3D_Classification.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "3D Classification.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "machine_shape": "hm",
      "mount_file_id": "1Yt50EayGY38jEb-yfDor6aqJvnrBEEDl",
      "authorship_tag": "ABX9TyPz+DELAiAsn8VlfuVR+/CH",
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Meddebma/pyradiomics/blob/master/3D_Classification.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wiFnM05m6uL2"
      },
      "source": [
        "%pip install \"monai--weekly[nibabel, skimage, pillow, tensorboard, gdown, ignite, torchvision, itk, tqdm, lmdb, psutil]\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8JIt-TFM-3ZL",
        "outputId": "77ce72d5-87f4-42c5-8cfd-7ecc4b85564a"
      },
      "source": [
        "# Copyright 2020 MONAI Consortium\n",
        "# Licensed under the Apache License, Version 2.0 (the \"License\");\n",
        "# you may not use this file except in compliance with the License.\n",
        "# You may obtain a copy of the License at\n",
        "#     http://www.apache.org/licenses/LICENSE-2.0\n",
        "# Unless required by applicable law or agreed to in writing, software\n",
        "# distributed under the License is distributed on an \"AS IS\" BASIS,\n",
        "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
        "# See the License for the specific language governing permissions and\n",
        "# limitations under the License.\n",
        "\n",
        "import logging\n",
        "import os\n",
        "import sys\n",
        "import tempfile\n",
        "import shutil\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "import torch\n",
        "from torch.utils.tensorboard import SummaryWriter\n",
        "import numpy as np\n",
        "\n",
        "import monai\n",
        "from monai.networks.nets import DenseNet121, Classifier\n",
        "from monai.apps import download_and_extract\n",
        "from monai.config import print_config\n",
        "from monai.utils import first\n",
        "from monai.visualize import GradCAM\n",
        "from monai.metrics import ConfusionMatrixMetric, get_confusion_matrix\n",
        "from monai.data import CacheDataset, DataLoader, ImageDataset, ITKReader, PILReader, Dataset, partition_dataset_classes\n",
        "from sklearn.metrics import (\n",
        "    classification_report, confusion_matrix,\n",
        "    ConfusionMatrixDisplay, \n",
        ")\n",
        "from monai.data import decollate_batch\n",
        "from monai.metrics import ROCAUCMetric\n",
        "from torch.utils.data import DataLoader \n",
        "from monai.transforms import Transform\n",
        "from monai.transforms import (\n",
        "    AddChannel,\n",
        "    Compose,\n",
        "    LoadImage,\n",
        "    RandRotate90,\n",
        "    Invertd,\n",
        "    Lambda,\n",
        "    Resize,\n",
        "    Orientation,\n",
        "    Orientationd,\n",
        "    EnsureChannelFirstd,\n",
        "    ScaleIntensity,\n",
        "    AsChannelFirst,\n",
        "    AsChannelLast,\n",
        "    Activations,\n",
        "    AddChannel,\n",
        "    AddChanneld,\n",
        "    ToTensord,\n",
        "    Resized,\n",
        "    AsDiscrete,\n",
        "    Spacingd,\n",
        "    EnsureChannelFirst,\n",
        "    RepeatChannel,\n",
        "    RemoveRepeatedChannel,\n",
        "    ScaleIntensityRanged,\n",
        "    SplitChannel,\n",
        "    SqueezeDim,\n",
        "    ToTensor,\n",
        "    Randomizable,\n",
        "    LoadImaged,\n",
        "    ToTensord,\n",
        "    MaskIntensityd,\n",
        "    Flip,\n",
        "    MaskIntensity,\n",
        "    ScaleIntensityd,\n",
        "    EnsureType,\n",
        "    EnsureTyped,\n",
        ")\n",
        "import numpy as np\n",
        "import statsmodels.formula.api as smf\n",
        "import statsmodels.api as sm\n",
        "import pandas as pd\n",
        "import seaborn as sns \n",
        "from glob import glob\n",
        "import logging\n",
        "from abc import ABC, abstractmethod\n",
        "from typing import Any, Callable, Dict, Generator, Hashable, Iterable, List, Optional, Tuple\n",
        "\n",
        "import numpy as np\n",
        "import torch\n",
        "\n",
        "import numpy as np\n",
        "from monai.transforms import Transform, AddChannel, AsChannelFirst\n",
        "\n",
        "\n",
        "logging.basicConfig(stream=sys.stdout, level=logging.INFO)\n",
        "print_config()"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "MONAI version: 0.7.dev2132\n",
            "Numpy version: 1.19.5\n",
            "Pytorch version: 1.9.0+cu102\n",
            "MONAI flags: HAS_EXT = False, USE_COMPILED = False\n",
            "MONAI rev id: bcbf21ec88dd95929b1d6439ebbf37b3cb266988\n",
            "\n",
            "Optional dependencies:\n",
            "Pytorch Ignite version: 0.4.5\n",
            "Nibabel version: 3.0.2\n",
            "scikit-image version: 0.16.2\n",
            "Pillow version: 7.1.2\n",
            "Tensorboard version: 2.5.0\n",
            "gdown version: 3.6.4\n",
            "TorchVision version: 0.10.0+cu102\n",
            "tqdm version: 4.62.0\n",
            "lmdb version: 0.99\n",
            "psutil version: 5.4.8\n",
            "pandas version: 1.1.5\n",
