aifnet_complete.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "6b038707",
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import os\n",
    "import zipfile\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "from tensorflow.keras import layers\n",
    "from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard\n",
    "import nibabel as nib\n",
    "from scipy import ndimage\n",
    "import glob\n",
    "import sys,os\n",
    "from natsort import natsorted\n",
    "import tensorflow_probability as tfp\n",
    "import random\n",
    "from aifnet_utils.preprocess import read_nifti_file, normalize, normalize_aif, process_scan, normalize_zero_one\n",
    "from aifnet_utils.losses import MaxCorrelation\n",
    "from aifnet_utils.data_loaders import read_isles_annotations, read_isles_volumes, ISLES18DataGen_aifvof_aug\n",
    "from aifnet_utils.data_loaders import delay_sequence_padding, anticipate_sequence_padding, late_bolus, early_bolus\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "da52553f",
   "metadata": {},
   "outputs": [],
   "source": [
    "keras.backend.set_image_data_format('channels_last')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "b58ab5b7",
   "metadata": {},
   "outputs": [],
   "source": [
    "root_dir     = '/media/sebastian/data/ASAP/ISLES2018_Training'\n",
    "#At insel: /media/sebastian/data/ASAP/ISLES2018_Training\n",
    "#Local: '/Users/sebastianotalora/work/postdoc/data/ISLES/'\n",
    "aif_annotations_path = '/home/sebastian/experiments/aifnet_replication/annotated_aif_vof_complete_revised.csv'\n",
    "min_num_volumes_ctp = 43\n",
    "ROOT_EXP = '/home/sebastian/experiments/aifnet_replication'#'/Users/sebastianotalora/work/postdoc/ctp/aifnet_replication'\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "4065099b",
   "metadata": {},
   "outputs": [],
   "source": [
    "aif_annotations, vof_annotations = read_isles_annotations(aif_annotations_path, root_dir, \n",
    "                                         min_num_volumes_ctp, return_aif_only = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "09afb354",
   "metadata": {},
   "outputs": [],
   "source": [
    "ctp_volumes = read_isles_volumes(root_dir, aif_annotations_path, min_num_volumes_ctp, take_two_slices_only=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "8badb987",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "94 94\n"
     ]
    }
   ],
   "source": [
    "print(len(ctp_volumes), len(aif_annotations))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "2be2a1bd",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_model_twoPvols(width=256, height=256, num_channels=43):\n",
    "    \"\"\"Build a 3D convolutional neural network model.\"\"\"\n",
    "    #width and height of the PCT is 256, the number of slices is variable, and the number of channels are\n",
    "    #the number of timepoints in the PCT sequence        \n",
    "    inputs = keras.Input((width, height, None , 43))\n",
    "\n",
    "    x = layers.Conv3D(filters=16, kernel_size=(3,3,1), activation=\"relu\", data_format='channels_last', padding='same')(inputs)\n",
    "    #x = layers.Dropout(0.3)(x)\n",
    "    \n",
    "    x = layers.Conv3D(filters=32, kernel_size=(3,3,3), activation=\"relu\", data_format='channels_last', padding='same')(x)\n",
    "    #x = layers.Dropout(0.3)(x)\n",
    "    \n",
    "    x = layers.Conv3D(filters=64, kernel_size=(3,3,3), activation=\"relu\", data_format='channels_last', padding='same')(x)\n",
    "    #x = layers.Dropout(0.3)(x)\n",
    "    \n",
    "    x = layers.Conv3D(filters=128, kernel_size=(3,3,3), activation=\"relu\", data_format='channels_last', padding='same')(x)\n",
    "\n",
    "    #x = layers.Conv3D(filters=256, kernel_size=(3,3,3), activation=\"relu\", data_format='channels_last', padding='same')(x)\n",
    "    #x = layers.Dropout(0.3)(x)\n",
