Pneumonia Image Detection

This repository contains a code implementation for detecting pneumonia in X-ray images using deep learning techniques. The code processes the input dataset, generates new images, creates and trains a model to classify the images, and evaluates the performance of the model. The code also supports using a pre-trained InceptionV3 model for transfer learning.

Functions

The following functions are implemented in the code:

• image_exploration: Creates a list of tuples containing class, path, and size for each image in the given path.
• images_creator: Generates a specified number of new images, creating a certain number of images for each image in the input list of tuples.
• image_generator: Generates new images from a given image using data augmentation techniques.
• format_example: Returns an image that is reshaped to IMG_SIZE.
• dataset_creator: Creates two lists of tuples with class and formatted image arrays, one for normal images and one for pneumonia images.
• dataset_loader: Loads image data from a list of tuples and returns two arrays, one for images and one for labels.
• convert_RGB: Converts a grayscale image to an RGB image by duplicating the number of channels and pixel values.

Usage

1. Set the dir_images variable to the directory containing your dataset of X-ray images.

2. Create a list of tuples with class, path, and size for each image in the dataset:

   images_list = image_exploration(dir_images)

3. Generate new images from the dataset:

   images_creator(images_list)

4. Combine the original images and the newly generated images into a single list, and create a dataset:

   ds_list = images_list + new_images_list
   ds_NORMAL, ds_PNEUMO = dataset_creator(ds_list)
   ds_total = ds_NORMAL + ds_PNEUMO

5. Load the dataset and split it into training, validation, and test sets:

   X, y = dataset_loader(ds_total)
   X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=val_size, random_state=42)
   X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=test_size, random_state=42)

6. Train and evaluate the custom model:

   # Train the model
   history = model_PROP.fit(X_train, y_train, validation_data=(X_val, y_val), epochs=EPOCHS, batch_size=BATCH_SIZE)
   
   # Evaluate the model
   accuracy_train = model_PROP.history.history['accuracy'][-1]
   loss0, accuracy_test = model_PROP.evaluate(X_test, y_test)
   
   print(f'ACC. TEST: {accuracy_test} -- ACC. TRAIN: {accuracy_train} ')

7. Train and evaluate the InceptionV3 model:

   # Train the model
   history = model_V3.fit(X_train, y_train, validation_data=(X_val, y_val), epochs=EPOCHS, batch_size=BATCH_SIZE)
   
   # Evaluate the model
   accuracy_train = model_V3.history.history['accuracy'][-1]
   loss0, accuracy_test = model_V3.evaluate(X_test, y_test)
   
   print(f'ACC. TEST: {accuracy_test} -- ACC. TRAIN: {accuracy_train} ')

Dependencies

• TensorFlow
• NumPy
• OpenCV
• Pillow
• Matplotlib
• tqdm
• scikit-learn