            "einops version: NOT INSTALLED or UNKNOWN VERSION.\n",
            "\n",
            "For details about installing the optional dependencies, please visit:\n",
            "    https://docs.monai.io/en/latest/installation.html#installing-the-recommended-dependencies\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BvUL9l6t_tDM"
      },
      "source": [
        "**Data Direction**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mHeS8Rt6_zWD"
      },
      "source": [
        "data_dir= \"/content/drive/My Drive/Spleen_AI\""
      ],
      "execution_count": 59,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "O1S2v-B-AITM"
      },
      "source": [
        "#labels_all = pd.read_excel (r'/content/drive/MyDrive/Spleen_AI/labels.xlsx')\n",
        "#l=labels_all[\"label\"].to_numpy()\n",
        "#print(l)\n",
        "#print(labels_all[\"label\"])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "GyjyR96rS0Bl",
        "outputId": "9a998eca-e3ca-464d-c09f-6fb011400ed8"
      },
      "source": [
        "images = sorted(glob(os.path.join(data_dir, \"Projekt2t/\", \"*.nii.gz\")))\n",
        "segmentations = sorted(glob(os.path.join(data_dir, \"Projekt2t/mask/*\", \"*.nii.gz\")))\n",
        "\n",
        "#for i in images:\n",
        " # print(os.path.basename(i))\n",
        "\n",
        "#labels= l\n",
        "labels= np.array([0,0,0,0,0,0,1,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1], dtype=np.int64)\n",
        "print(len(images))\n",
        "print(len(segmentations))\n",
        "print(len(labels))\n"
      ],
      "execution_count": 60,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "23\n",
            "23\n",
            "23\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JgOMkqS2-i4q"
      },
      "source": [
        "**Check Data Shape**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 306
        },
        "id": "B6e1LbkoDgKM",
        "outputId": "fe6ca7a9-fc90-433c-e4da-510d4b1cc4cd"
      },
      "source": [
        "device = torch.device(\"cuda\")\n",
        "#AddChannel(), Orientation(axcodes=\"PRI\"),Flip(spatial_axis=0),\n",
        "transforms=Compose([ Flip(spatial_axis=(0,-1)), ToTensor()])\n",
        "check = ImageDataset(image_files=images[:2], labels=labels[:2],\n",
        "                        transform=transforms)\n",
        "check_loader = DataLoader(check, batch_size=1,\n",
        "                          num_workers=1, pin_memory=torch.cuda.is_available())\n",
        "for data,i in check_loader:\n",
        "  inputs = data[0].to(device)\n",
        "  plt.imshow(data[0][:, :, 35], cmap=\"gray\")\n",
        "  print(inputs.shape)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "torch.Size([512, 512, 50])\n",
            "torch.Size([512, 512, 138])\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AJJwFbO47yGV"
      },
      "source": [
        "**Loading Data**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8iNiaVK_dwM_",
        "outputId": "683ae3f3-a375-45bd-e753-2fe7f5b56adc"
      },
      "source": [
        "train_files = [{\"image\": i, \"seg\": s, \"label\": l} for i, s, l in zip(images[:18], segmentations[:18], labels[:18])]\n",
        "val_files  = [{\"image\": i, \"seg\": s, \"label\": l} for i, s, l in zip(images[18:], segmentations[18:], labels[18:])]\n",
        "\n",
        "transforms = Compose([\n",
        "    LoadImaged(keys=[\"image\", \"seg\"]),\n",
        "    AddChanneld(keys=[\"image\", \"seg\"]),\n",
        "    Spacingd(keys=[\"image\", \"seg\"], pixdim=(1.5, 1.5, 2.0), mode=(\"bilinear\", \"nearest\")),\n",
        "    Orientationd(keys=[\"image\", \"seg\"], axcodes=\"PLI\"),\n",
        "    Resized(keys=[\"image\", \"seg\"], spatial_size=(96, 96, 96)),\n",
        "    ScaleIntensityRanged(keys=['image'], a_min=-57, a_max=164, b_min=0.0, b_max=1.0, clip=True),\n",
        "    MaskIntensityd(keys=\"image\", mask_key=\"seg\"),\n",
        "    EnsureTyped(keys=[\"image\", \"seg\"]),\n",
        "])\n",
        "\n",
        "# create a training data loader\n",
        "train_ds = CacheDataset(data=train_files, transform=transforms)\n",
        "train_loader = DataLoader(train_ds, batch_size=2, shuffle=True, num_workers=4, pin_memory=torch.cuda.is_available())\n",
        "\n",
        "# create a validation data loader\n",
        "val_ds = CacheDataset(data=val_files, transform=transforms)\n",
        "val_loader = DataLoader(val_ds, batch_size=2, shuffle=True, num_workers=4, pin_memory=torch.cuda.is_available())\n",
        "\n",
        "post_pred = Compose([EnsureType(), Activations(softmax=True)])\n",
        "post_label = Compose([EnsureType(), AsDiscrete(to_onehot=True, n_classes=2)])\n"
      ],
      "execution_count": 80,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loading dataset: 100%|██████████| 18/18 [00:39<00:00,  2.18s/it]\n",
            "Loading dataset: 100%|██████████| 5/5 [00:14<00:00,  2.92s/it]\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 434
        },
        "id": "xxkzWn0Yco_B",
        "outputId": "ce6b2ad8-a77f-496d-f3fe-2f94ecf7a234"
      },
      "source": [
        "device = torch.device(\"cuda\")\n",
        "for data in train_loader:\n",
        "  image, label = data[\"image\"].to(device), data[\"seg\"].to(device)\n",