    "    Lout = layers.Conv3D(filters=1, kernel_size=(3,3,3), activation=\"relu\", data_format='channels_last', padding='same')(x)\n",
    "    Lout2 = layers.Conv3D(filters=1, kernel_size=(3,3,3), activation=\"relu\", data_format='channels_last', padding='same')(x)\n",
    "\n",
    "    P_vol_aif = tf.keras.activations.sigmoid(Lout,name=\"aif\")\n",
    "    P_vol_vof = tf.keras.activations.sigmoid(Lout2,name=\"vof\")\n",
    "    #Voxelwise multiplication of P_vol and each of the CTP time points    \n",
    "    voxelwise_mult_each_ctp_aif = tf.keras.layers.Multiply()([inputs,P_vol_aif])    \n",
    "    voxelwise_mult_each_ctp_vof = tf.keras.layers.Multiply()([inputs,P_vol_vof])        \n",
    "    #The 3D average pooling block averages the volumetric information along the x-y-z axes, \n",
    "    #such that the predicted vascular function y(t) is a 1D vector of length T.\n",
    "    outputs_aif = layers.GlobalAveragePooling3D(data_format='channels_last', name='aif_loss')(voxelwise_mult_each_ctp_aif)\n",
    "    outputs_vof = layers.GlobalAveragePooling3D(data_format='channels_last', name='vof_loss')(voxelwise_mult_each_ctp_vof)    \n",
    "    #outputs_aif = tf.keras.activations.linear\n",
    "\n",
    "    # Define the model.\n",
    "    model = keras.Model(inputs, [outputs_aif,outputs_vof], name=\"aifnet\")\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "602e8165",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"aifnet\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            [(None, 256, 256, No 0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv3d (Conv3D)                 (None, 256, 256, Non 6208        input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "conv3d_1 (Conv3D)               (None, 256, 256, Non 13856       conv3d[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "conv3d_2 (Conv3D)               (None, 256, 256, Non 55360       conv3d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv3d_3 (Conv3D)               (None, 256, 256, Non 221312      conv3d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv3d_4 (Conv3D)               (None, 256, 256, Non 3457        conv3d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv3d_5 (Conv3D)               (None, 256, 256, Non 3457        conv3d_3[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "tf.keras.activations.sigmoid (T (None, 256, 256, Non 0           conv3d_4[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "tf.keras.activations.sigmoid_1  (None, 256, 256, Non 0           conv3d_5[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "multiply (Multiply)             (None, 256, 256, Non 0           input_1[0][0]                    \n",
      "                                                                 tf.keras.activations.sigmoid[0][0\n",
      "__________________________________________________________________________________________________\n",
      "multiply_1 (Multiply)           (None, 256, 256, Non 0           input_1[0][0]                    \n",
      "                                                                 tf.keras.activations.sigmoid_1[0]\n",
      "__________________________________________________________________________________________________\n",
      "aif_loss (GlobalAveragePooling3 (None, 43)           0           multiply[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "vof_loss (GlobalAveragePooling3 (None, 43)           0           multiply_1[0][0]                 \n",
      "==================================================================================================\n",
      "Total params: 303,650\n",
      "Trainable params: 303,650\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Build model.\n",
    "model = get_model_twoPvols(width=256, height=256, num_channels=43)\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "5926a438",
   "metadata": {},
   "outputs": [],
