{"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":"code","source":"import os\nimport json\nimport glob\nimport random\nimport collections\n\nimport numpy as np\nimport pandas as pd\nimport pydicom\nfrom pydicom.pixel_data_handlers.util import apply_voi_lut\nimport cv2\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nimport random\nfrom tqdm.notebook import tqdm\n\nimport tensorflow as tf\nfrom tensorflow import keras\nfrom sklearn.metrics import roc_auc_score\nfrom sklearn.model_selection import train_test_split\nfrom tensorflow.keras.utils import to_categorical\nfrom tensorflow.keras import layers\n\n\n\nTYPES = [\"FLAIR\", \"T1w\", \"T2w\", \"T1wCE\"]\nWHITE_THRESHOLD = 10 # out of 255\nEXCLUDE = [109, 123, 709]\n\n\ntrain_df = pd.read_csv(\"../input/rsna-miccai-brain-tumor-radiogenomic-classification/train_labels.csv\")\ntest_df = pd.read_csv('../input/rsna-miccai-brain-tumor-radiogenomic-classification/sample_submission.csv')\ntrain_df = train_df[~train_df.BraTS21ID.isin(EXCLUDE)]","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2021-10-14T07:18:06.856408Z","iopub.execute_input":"2021-10-14T07:18:06.856749Z","iopub.status.idle":"2021-10-14T07:18:06.876663Z","shell.execute_reply.started":"2021-10-14T07:18:06.856705Z","shell.execute_reply":"2021-10-14T07:18:06.875941Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def load_dicom(path, size = 224):\n    ''' \n    Reads a DICOM image, standardizes so that the pixel values are between 0 and 1, then rescales to 0 and 255\n    \n    Note super sure if this kind of scaling is appropriate, but everyone seems to do it. \n    '''\n    dicom = pydicom.read_file(path)\n    data = dicom.pixel_array\n    if np.max(data) != 0:\n        data = data / np.max(data)\n    data = (data * 255).astype(np.uint8)\n    return cv2.resize(data, (size, size))\n\ndef get_all_image_paths(brats21id, image_type, folder='train'): \n    '''\n    Returns an arry of all the images of a particular type for a particular patient ID\n    '''\n    assert(image_type in TYPES)\n    \n    patient_path = os.path.join(\n        \"../input/rsna-miccai-brain-tumor-radiogenomic-classification/%s/\" % folder, \n        str(brats21id).zfill(5),\n    )\n\n    paths = sorted(\n        glob.glob(os.path.join(patient_path, image_type, \"*\")), \n        key=lambda x: int(x[:-4].split(\"-\")[-1]),\n    )\n    \n    num_images = len(paths)\n    \n    start = int(num_images * 0.25)\n    end = int(num_images * 0.75)\n\n    interval = 3\n    \n    if num_images < 10: \n        interval = 1\n    \n    return np.array(paths[start:end:interval])\n\ndef get_all_images(brats21id, image_type, folder='train', size=225):\n    return [load_dicom(path, size) for path in get_all_image_paths(brats21id, image_type, folder)]\n","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:18:06.878127Z","iopub.execute_input":"2021-10-14T07:18:06.878472Z","iopub.status.idle":"2021-10-14T07:18:06.887877Z","shell.execute_reply.started":"2021-10-14T07:18:06.878437Z","shell.execute_reply":"2021-10-14T07:18:06.886854Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"IMAGE_SIZE = 128\n\ndef get_all_data_for_train(image_type):\n    global train_df\n    \n    X = []\n    y = []\n    train_ids = []\n\n    for i in tqdm(train_df.index):\n        x = train_df.loc[i]\n        images = get_all_images(int(x['BraTS21ID']), image_type, 'train', IMAGE_SIZE)\n        label = x['MGMT_value']\n\n        X += images\n        y += [label] * len(images)\n        train_ids += [int(x['BraTS21ID'])] * len(images)\n        assert(len(X) == len(y))\n    return np.array(X), np.array(y), np.array(train_ids)\n\ndef get_all_data_for_test(image_type):\n    global test_df\n    \n    X = []\n    test_ids = []\n\n    for i in tqdm(test_df.index):\n        x = test_df.loc[i]\n        images = get_all_images(int(x['BraTS21ID']), image_type, 'test', IMAGE_SIZE)\n        X += images\n        test_ids += [int(x['BraTS21ID'])] * len(images)\n\n    return np.array(X), np.array(test_ids)","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:18:06.890266Z","iopub.execute_input":"2021-10-14T07:18:06.890689Z","iopub.status.idle":"2021-10-14T07:18:06.901657Z","shell.execute_reply.started":"2021-10-14T07:18:06.890654Z","shell.execute_reply":"2021-10-14T07:18:06.900821Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"X, y, trainidt = get_all_data_for_train('T1wCE')\nX_test, testidt = get_all_data_for_test('T1wCE')\nX.shape, y.shape, trainidt.