breast_eff_net_v2_b2.py

#Full research work coded and implemented by Sheekar Banerjee. The runtime was partially tested by Md. Kamrul Hasan Monir

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import glob, os

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, activations, optimizers, losses, metrics, initializers
from tensorflow.keras.preprocessing import image, image_dataset_from_directory
from tensorflow.keras.applications import MobileNetV3Small, MobileNet, InceptionV3
from tensorflow.keras.applications.efficientnet import preprocess_input, decode_predictions

seed = 42
tf.random.set_seed(seed)
np.random.seed(seed)

dir_path = '/content/drive/MyDrive/Breast Cancer/breast_cancer_data'
IMAGE_SHAPE = (224, 224)

# create prepare_image method
# used to preprocess the image for efficientNet model
def prepare_image(file):
    img = image.load_img(file, target_size=IMAGE_SHAPE)
    img_array = image.img_to_array(img)
    return tf.keras.applications.efficientnet.preprocess_input (img_array)

directories = os.listdir(dir_path) # read the folders

files = [] # save all images for each folder
labels = [] # set for each image the name of it

# read files for each directory
for folder in directories:

    fileList = glob.glob(dir_path + '/'+ folder + '/*')
    labels.extend([folder for l in fileList])
    files.extend(fileList)

len(files), len(labels)

# create two lists to hold only non-mask images and label for each one
selected_files = []
selected_labels = []

for file, label in zip(files, labels):
    if 'mask' not in file:
        selected_files.append(file)
        selected_labels.append(label)


len(selected_files), len(selected_labels)

# the dictionary holds list of images and for each one has its target/label
images = {
    'image': [],
    'target': []
}

print('Preparing the image...')

for i, (file, label) in enumerate(zip(selected_files, selected_labels)):
    images['image'].append(prepare_image(file))
    images['target'].append(label)

print('Finished.')

# convert lists to arrays
images['image'] = np.array(images['image'])
images['target'] = np.array(images['target'])

# encode the target
from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()

images['target'] = le.fit_transform(images['target'])

classes = le.classes_ # get the classes for each target
print(f'the target classes are: {classes}')

from sklearn.model_selection import train_test_split

x_train, x_test, y_train, y_test = train_test_split(images['image'], images['target'], test_size=.10)

x_train.shape, x_test.shape, y_train.shape, y_test.shape

from keras.src.applications import EfficientNetV2B2
#Build Model

from keras.applications.efficientnet import EfficientNetV2B2

base_model = EfficientNetV2B2(
    include_top=False,
    weights='imagenet',
    input_shape=(*IMAGE_SHAPE, 3),
    classes=3)

# Freeze the base_model
base_model.trainable = False

# append my own layers on the top of the model for Transfer Learning
x = base_model.output

# 1st conv block
x = layers.Conv2D(256, 3, padding='same')(x)
x = layers.BatchNormalization()(x)
x = layers.Activation('relu')(x)
x = layers.GlobalAveragePooling2D(keepdims = True)(x)

# 2nd conv block
x = layers.Conv2D(128, 3, padding='same')(x)
x = layers.BatchNormalization()(x)
x = layers.Activation('relu')(x)
x = layers.GlobalAveragePooling2D(keepdims = True)(x)

# 1st FC layer
x = layers.Flatten()(x)
x = layers.Dense(64)(x)
x = layers.BatchNormalization()(x)
x = layers.Activation('relu')(x)

# 2nd FC layer
x = layers.Dense(32, activation = 'relu')(x)
x = layers.BatchNormalization()(x)
x = layers.Activation('relu')(x)
x = layers.Dropout(.2)(x)

x = layers.Dense(3, 'softmax')(x)

incept_model = keras.models.Model(inputs = base_model.input, outputs = x)

# compile the model
incept_model.compile(optimizer=optimizers.RMSprop(.001), loss = losses.sparse_categorical_crossentropy, metrics= [metrics.SparseCategoricalAccuracy()])

incept_model.summary()

earlyStop = keras.callbacks.EarlyStopping(patience=60)
best_model = keras.callbacks.ModelCheckpoint(filepath='breast_VGG16_sheekar.h5', save_best_only=True)

with tf.device('/gpu:0'):
    history = incept_model.fit(x_train, y_train, batch_size=32, epochs=100, validation_data=(x_test, y_test), callbacks=[earlyStop, best_model])

# hist = history.history

# plt.plot(hist['loss'], label=  'loss')
# plt.plot(hist['val_loss'], label = 'val_loss')
# plt.plot(hist['sparse_categorical_accuracy'], label='accuracy')
# plt.plot(hist['val_sparse_categorical_accuracy'], label='val_accuracy')
# plt.legend()

# Summarize history for accu

plt.plot(history.history['sparse_categorical_accuracy'])
plt.plot(history.history['val_sparse_categorical_accuracy'])
plt.title('Model AUC')
plt.ylabel('AUC')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left', bbox_to_anchor=(1,1))
plt.show()

# Summarize history for loss

plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('Model Loss')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left', bbox_to_anchor=(1,1))
plt.show()

# used to predict the model and visualize the orignal image with title of true and pred values
def predict_image(img_path, label):
    img1 = prepare_image(img_path) # preprocess the image
    res = incept_model.predict(np.expand_dims(img1, axis = 0)) # predict the image
    pred = classes[np.argmax(res)]

    # Visualize the image
    img = image.load_img(img_path)
    plt.imshow(np.array(img))
    plt.title(f'True: {label}\nPredicted: {pred}')

predict_image('/content/drive/MyDrive/Breast Cancer/benign.png', 'benign')

predict_image('/content/drive/MyDrive/Breast Cancer/malignant.png', 'malignant')

predict_image('/content/drive/MyDrive/Breast Cancer/normal.png', 'normal')