   "source": [
    "initial_learning_rate = 0.001\n",
    "lr_schedule = keras.optimizers.schedules.ExponentialDecay(\n",
    "    initial_learning_rate, decay_steps=100, decay_rate=0.96, staircase=True\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "8e138494",
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizer_aifnet = optimizer = keras.optimizers.SGD(learning_rate=lr_schedule) #keras.optimizers.Adam(learning_rate=initial_learning_rate)\n",
    "model.compile(\n",
    "    loss=['mse','mse'],\n",
    "    optimizer=optimizer_aifnet,\n",
    "    metrics=['mae'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "a3907124",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define callbacks.\n",
    "checkpoint_cb = keras.callbacks.ModelCheckpoint(\n",
    "    \"aif_onehead_classification.h5\", save_best_only=True\n",
    ")\n",
    "early_stopping_cb = keras.callbacks.EarlyStopping(monitor=\"train_acc\", patience=15)\n",
    "path_checkpointer_model = ROOT_EXP +'/results/' \n",
    "path_checkpointer_model += 'aifnet_2Pvols_augment' + str(initial_learning_rate)  + '.hdf5'\n",
    "path_tensorboard_log    = ROOT_EXP + '/results/logsTensorBoard/'\n",
    "path_tensorboard_log    += 'aifnet_2Pvols_augment' + str(initial_learning_rate) \n",
    "\n",
    "checkpointer = ModelCheckpoint(filepath=path_checkpointer_model, monitor='train_mse', \n",
    "                               verbose=1, save_best_only=True)\n",
    "tb_callback = TensorBoard(log_dir=path_tensorboard_log, histogram_freq=0, write_graph=True, write_grads=False,\n",
    "                          write_images=False, embeddings_freq=0, embeddings_layer_names=None,\n",
    "                          embeddings_metadata=None)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "0dd94a9c",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_datagen = ISLES18DataGen_aifvof_aug(ctp_volumes=ctp_volumes, annotations_aif=aif_annotations,\n",
    "                               annotations_vof=vof_annotations,minimum_number_volumes_ctp = 43, batch_size=1,\n",
    "                                         time_arrival_augmentation = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "7619cc03",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.08257481, 0.07730358, 0.11813583, 0.        , 0.05550066,\n",
       "        0.06830165, 0.04794057, 0.09128157, 0.221726  , 0.23537673,\n",
       "        0.48170579, 0.78750786, 0.86563477, 0.96241198, 1.        ,\n",
       "        0.99544322, 0.99889876, 0.96218275, 0.86189626, 0.65192204,\n",
       "        0.56970011, 0.50125448, 0.45215097, 0.39599262, 0.31808581,\n",
       "        0.29166209, 0.26091968, 0.22856351, 0.26744796, 0.15474904,\n",
       "        0.14516063, 0.19553963, 0.19290133, 0.14361789, 0.14484713,\n",
       "        0.16653791, 0.16653791, 0.16653791, 0.16653791, 0.16653791,\n",
       "        0.16653791, 0.16653791, 0.16653791],\n",
       "       [0.05415246, 0.02504378, 0.03393869, 0.        , 0.00204326,\n",
       "        0.02643459, 0.10194071, 0.05102015, 0.07753443, 0.16451013,\n",
       "        0.28472336, 0.45433134, 0.58080454, 0.78210145, 0.85873841,\n",
       "        0.85581729, 0.99849415, 1.        , 0.9553362 , 0.82155794,\n",
       "        0.79300284, 0.69930386, 0.59995823, 0.42033167, 0.37742713,\n",
       "        0.33015648, 0.28088619, 0.20560513, 0.20963086, 0.18939432,\n",
       "        0.21431999, 0.18197238, 0.19495682, 0.14332822, 0.13207351,\n",
       "        0.14165006, 0.14165006, 0.14165006, 0.14165006, 0.14165006,\n",
       "        0.14165006, 0.14165006, 0.14165006]])"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y = np.array(train_datagen.__getitem__(0)[1])\n",
    "y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "9abc1fde",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1, 2, 43)\n",