        "  plt.figure(\"check\", (18, 6))\n",
        "  plt.subplot(1, 2, 1)\n",
        "  plt.imshow(data[\"image\"][0, 0, :, :, 50], cmap=\"gray\")\n",
        "  plt.subplot(1, 2, 2)\n",
        "  plt.imshow(data[\"seg\"][0, 0, :, :, 50])\n",
        "\n",
        "  #print(inputs.shape)"
      ],
      "execution_count": 76,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n",
            "Adding an axes using the same arguments as a previous axes currently reuses the earlier instance.  In a future version, a new instance will always be created and returned.  Meanwhile, this warning can be suppressed, and the future behavior ensured, by passing a unique label to each axes instance.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1296x432 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o46nSnXueNbT"
      },
      "source": [
        "for (images, segmentations, labels) in train_loader:\n",
        "      \n",
        "      print(idx,  images.shape, segmentations.shape , labels)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RtjR-k758HU7"
      },
      "source": [
        "**Training**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CGT1akrQ-MKe"
      },
      "source": [
        "# Create DenseNet121, CrossEntropyLoss and Adam optimizer\n",
        "device = torch.device(\"cuda\")\n",
        "model = monai.networks.nets.DenseNet121(spatial_dims=3, in_channels=1, out_channels=2).to(device)\n",
        "loss_function = torch.nn.CrossEntropyLoss()\n",
        "optimizer = torch.optim.Adam(model.parameters(), 1e-5)\n",
        "auc_metric = ROCAUCMetric()\n"
      ],
      "execution_count": 81,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 838
        },
        "id": "eqZk0HD7xqy1",
        "outputId": "c18b5566-1b7a-4fcb-8583-63762e6a69e2"
      },
      "source": [
        "# start a typical PyTorch training\n",
        "val_interval = 2\n",
        "best_metric = -1\n",
        "best_metric_epoch = -1\n",
        "writer = SummaryWriter()\n",
        "max_epochs = 10\n",
        "for epoch in range(max_epochs):\n",
        "        print(\"-\" * 10)\n",
        "        print(f\"epoch {epoch + 1}/{max_epochs}\")\n",
        "        model.train()\n",
        "        epoch_loss = 0\n",
        "        step = 0\n",
        "        for batch_data in train_loader:\n",
        "            step += 1\n",
        "            inputs, labels = batch_data[\"image\"].to(device), batch_data[\"label\"].to(device)\n",
        "            optimizer.zero_grad()\n",
        "            outputs = model(inputs)\n",
        "            loss = loss_function(outputs, labels)\n",
        "            loss.backward()\n",
        "            optimizer.step()\n",
        "            epoch_loss += loss.item()\n",
        "            epoch_len = len(train_ds) // train_loader.batch_size\n",
        "            print(f\"{step}/{epoch_len}, train_loss: {loss.item():.4f}\")\n",
        "            writer.add_scalar(\"train_loss\", loss.item(), epoch_len * epoch + step)\n",
        "        epoch_loss /= step\n",
        "        print(f\"epoch {epoch + 1} average loss: {epoch_loss:.4f}\")\n",
        "\n",
        "        if (epoch + 1) % val_interval == 0:\n",
        "            model.eval()\n",
        "            with torch.no_grad():\n",
        "                y_pred = torch.tensor([], dtype=torch.float32, device=device)\n",
        "                y = torch.tensor([], dtype=torch.long, device=device)\n",
        "                for val_data in val_loader:\n",
        "                    val_images, val_labels = val_data[\"image\"].to(device), val_data[\"label\"].to(device)\n",
        "                    y_pred = torch.cat([y_pred, model(val_images)], dim=0)\n",
        "                    y = torch.cat([y, val_labels], dim=0)\n",
        "\n",
        "                acc_value = torch.eq(y_pred.argmax(dim=1), y)\n",
        "                acc_metric = acc_value.sum().item() / len(acc_value)\n",
        "                y_onehot = [post_label(i) for i in decollate_batch(y)]\n",
        "                y_pred_act = [post_pred(i) for i in decollate_batch(y_pred)]\n",
        "                auc_metric(y_pred_act, y_onehot)\n",
        "                auc_result = auc_metric.aggregate()\n",
        "                auc_metric.reset()\n",
        "                del y_pred_act, y_onehot\n",
        "                if acc_metric > best_metric:\n",
        "                    best_metric = acc_metric\n",
        "                    best_metric_epoch = epoch + 1\n",
        "                    torch.save(model.state_dict(), \"/content/drive/My Drive/Spleen_AI/best_metric_model_classification3d.pth\")\n",
        "                    print(\"saved new best metric model\")\n",
        "                print(\n",
        "                    \"current epoch: {} current accuracy: {:.4f} current AUC: {:.4f} best accuracy: {:.4f} at epoch {}\".format(\n",
        "                        epoch + 1, acc_metric, auc_result, best_metric, best_metric_epoch\n",
        "                    )\n",
        "                )\n",
        "                writer.add_scalar(\"val_accuracy\", acc_metric, epoch + 1)\n",
        "print(f\"train completed, best_metric: {best_metric:.4f} at epoch: {best_metric_epoch}\")\n",
        "writer.close()"