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:18:06.903228Z","iopub.execute_input":"2021-10-14T07:18:06.903605Z","iopub.status.idle":"2021-10-14T07:19:03.133584Z","shell.execute_reply.started":"2021-10-14T07:18:06.903570Z","shell.execute_reply":"2021-10-14T07:19:03.132655Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"X.shape, y.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:19:03.134814Z","iopub.execute_input":"2021-10-14T07:19:03.135170Z","iopub.status.idle":"2021-10-14T07:19:03.140509Z","shell.execute_reply.started":"2021-10-14T07:19:03.135133Z","shell.execute_reply":"2021-10-14T07:19:03.139713Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"X_train, X_valid, y_train, y_valid, trainidt_train, trainidt_valid = train_test_split(X, y, trainidt, test_size=0.1, random_state=140)\n\nsplit = int(X.shape[0] * 0.9)\n\n# X_train = X[:split]\n# X_valid = X[split:]\n\n# y_train = y[:split]\n# y_valid = y[split:]\n\n# trainidt_train = trainidt[:split]\n# trainidt_valid = trainidt[split:]\n\nX_train = tf.expand_dims(X_train, axis=-1)\nX_valid = tf.expand_dims(X_valid, axis=-1)\n\ny_train = to_categorical(y_train)\ny_valid = to_categorical(y_valid)\n\nX_train.shape, y_train.shape, X_valid.shape, y_valid.shape, trainidt_train.shape, trainidt_valid.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:19:03.141834Z","iopub.execute_input":"2021-10-14T07:19:03.142474Z","iopub.status.idle":"2021-10-14T07:19:03.473066Z","shell.execute_reply.started":"2021-10-14T07:19:03.142439Z","shell.execute_reply":"2021-10-14T07:19:03.472230Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"global_average_layer = tf.keras.layers.GlobalAveragePooling2D()","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:19:03.474367Z","iopub.execute_input":"2021-10-14T07:19:03.474713Z","iopub.status.idle":"2021-10-14T07:19:03.480494Z","shell.execute_reply.started":"2021-10-14T07:19:03.474660Z","shell.execute_reply":"2021-10-14T07:19:03.479695Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data_augmentation = tf.keras.Sequential([\n  tf.keras.layers.experimental.preprocessing.RandomFlip('horizontal'),\n  tf.keras.layers.experimental.preprocessing.RandomRotation(0.2),\n])","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:19:03.482901Z","iopub.execute_input":"2021-10-14T07:19:03.483441Z","iopub.status.idle":"2021-10-14T07:19:03.499384Z","shell.execute_reply.started":"2021-10-14T07:19:03.483403Z","shell.execute_reply":"2021-10-14T07:19:03.498633Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"np.random.seed(0)\nrandom.seed(12)\ntf.random.set_seed(12)\n\ninpt = keras.Input(shape=X_train.shape[1:])\n\npre = keras.layers.experimental.preprocessing.Rescaling(1./255)(inpt)\n# h = data_augmentation(h)\n\n# convolutional layer!\nh = keras.layers.Conv2D(32, kernel_size=(3, 3),strides=(1,1), activation=\"relu\", name=\"Conv_1\", padding=\"valid\")(pre) \nh = tf.keras.layers.BatchNormalization(axis=-1)(h)\nh = keras.layers.Conv2D(32, kernel_size=(3, 3),strides=(1,1), activation=\"relu\", name=\"Conv_1b\", padding=\"same\")(h) \nh = tf.keras.layers.BatchNormalization(axis=-1)(h)\nh = keras.layers.MaxPool2D(pool_size=(2,2))(h)\nh = keras.layers.Conv2D(64, kernel_size=(3, 3),strides=(1,1), activation=\"relu\", name=\"Conv_2\", padding=\"valid\")(h) \nh = tf.keras.layers.BatchNormalization(axis=-1)(h)\n# pooling layer\nh = keras.layers.MaxPool2D(pool_size=(2,2))(h) \nh = tf.keras.layers.BatchNormalization(axis=-1)(h)\n# convolutional layer!\nh = keras.layers.Conv2D(64, kernel_size=(3, 3), activation=\"relu\", name=\"Conv_3\",padding =\"same\")(h)\n# h = tf.keras.layers.BatchNormalization(axis=-1)(h)\n# pooling layer\n# h = keras.layers.MaxPool2D(pool_size=(1,1))(h)\nh = tf.keras.layers.BatchNormalization(axis=-1)(h)\n# h = keras.layers.Conv2D(128, kernel_size=(3, 3), activation=\"relu\", name=\"Conv_4\",padding =\"valid\")(h)\n# h = tf.keras.layers.BatchNormalization(axis=-1)(h)\n# h = keras.layers.Conv2D(128, kernel_size=(3, 3), activation=\"relu\", name=\"Conv_5\",padding =\"same\")(h)\n# h = tf.keras.layers.BatchNormalization(axis=-1)(h)\n# h = keras.layers.MaxPool2D(pool_size=(2,2))(h)\n# h = tf.keras.layers.BatchNormalization(axis=-1)(h)\nh = keras.layers.Dropout(0.4)(h)   \n\nh = keras.layers.Flatten()(h) \n\nh1 = keras.layers.experimental.preprocessing.Rescaling(1./255)(pre)\nh1 =  keras.layers.MaxPool2D(strides=(5,5))(h1)\nh1 = keras.layers.Conv2D(32, kernel_size=(3, 3), activation=\"swish\", name=\"Conv_4b\")(h1)\nh1 = tf.keras.layers.BatchNormalization(axis=-1)(h1)\nh1 =  keras.layers.MaxPool2D(strides=(5,5))(h1)\nh1 = keras.layers.Dropout(0.3)(h1)\nh1 = keras.layers.Conv2D(32, kernel_size=(3, 3), activation=\"swish\", name=\"Conv_5b\")(h1)\nh1 = tf.keras.layers.BatchNormalization(axis=-1)(h1)\nh1 =  keras.layers.MaxPool2D(strides=(5,5))(h1)\nh1 = keras.layers.Dropout(0.3)(h1)\nh1 = keras.layers.Flatten()(h1) \n\nmerge = keras.layers.Concatenate()([h,h1])\n# h = global_average_layer(h)\nout = keras.layers.Dropout(0.4)(merge)\nout = keras.layers.Dense(150, activation='relu')(out)   \n\n# out = keras.layers.Dense(64, activation='relu')(out)   \n# out = keras.layers.Dropout(0.2)(out)\n# out = keras.layers.Dense(32, activation='relu')(out)   \n# out = keras.layers.Dropout(0.2)(out)\noutput = keras.layers.Dense(2, activation=\"softmax\")(out)\n\nmodel = keras.Model(inpt, output)\n\nfrom keras.optimizers import SGD\n# opt = SGD(lr=0.1)\n\ncheckpoint_filepath = 'best_model.h5'\nmodel_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(\nfilepath=checkpoint_filepath,\nsave_weights_only=False,\nmonitor='val_auc',\nmode='max',\nsave_best_only=True,\nsave_freq='epoch')\n\nmodel.compile(loss='categorical_crossentropy',\n             optimizer=tf.keras.optimizers.SGD(learning_rate =0.0001),\n             metrics=[tf.keras.metrics.AUC()])\n\nhistory = model.fit(x=X_train, y = y_train, epochs=95, callbacks=[model_checkpoint_callback], validation_data= (X_valid, y_valid))","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:19:03.501732Z","iopub.execute_input":"2021-10-14T07:19:03.501998Z","iopub.status.idle":"2021-10-14T07:33:55.020558Z","shell.execute_reply.started":"2021-10-14T07:19:03.501974Z","shell.execute_reply":"2021-10-14T07:33:55.019740Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model_best = tf.keras.models.load_model(filepath=checkpoint_filepath)","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:35:03.966179Z","iopub.execute_input":"2021-10-14T07:35:03.966607Z","iopub.status.idle":"2021-10-14T07:35:04.268069Z","shell.execute_reply.started":"2021-10-14T07:35:03.966567Z","shell.execute_reply":"2021-10-14T07:35:04.267234Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred = model_best.predict(X_valid)\n\npred = np.argmax(y_pred, axis=1)\n\nresult=pd.DataFrame(trainidt_valid)\nresult[1]=pred\n\nresult.columns=['BraTS21ID','MGMT_value']\nresult2 = result.groupby('BraTS21ID',as_index=False).mean()\n\nresult2 = result2.merge(train_df, on='BraTS21ID')\nroc_auc_score(result2.MGMT_value_y, result2.MGMT_value_x,)","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:35:05.184173Z","iopub.execute_input":"2021-10-14T07:35:05.184572Z","iopub.status.idle":"2021-10-14T07:35:05.606165Z","shell.execute_reply.started":"2021-10-14T07:35:05.184489Z","shell.execute_reply":"2021-10-14T07:35:05.605344Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"len(result2)","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:33:55.665055Z","iopub.execute_input":"2021-10-14T07:33:55.665389Z","iopub.status.idle":"2021-10-14T07:33:55.674376Z","shell.execute_reply.started":"2021-10-14T07:33:55.665353Z","shell.execute_reply":"2021-10-14T07:33:55.673470Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"sample = pd.read_csv('../input/rsna-miccai-brain-tumor-radiogenomic-classification/sample_submission.csv')\n\ny_pred = model_best.predict(X_test)\n\npred = np.argmax(y_pred, axis=1)\n\nresult=pd.DataFrame(testidt)\nresult[1]=pred\n\nresult.columns=['BraTS21ID','MGMT_value']\nresult2 = result.groupby('BraTS21ID',as_index=False).mean()\nresult2['BraTS21ID'] = sample['BraTS21ID']\nresult2['MGMT_value'] = result2['MGMT_value'].apply(lambda x:round(x*10)/10)\nresult2.to_csv('submission.csv',index=False)\nresult2","metadata":{"execution":{"iopub.status.busy":"2021-10-14T07:33:55.675948Z","iopub.execute_input":"2021-10-14T07:33:55.676579Z","iopub.status.idle":"2021-10-14T07:33:56.228743Z","shell.execute_reply.started":"2021-10-14T07:33:55.676545Z","shell.execute_reply":"2021-10-14T07:33:56.227970Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}