      "(1, 256, 256, 2, 43)\n",
      "[[[0.         0.         0.         0.         0.         0.\n",
      "   0.         0.123592   0.17131565 0.19008231 0.2277082  0.17284587\n",
      "   0.21105302 0.26207688 0.22799301 0.45720455 0.4913874  0.49300343\n",
      "   0.70448294 0.88714153 0.98140581 0.98507468 0.89363239 1.\n",
      "   0.95171536 0.80248594 0.69075235 0.76134212 0.75633998 0.67696369\n",
      "   0.5722515  0.46265193 0.38285113 0.38044071 0.56620863 0.43038192\n",
      "   0.35584986 0.35979398 0.3315117  0.36523448 0.41092181 0.43156517\n",
      "   0.33515234]\n",
      "  [0.10338247 0.10338247 0.10338247 0.10338247 0.10338247 0.10338247\n",
      "   0.10338247 0.04753512 0.01045466 0.01337223 0.07976035 0.01473522\n",
      "   0.         0.02586912 0.07904202 0.14527446 0.292079   0.48665035\n",
      "   0.68229571 0.6472989  0.74850151 0.91326647 1.         0.83873138\n",
      "   0.81583837 0.83897816 0.69090817 0.45762017 0.37877654 0.41745783\n",
      "   0.45702597 0.38181202 0.32355322 0.28945378 0.25921836 0.24872637\n",
      "   0.27938708 0.31982475 0.27394536 0.28936071 0.24303166 0.16613033\n",
      "   0.2315624 ]]]\n"
     ]
    }
   ],
   "source": [
    "sample_item = 69\n",
    "x = np.array(train_datagen.__getitem__(sample_item)[0])\n",
    "#x=x[np.newaxis,:,:,:,:]\n",
    "y = np.array(train_datagen.__getitem__(sample_item)[1])\n",
    "y = y[np.newaxis,:]\n",
    "print(y.shape)\n",
    "print(x.shape)\n",
    "print(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "30019dab",
   "metadata": {},
   "outputs": [],
   "source": [
    "load_previous_weights = True\n",
    "path_previous_weights = ROOT_EXP + '/results/trained_models/sgd_weights_8epochsaifnet_2Pvols_augment0.001.hdf5'\n",
    "if load_previous_weights:\n",
    "    model.load_weights(path_previous_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "eda6b1d6",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_predictions(model,x,y,normalize_preds=True, type_pred='AIF'):\n",
    "    pred = model.predict(np.expand_dims(normalize(x), axis=0))    \n",
    "    if type_pred=='AIF':\n",
    "        pred = pred[0][0]\n",
    "        plt.title('AIF Function Predictions')        \n",
    "    else:\n",
    "        pred = pred[1][0]\n",
    "        plt.title('VOF Function Predictions')\n",
    "\n",
    "    if normalize_preds:\n",
    "        print('Normalizing')\n",
    "        pred = normalize_zero_one(pred)\n",
    "        y = normalize_zero_one(y)\n",
    "    plt.plot(y)\n",
    "    plt.plot(pred)    \n",
    "    plt.legend(['y', 'prediction'])\n",
    "    plt.xlabel('Time (s)')\n",
    "    plt.ylabel('Normalized Density (HU)')\n",
    "    plt.show()\n",
    "    return"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "62d4183f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.6798698306083679,\n",
       " 0.34297728538513184,\n",
       " 0.3368925452232361,\n",
       " 0.5476176738739014,\n",
       " 0.5420355200767517]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.train_on_batch(x = train_datagen.__getitem__(9)[0], y = y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "f46ba06e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['loss', 'aif_loss_loss', 'vof_loss_loss', 'aif_loss_mae', 'vof_loss_mae']"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.metrics_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "3c488f5f",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/sebastian/anaconda3/envs/aifnet/lib/python3.9/site-packages/tensorflow/python/keras/utils/generic_utils.py:494: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.\n",
      "  warnings.warn('Custom mask layers require a config and must override '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/4\n",
      "94/94 [==============================] - 66s 644ms/step - loss: 0.3645 - aif_loss_loss: 0.1834 - vof_loss_loss: 0.1811 - aif_loss_mae: 0.3243 - vof_loss_mae: 0.3221\n",
