Team 32 : Chest X-Ray disease classification Project Members: Adiki Madhu Babu, Mansi Arora, Srinivas Sekar

Note: Use python3.6

Instructions to run each file are given below

1. EDA.ipynb : This jupyter notebook takes in the raw csv files containing info about image labels, blacklisted images and outputs two pickle objects containing training and testing images' names and labels.

            INPUT : Make sure you have the following Folder structure and files
            1. ../data/Data_Entry_2017.csv
            2. ../data/BBox_List_2017.csv
            3. ../data/blacklist_non_PA_AP_view.csv
            4. ../data/blacklist_rotated_images.csv
            5. ../data/blacklist_other_images_with_lower_quality.csv
            6. ../data/train_val_list.txt
            7. ../data/test_list.txt
   
            OUTPUT:
            1. train_val_filtered.pkl
            2. test_filtered.pkl
   

2. torch_V100_423.py : The main training file. The code is written in Pytorch and assuumes availability of GPU and cuda().

            INPUT:
            1. imagepath = '../data/images/' : Make sure you have all your
                            X-ray images in this folder structure
            2. densenet_model.pkl file*
            3. train_val_filtered.pkl
            4. test_filtered.pkl : Both this files were generated by the running
                                    EDA.ipynb
   
            * to generate this file, follow the undermentioned instructions
            model = get_symbol() // Need internet connection for this
            with open('densenet_model.pkl', 'wb') as f:
                pickle.dump(model, f)
   
            OUTPUT:
            1. Model_all_50_4-23-18.model : The trained model
            2. valid_loss_all_50_4-23-18.csv : CSV files with losses at each
                epoch saved
   

3. Trying_CAM_final.ipynb : The file takes the trained model and performs CAM localization on a given image

                INPUT:
                1. Model_65_XX_XX-XX_XX.model* : The trained model generated by
                    torch_V100_423.py
                2. path = "images/"
                3. images/train_val_filtered.pkl
                4. images/test_filtered.pkl
                5. All images in images/
   
                OUTPUT:
                1. CAM.jpg : a localized image
                2. test.jpg : saves the randomly selected image
   
                * Make sure the name matches the output model's name of
                    torch_V100_423.py
   

4. out_preds.py: Reads the test images and outputs prediction probabilities and labels in csv format (Used to plot ROC curves)

                INPUT:
                1. ../data/images/ : ALl images go here
                2. train_val_filtered.pkl
                3. test_filtered.pkl
                4. Model_65_XX_XX-XX_XX.model
   
                OUTPUT:
                1. out_preds.csv: predictions for test images
                2. out_gt.csv: labels for test images
   

5. plot_ROC_curve.ipynb: Given predictions and true labels, plots ROC curve

                INPUT:
                1. out_preds.csv
                2. out_gt.csv
   
                OUTPUT:
                1. Inline plot of ROC curve for all diseases
   

6. api.py : It renders the index.html file in the templates/ folder

        RUN : python api.py
   
        INPUT:
        1. Model_65_XX_XX-XX_XX.model
        2. templates/index.html
   

7. Team32ChestXRayPPT.ppsx: recorded and narrated ppt

8. Team32ChestXRayPPT.pdf : Final report