h360-final-project-pneumonia-detection-using-cnn.ipynb

{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# Pneumonia Detection using CNN\nBailey Hirota, Fatima Irfan, Joon Luther\n\nThis notebook uses data from the Chest X-Ray Images (Pneumonia) dataset and performs classification using a CNN model. It is adapted from https://www.kaggle.com/code/madz2000/pneumonia-detection-using-cnn-92-6-accuracy. \n\n","metadata":{}},{"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load in \n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the \"../input/\" directory.\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n    for filename in filenames:\n        os.path.join(dirname, filename)\n\n# Any results you write to the current directory are saved as output.","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_kg_hide-output":true,"execution":{"iopub.status.busy":"2023-05-09T21:31:12.893211Z","iopub.execute_input":"2023-05-09T21:31:12.893646Z","iopub.status.idle":"2023-05-09T21:31:13.946665Z","shell.execute_reply.started":"2023-05-09T21:31:12.893603Z","shell.execute_reply":"2023-05-09T21:31:13.942798Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Importing the necessary libraries","metadata":{}},{"cell_type":"code","source":"import matplotlib.pyplot as plt\nimport seaborn as sns\nimport keras\nfrom keras.models import Sequential\nfrom keras.layers import Dense, Conv2D , MaxPool2D , Flatten , Dropout , BatchNormalization\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import classification_report,confusion_matrix\nfrom keras.callbacks import ReduceLROnPlateau\nimport cv2\nimport os","metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","execution":{"iopub.status.busy":"2023-05-09T21:31:13.948639Z","iopub.status.idle":"2023-05-09T21:31:13.949104Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Description of the Pneumonia Dataset\n**The dataset is organized into 3 folders (train, test, val) and contains subfolders for each image category (Pneumonia/Normal). There are 5,863 X-Ray images (JPEG) and 2 categories (Pneumonia/Normal).\nChest X-ray images (anterior-posterior) were selected from retrospective cohorts of pediatric patients of one to five years old from Guangzhou Women and Children’s Medical Center, Guangzhou. All chest X-ray imaging was performed as part of patients’ routine clinical care.\nFor the analysis of chest x-ray images, all chest radiographs were initially screened for quality control by removing all low quality or unreadable scans. The diagnoses for the images were then graded by two expert physicians before being cleared for training the AI system. In order to account for any grading errors, the evaluation set was also checked by a third expert.**","metadata":{}},{"cell_type":"code","source":"labels = ['PNEUMONIA', 'NORMAL']\nimg_size = 150\ndef get_training_data(data_dir):\n    data = [] \n    for label in labels: \n        path = os.path.join(data_dir, label)\n        class_num = labels.index(label)\n        for img in os.listdir(path):\n            try:\n                img_arr = cv2.imread(os.path.join(path, img), cv2.IMREAD_GRAYSCALE)\n                resized_arr = cv2.resize(img_arr, (img_size, img_size)) # Reshaping images to preferred size\n                data.append([resized_arr, class_num])\n            except Exception as e:\n                print(e)\n    return np.array(data)","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.950871Z","iopub.status.idle":"2023-05-09T21:31:13.951766Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Loading the Dataset","metadata":{}},{"cell_type":"code","source":"train = get_training_data('../input/chest-xray-pneumonia/chest_xray/chest_xray/train')\ntest = get_training_data('../input/chest-xray-pneumonia/chest_xray/chest_xray/test')\nval = get_training_data('../input/chest-xray-pneumonia/chest_xray/chest_xray/val')","metadata":{"_kg_hide-output":true,"execution":{"iopub.status.busy":"2023-05-09T21:31:13.953510Z","iopub.status.idle":"2023-05-09T21:31:13.954056Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Data Visualization & Preprocessing","metadata":{}},{"cell_type":"code","source":"l = []\nfor i in train:\n    if(i[1] == 0):\n        l.append(\"Pneumonia\")\n    else:\n        l.append(\"Normal\")\nsns.set_style('darkgrid')\nsns.countplot(l)        ","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.955508Z","iopub.status.idle":"2023-05-09T21:31:13.956056Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**The data seems imbalanced. To