X-Ray Image Classification for Pneumonia Detection

Chest X-Ray image classification and prediction using Convolutional Neural Networks (CNNs) in Tensorflow

                                        

Background

The X-Ray image classifier is a seminal work that is capable of detecting whether or not a patient has pneumonia using the patient's lung X-rays. This model uses transfer learning techniques, encircling the DenseNet201 architecture, and performed well in predicting X-ray images correctly with a validation accuracy of 95.19 % .However, this cannot be used as a full proof method to detect pneumonia, as the symptoms of the patient are not taken into account.

Download Dependencies

pip install numpy
pip install tensorflow
pip install matplotlib
pip install os

Code

1. Importing Dependancies

import numpy as np
import tensorflow as tf
from tensorflow.keras.applications.densenet import DenseNet201
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import matplotlib.pyplot as plt
import os

2. Getting path of the images stored in google drive

train_norm_dir = os.path.join('/content/drive/My Drive/chest-xray-pneumonia/train/NORMAL/')
train_pneu_dir = os.path.join('/content/drive/My Drive/chest-xray-pneumonia/train/PNEUMONIA/')

val_norm_dir = os.path.join('/content/drive/My Drive/chest-xray-pneumonia/test/NORMAL/')
val_pneu_dir = os.path.join('/content/drive/My Drive/chest-xray-pneumonia/test/PNEUMONIA/')

2.1. Getting names of the images stored in google drive

train_norm_names = os.listdir(train_norm_dir)
train_pneu_names = os.listdir(train_pneu_dir)

val_norm_names = os.listdir(val_norm_dir)
val_pneu_names = os.listdir(val_pneu_dir)

3. Define the model

pre_model= DenseNet201(include_top=False,
                             weights='imagenet',
                             input_shape=(224,224,3))
for layer in pre_model.layers:
  layer.trainable=True

last_output = pre_model.output

x = tf.keras.layers.Flatten()(last_output)
x = tf.keras.layers.Dense(256,activation='relu')(x)
x = tf.keras.layers.Dropout(0.5)(x)
x = tf.keras.layers.Dense(2,activation='softmax')(x)

model = tf.keras.Model(pre_model.input,x)

4. Compile the model

opt = tf.keras.optimizers.Adam(learning_rate=0.0001)
model.compile(optimizer=opt,loss='categorical_crossentropy',metrics=['accuracy'])

5. Define the ImageDataGenerators and perform Image Augmentation

train_datagen = ImageDataGenerator(rescale=1.0/255.,
                                   rotation_range=20,
                                   brightness_range=[0.7,1.2],
                                   zoom_range=0.2
                                  )
val_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory('/content/drive/My Drive/chest-xray-pneumonia/train',
                                                    batch_size = 16,
                                                    class_mode='categorical',
                                                    target_size=(224,224))

val_generator = val_datagen.flow_from_directory('/content/drive/My Drive/chest-xray-pneumonia/test',
                                                    batch_size = 16,
                                                    class_mode='categorical',
                                                    target_size=(224,224))

6. Train the model

history = model.fit(train_generator,
                    validation_data=val_generator,
                    steps_per_epoch=326,
                    validation_steps=val_generator.samples//val_generator.batch_size,
                    epochs=25)            

6. Evaluate the network

print("[INFO] evaluating network...")
predIdxs = model.predict(tsX, batch_size=16)

predIdxs = np.argmax(predIdxs, axis=1)

print(classification_report(tsY.argmax(axis=1), predIdxs,
	target_names=lb.classes_))

How to use

Step 1 : Create folders in GDrive as

     Main Folder -> train ->PNEUMONIA
                          -> NORMAL
                 -> test ->PNEUMONIA
                          -> NORMAL

Step 2 : Mount drive in google colab

Step 3 : Run all the blocks of Code

Results