity of the five types of diabetes: No, Mild, Moderate, Severe and Proliferative DR.

The data used in this project can be obtained from Google driver, which is called data directory. Just download the folder from Google driver and extract it into the Diabetic-Retinopathy-Detection folder.

Directory Tree

In this project, you can put the folder on the specified path according to the pattern in the following directory tree for training and testing.

├─ ALL_model.py
├─ data
│  ├─ 10003_left.jpeg
│  ├─ 10003_right.jpeg
│  ├─ 10007_left.jpeg
│  ├─ 10007_right.jpeg
│  ├─ 10009_left.jpeg
│  ├─ 10009_right.jpeg
│  ├─ 1000_left.jpeg
│  ├─ ...
│  ├─ ...
│  └─ 99_right.jpeg
├─ dataloader.py
├─ history_csv
│  ├─ ResNet18_nonpretrained.csv
│  ├─ ResNet18_pretrained.csv
│  ├─ ResNet50_nonpretrained.csv
│  └─ ResNet50_pretrained.csv
├─ model_testing.py
├─ model_weight
│  ├─ ResNet18_nonpretrained.rar
│  ├─ ResNet18_pretrained.rar
│  ├─ ResNet50_nonpretrained.rar
│  └─ ResNet50_pretrained.rar
├─ Plot_History_Result.py
├─ ResNet18_nonpretrained_model.py
├─ ResNet18_pretrained_model.py
├─ ResNet50_nonpretrained_model.py
├─ ResNet50_pretrained_model.py
├─ test_img.csv
├─ test_label.csv
├─ train_img.csv
├─ train_label.csv
└─ README.md

Performance Metrics

In this project, Crossentropy is the loss function, and Accuracy is the classification metrics.

Crossentropy

• M: number of classes
• log: the natural log
• y: binary indicator (0 or 1) if class label c is the correct classification for observation o
• p: predicted probability observation o is of class c

---

Accuracy

• True Positive(TP) signifies how many positive class samples your model predicted correctly.
• True Negative(TN) signifies how many negative class samples your model predicted correctly.
• False Positive(FP) signifies how many negative class samples your model predicted incorrectly. This factor represents Type-I error in statistical nomenclature. This error positioning in the confusion matrix depends on the choice of the null hypothesis.
• False Negative(FN) signifies how many positive class samples your model predicted incorrectly. This factor represents Type-II error in statistical nomenclature. This error positioning in the confusion matrix also depends on the choice of the null hypothesis.

Training

A total of four methods are provided in this project to train the model, corresponding to the four files ResNet18_nonpretrained_model.py, ResNet18_pretrained_model.py, ResNet50_nonpretrained_model.py and ResNet50_pretrained_model.py.

You can get some detailed introduction and experimental results in this link.

You can config the training parameters through the following argparse, and use the following instructions to train different model.

parser = argparse.ArgumentParser()
parser.add_argument('--epochs', type=int, default='10', help='training epochs')
parser.add_argument('--image_size', type=int, default='224', help='model input image size')
parser.add_argument('--n_channels', type=int, default='3', help='model input image channels')
parser.add_argument('--train_batch_size', type=int, default='256', help='batch size to training')
parser.add_argument('--test_batch_size', type=int, default='281', help='batch size to testing')
parser.add_argument('--number_worker', type=int, default='4', help='number worker')
parser.add_argument('--learning_rate', type=float, default='5e-3', help='learning rate')
parser.add_argument('--save_model', action='store_true', help='check if you want to save the model.')
parser.add_argument('--save_csv', action='store_true', help='check if you want to save the training history.')
opt = parser.parse_args()

ResNet18 pretrained model

python ResNet18_pretrained_model.py --epochs 10 --save_model --save_csv

ResNet50 pretrained model

python ResNet50_pretrained_model.py --epochs 10 --save_model --save_csv

ResNet18 non-pretrained model

python ResNet18_nonpretrained_model.py --epochs 5 --save_model --save_csv

ResNet50 non-pretrained model

python ResNet50_nonpretrained_model.py --epochs 5 --save_model --save_csv

Testing

You can display the testing results in different models by using the following commands which contains pretrained and non-pretrained models. The model weight could be downloaded from the link.

The detailed experimental result are in the link.

Then you will get the best result like this, each of the values were the testing accuracy.

	Pretrained	None-Pretrained
ResNet18	0.758281	0.705448
ResNet50	0.774843	0.702573

Reference

• https://www.kaggle.com/c/diabetic-retinopathy-detection#description
• https://arxiv.org/pdf/1712.09913.pdf
• https://ieeexplore.ieee.org/document/7780459
• https://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html#sphx-glr-auto-examples-model-selection-plot-confusion-matrix-py