increase the no. of training examples, we will use data augmentation**","metadata":{}},{"cell_type":"markdown","source":"**Previewing the images of both the classes**","metadata":{}},{"cell_type":"code","source":"plt.figure(figsize = (5,5))\nplt.imshow(train[0][0], cmap='gray')\nplt.title(labels[train[0][1]])\n\nplt.figure(figsize = (5,5))\nplt.imshow(train[-1][0], cmap='gray')\nplt.title(labels[train[-1][1]])","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.957643Z","iopub.status.idle":"2023-05-09T21:31:13.958212Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"x_train = []\ny_train = []\n\nx_val = []\ny_val = []\n\nx_test = []\ny_test = []\n\nfor feature, label in train:\n    x_train.append(feature)\n    y_train.append(label)\n\nfor feature, label in test:\n    x_test.append(feature)\n    y_test.append(label)\n    \nfor feature, label in val:\n    x_val.append(feature)\n    y_val.append(label)","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.959550Z","iopub.status.idle":"2023-05-09T21:31:13.960116Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**We perform a grayscale normalization to reduce the effect of illumination's differences.Moreover the CNN converges faster on [0..1] data than on [0..255].**","metadata":{}},{"cell_type":"code","source":"# Normalize the data\nx_train = np.array(x_train) / 255\nx_val = np.array(x_val) / 255\nx_test = np.array(x_test) / 255","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.961462Z","iopub.status.idle":"2023-05-09T21:31:13.961985Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# resize data for deep learning \nx_train = x_train.reshape(-1, img_size, img_size, 1)\ny_train = np.array(y_train)\n\nx_val = x_val.reshape(-1, img_size, img_size, 1)\ny_val = np.array(y_val)\n\nx_test = x_test.reshape(-1, img_size, img_size, 1)\ny_test = np.array(y_test)","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.963337Z","iopub.status.idle":"2023-05-09T21:31:13.963749Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Data Augmentation\n**In order to avoid overfitting problem, we need to expand artificially our dataset. We can make your existing dataset even larger. The idea is to alter the training data with small transformations to reproduce the variations.\nApproaches that alter the training data in ways that change the array representation while keeping the label the same are known as data augmentation techniques. Some popular augmentations people use are grayscales, horizontal flips, vertical flips, random crops, color jitters, translations, rotations, and much more.\nBy applying just a couple of these transformations to our training data, we can easily double or triple the number of training examples and create a very robust model.**","metadata":{}},{"cell_type":"code","source":"# With data augmentation to prevent overfitting and handling the imbalance in dataset\n\ndatagen = ImageDataGenerator(\n        featurewise_center=False,  # set input mean to 0 over the dataset\n        samplewise_center=False,  # set each sample mean to 0\n        featurewise_std_normalization=False,  # divide inputs by std of the dataset\n        samplewise_std_normalization=False,  # divide each input by its std\n        zca_whitening=False,  # apply ZCA whitening\n        rotation_range = 30,  # randomly rotate images in the range (degrees, 0 to 180)\n        zoom_range = 0.2, # Randomly zoom image \n        width_shift_range=0.4,  # randomly shift images horizontally (fraction of total width)\n        height_shift_range=0.1,  # randomly shift images vertically (fraction of total height)\n        horizontal_flip = True,  # randomly flip images\n        vertical_flip=False)  # randomly flip images\n\n\ndatagen.fit(x_train)","metadata":{"execution":{"iopub.status.busy":"2023-05-09T22:25:24.916669Z","iopub.execute_input":"2023-05-09T22:25:24.917348Z","iopub.status.idle":"2023-05-09T22:25:25.430111Z","shell.execute_reply.started":"2023-05-09T22:25:24.917285Z","shell.execute_reply":"2023-05-09T22:25:25.429299Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"For the data augmentation, we choose to vary:\n1. rotation_range\n2. zoom_range\n3. width_shift_range \n4. height_shift_range \nOnce our model is ready, we fit the training dataset.","metadata":{}},{"cell_type":"markdown","source":"# Training the Model\nWe implement grid search here on different parameters to find optimality and then build the model using optimal parameters.