      "WARNING:tensorflow:Can save best model only with train_mse available, skipping.\n",
      "Epoch 2/4\n",
      "94/94 [==============================] - 47s 500ms/step - loss: 0.3701 - aif_loss_loss: 0.1859 - vof_loss_loss: 0.1842 - aif_loss_mae: 0.3269 - vof_loss_mae: 0.3252\n",
      "WARNING:tensorflow:Can save best model only with train_mse available, skipping.\n",
      "Epoch 3/4\n",
      "94/94 [==============================] - 46s 467ms/step - loss: 0.3696 - aif_loss_loss: 0.1855 - vof_loss_loss: 0.1841 - aif_loss_mae: 0.3266 - vof_loss_mae: 0.3252\n",
      "WARNING:tensorflow:Can save best model only with train_mse available, skipping.\n",
      "Epoch 4/4\n",
      "94/94 [==============================] - 46s 484ms/step - loss: 0.3589 - aif_loss_loss: 0.1800 - vof_loss_loss: 0.1789 - aif_loss_mae: 0.3208 - vof_loss_mae: 0.3196\n",
      "WARNING:tensorflow:Can save best model only with train_mse available, skipping.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7f11f0170430>"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "nb_epochs=4\n",
    "model.fit(train_datagen,batch_size=1,callbacks=[checkpointer,tb_callback],epochs=nb_epochs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "5b9e1770",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save_weights('/home/sebastian/experiments/aifnet_replication/results/trained_models/sgd_weights_12epochs' + path_checkpointer_model.split('/')[-1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "29ad7f3a",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[array([[0.09072638, 0.09076854, 0.09071487, 0.09071849, 0.09074938,\n",
       "         0.09072965, 0.09070836, 0.09072001, 0.09074463, 0.09072916,\n",
       "         0.09069126, 0.09109224, 0.09105828, 0.09080672, 0.09100239,\n",
       "         0.09133022, 0.09183225, 0.09231695, 0.09293491, 0.09372507,\n",
       "         0.09467722, 0.09556019, 0.09638402, 0.09714124, 0.09749305,\n",
       "         0.09750446, 0.09739155, 0.09723754, 0.09685588, 0.09660017,\n",
       "         0.09618226, 0.09577855, 0.09543174, 0.0949845 , 0.09476492,\n",
       "         0.09457878, 0.09425073, 0.09402992, 0.0938217 , 0.09379556,\n",
       "         0.09371489, 0.09368948, 0.09361182]], dtype=float32),\n",
       " array([[0.09176896, 0.09181137, 0.09175752, 0.09176183, 0.09179178,\n",
       "         0.09177244, 0.09175107, 0.09176152, 0.09178658, 0.09177092,\n",
       "         0.09173323, 0.09215069, 0.09210767, 0.09184525, 0.09204162,\n",
       "         0.09237596, 0.09288664, 0.0933788 , 0.09401113, 0.09481792,\n",
       "         0.09578607, 0.09668726, 0.09752777, 0.09829843, 0.09865735,\n",
       "         0.09867033, 0.09855677, 0.09839832, 0.09800951, 0.09775203,\n",
       "         0.09732392, 0.096913  , 0.09655899, 0.09610227, 0.09587947,\n",
       "         0.09568925, 0.09535334, 0.09512939, 0.09491751, 0.09488763,\n",
       "         0.09480622, 0.09478065, 0.09469891]], dtype=float32)]"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict(np.expand_dims(normalize(ctp_volumes[10]['ctpvals']), axis=0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "0d61a4a8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "345969\n",
      "Normalizing\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "case_number = 60\n",
    "case_id = ctp_volumes[case_number]['image'].split('.')[-2]\n",
    "type_predictions = 'AIF'\n",
    "x = ctp_volumes[case_number]['ctpvals']\n",
    "if type_predictions == 'AIF':\n",
    "    y = aif_annotations[case_id]\n",
    "if type_predictions == 'VOF':\n",
    "    y = vof_annotations[case_id]\n",
    "print(case_id)\n",
    "plot_predictions(model,x,y,normalize_preds=True, type_pred=type_predictions)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}