\n\n## Grid Search on Parameters (OPTIONAL)\n","metadata":{}},{"cell_type":"code","source":"from keras.wrappers.scikit_learn import KerasClassifier\nfrom sklearn.model_selection import GridSearchCV\n\n# define the function to create the model\ndef create_model(activation='relu', dropout_rate=0.2, strides=1, padding='same'):\n    model = Sequential()\n    model.add(Conv2D(32, (3,3), strides=strides, padding=padding, activation=activation, input_shape=(150,150,1)))\n    model.add(BatchNormalization())\n    model.add(MaxPool2D((2,2), strides=2, padding='same'))\n    model.add(Conv2D(64, (3,3), strides=strides, padding=padding, activation=activation))\n    model.add(Dropout(dropout_rate))\n    model.add(BatchNormalization())\n    model.add(MaxPool2D((2,2), strides=2, padding='same'))\n    model.add(Conv2D(64, (3,3), strides=strides, padding=padding, activation=activation))\n    model.add(BatchNormalization())\n    model.add(MaxPool2D((2,2), strides=2, padding='same'))\n    model.add(Conv2D(128, (3,3), strides=strides, padding=padding, activation=activation))\n    model.add(Dropout(dropout_rate))\n    model.add(BatchNormalization())\n    model.add(MaxPool2D((2,2), strides=2, padding='same'))\n    model.add(Conv2D(256, (3,3), strides=strides, padding=padding, activation=activation))\n    model.add(Dropout(dropout_rate))\n    model.add(BatchNormalization())\n    model.add(MaxPool2D((2,2), strides=2, padding='same'))\n    model.add(Flatten())\n    model.add(Dense(units=128, activation=activation))\n    model.add(Dropout(dropout_rate))\n    model.add(Dense(units=1, activation='sigmoid'))\n    model.compile(optimizer=\"rmsprop\", loss='binary_crossentropy', metrics=['accuracy'])\n    return model\n\n# create the Keras classifier\nmodel = KerasClassifier(build_fn=create_model, epochs=10, batch_size=32, verbose=0)\n\n# define the grid search parameters\nparam_grid = {\n    'activation': ['relu', 'tanh'],\n    'dropout_rate': [0.1, 0.2],\n    'strides': [1, 2],\n    'padding': ['same']\n}\n\n# create the grid search object\ngrid = GridSearchCV(estimator=model, param_grid=param_grid, n_jobs=-1, cv=3)\n\n# fit the grid search object to the training data\ngrid_result = grid.fit(x_train, y_train)\n\n# print the best results\nprint(\"Best: %f using %s\" % (grid_result.best_score_, grid_result.best_params_))\n","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.966356Z","iopub.status.idle":"2023-05-09T21:31:13.966759Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Training the Model 2\n\nWe observe the best accuracy using tanh activation, 0.2 dropout, same padding, and a stride of 1. Therefore, we build our model with those specifications on Conv2D filters.","metadata":{}},{"cell_type":"code","source":"model = Sequential()\nmodel.add(Conv2D(32 , (3,3) , strides = 1 , padding = 'same' , activation = 'tanh' , input_shape = (150,150,1)))\nmodel.add(BatchNormalization())\nmodel.add(MaxPool2D((2,2) , strides = 2 , padding = 'same'))\nmodel.add(Conv2D(64 , (3,3) , strides = 1 , padding = 'same' , activation = 'tanh'))\nmodel.add(Dropout(0.2))\nmodel.add(BatchNormalization())\nmodel.add(MaxPool2D((2,2) , strides = 2 , padding = 'same'))\nmodel.add(Conv2D(64 , (3,3) , strides = 1 , padding = 'same' , activation = 'tanh'))\nmodel.add(BatchNormalization())\nmodel.add(MaxPool2D((2,2) , strides = 2 , padding = 'same'))\nmodel.add(Conv2D(128 , (3,3) , strides = 1 , padding = 'same' , activation = 'tanh'))\nmodel.add(Dropout(0.2))\nmodel.add(BatchNormalization())\nmodel.add(MaxPool2D((2,2) , strides = 2 , padding = 'same'))\nmodel.add(Conv2D(256 , (3,3) , strides = 1 , padding = 'same' , activation = 'tanh'))\nmodel.add(Dropout(0.2))\nmodel.add(BatchNormalization())\nmodel.add(MaxPool2D((2,2) , strides = 2 , padding = 'same'))\nmodel.add(Flatten())\nmodel.add(Dense(units = 128 , activation = 'relu'))\nmodel.add(Dropout(0.2))\nmodel.add(Dense(units = 1 , activation = 'sigmoid'))\nmodel.compile(optimizer = \"rmsprop\" , loss = 'binary_crossentropy' , metrics = ['accuracy'])\nmodel.summary()","metadata":{"execution":{"iopub.status.busy":"2023-05-09T22:25:28.638866Z","iopub.execute_input":"2023-05-09T22:25:28.639448Z","iopub.status.idle":"2023-05-09T22:25:29.118574Z","shell.execute_reply.started":"2023-05-09T22:25:28.639406Z","shell.execute_reply":"2023-05-09T22:25:29.117648Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"learning_rate_reduction = ReduceLROnPlateau(monitor='val_accuracy', patience = 2, verbose=1,factor=0.3, min_lr=0.000001)","metadata":{"execution":{"iopub.status.busy":"2023-05-09T22:25:34.035740Z","iopub.execute_input":"2023-05-09T22:25:34.036066Z","iopub.status.idle":"2023-05-09T22:25:34.042691Z","shell.execute_reply.started":"2023-05-09T22:25:34.036024Z","shell.execute_reply":"2023-05-09T22:25:34.041617Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history = model.fit(datagen.flow(x_train,y_train, batch_size = 32) ,epochs = 12 , validation_data = datagen.flow(x_val, y_val) ,callbacks = [learning_rate_reduction])","metadata":{"execution":{"iopub.status.busy":"2023-05-09T22:25:35.700509Z","iopub.execute_input":"2023-05-09T22:25:35.700859Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"Loss of the model is - \" , model.evaluate(x_test,y_test)[0])\nprint(\"Accuracy of the model is - \" , model.evaluate(x_test,y_test)[1]*100 , \"%\")","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Analysis after Model Training","metadata":{}},{"cell_type":"code","source":"epochs = [i for i in range(12)]\nfig , ax = plt.subplots(1,2)\ntrain_acc = history.history['accuracy']\ntrain_loss = history.history['loss']\nval_acc = history.history['val_accuracy']\nval_loss = history.history['val_loss']\nfig.set_size_inches(20,10)\n\nax[0].plot(epochs , train_acc , 'go-' , label = 'Training Accuracy')\nax[0].plot(epochs , val_acc , 'ro-' , label = 'Validation Accuracy')\nax[0].set_title('Training & Validation Accuracy')\nax[0].legend()\nax[0].set_xlabel(\"Epochs\")\nax[0].set_ylabel(\"Accuracy\")\n\nax[1].plot(epochs , train_loss , 'g-o' , label = 'Training Loss')\nax[1].plot(epochs , val_loss , 'r-o' , label = 'Validation Loss')\nax[1].set_title('Testing Accuracy & Loss')\nax[1].legend()\nax[1].set_xlabel(\"Epochs\")\nax[1].set_ylabel(\"Training & Validation Loss\")\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.973360Z","iopub.status.idle":"2023-05-09T21:31:13.973735Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"predictions = model.predict_classes(x_test)\npredictions = predictions.reshape(1,-1)[0]\npredictions[:15]","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.974901Z","iopub.status.idle":"2023-05-09T21:31:13.975318Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(classification_report(y_test, predictions, target_names = ['Pneumonia (Class 0)','Normal (Class 1)']))","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.976306Z","iopub.status.idle":"2023-05-09T21:31:13.976678Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cm = confusion_matrix(y_test,predictions)\ncm","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.977696Z","iopub.status.idle":"2023-05-09T21:31:13.978067Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cm = pd.DataFrame(cm , index = ['0','1'] , columns = ['0','1'])","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.979246Z","iopub.status.idle":"2023-05-09T21:31:13.979625Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.figure(figsize = (10,10))\nsns.heatmap(cm,cmap= \"Blues\", linecolor = 'black' , linewidth = 1 , annot = True, fmt='',xticklabels = labels,yticklabels = labels)","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.980668Z","iopub.status.idle":"2023-05-09T21:31:13.981040Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"correct = np.nonzero(predictions == y_test)[0]\nincorrect = np.nonzero(predictions != y_test)[0]","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.982029Z","iopub.status.idle":"2023-05-09T21:31:13.982449Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Some of the Correctly Predicted Classes**","metadata":{}},{"cell_type":"code","source":"i = 0\nfor c in correct[:6]:\n    plt.subplot(3,2,i+1)\n    plt.xticks([])\n    plt.yticks([])\n    plt.imshow(x_test[c].reshape(150,150), cmap=\"gray\", interpolation='none')\n    plt.title(\"Predicted Class {},Actual Class {}\".format(predictions[c], y_test[c]))\n    plt.tight_layout()\n    i += 1","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.983501Z","iopub.status.idle":"2023-05-09T21:31:13.983867Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Some of the Incorrectly Predicted Classes**","metadata":{}},{"cell_type":"code","source":"i = 0\nfor c in incorrect[:6]:\n    plt.subplot(3,2,i+1)\n    plt.xticks([])\n    plt.yticks([])\n    plt.imshow(x_test[c].reshape(150,150), cmap=\"gray\", interpolation='none')\n    plt.title(\"Predicted Class {},Actual Class {}\".format(predictions[c], y_test[c]))\n    plt.tight_layout()\n    i += 1","metadata":{"execution":{"iopub.status.busy":"2023-05-09T21:31:13.985140Z","iopub.status.idle":"2023-05-09T21:31:13.985510Z"},"trusted":true},"execution_count":null,"outputs":[]}]}