      ],
      "execution_count": 82,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "----------\n",
            "epoch 1/10\n",
            "1/9, train_loss: 0.6220\n",
            "2/9, train_loss: 0.6585\n",
            "3/9, train_loss: 0.6015\n",
            "4/9, train_loss: 0.7530\n",
            "5/9, train_loss: 0.6869\n",
            "6/9, train_loss: 0.6901\n",
            "7/9, train_loss: 0.6806\n",
            "8/9, train_loss: 0.6628\n",
            "9/9, train_loss: 0.5868\n",
            "epoch 1 average loss: 0.6602\n",
            "----------\n",
            "epoch 2/10\n",
            "1/9, train_loss: 0.6469\n",
            "2/9, train_loss: 0.5941\n",
            "3/9, train_loss: 0.8241\n",
            "4/9, train_loss: 0.6381\n",
            "5/9, train_loss: 0.5654\n",
            "6/9, train_loss: 0.6721\n",
            "7/9, train_loss: 0.5728\n",
            "8/9, train_loss: 0.5792\n",
            "9/9, train_loss: 0.5696\n",
            "epoch 2 average loss: 0.6291\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "error",
          "ename": "AssertionError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mAssertionError\u001b[0m                            Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-82-35cb825d6bff>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     41\u001b[0m                 \u001b[0my_pred_act\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mpost_pred\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mdecollate_batch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     42\u001b[0m                 \u001b[0mauc_metric\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred_act\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_onehot\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 43\u001b[0;31m                 \u001b[0mauc_result\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mauc_metric\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0maggregate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     44\u001b[0m                 \u001b[0mauc_metric\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreset\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     45\u001b[0m                 \u001b[0;32mdel\u001b[0m \u001b[0my_pred_act\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_onehot\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/monai/metrics/rocauc.py\u001b[0m in \u001b[0;36maggregate\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m     60\u001b[0m             \u001b[0;32mraise\u001b[0m \u001b[0mValueError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"y_pred and y must be PyTorch Tensor.\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     61\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 62\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mcompute_roc_auc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0my_pred\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maverage\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0maverage\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     63\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     64\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/monai/metrics/rocauc.py\u001b[0m in \u001b[0;36mcompute_roc_auc\u001b[0;34m(y_pred, y, average)\u001b[0m\n\u001b[1;32m    151\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0m_calculate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mflatten\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mflatten\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    152\u001b[0m     \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_pred\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtranspose\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_pred\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtranspose\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 153\u001b[0;31m     \u001b[0mauc_values\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0m_calculate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred_\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0my_pred_\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mzip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    154\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0maverage\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0mAverage\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mNONE\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    155\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mauc_values\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/monai/metrics/rocauc.py\u001b[0m in \u001b[0;36m<listcomp>\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m    151\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0m_calculate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mflatten\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mflatten\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    152\u001b[0m     \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_pred\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtranspose\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_pred\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtranspose\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 153\u001b[0;31m     \u001b[0mauc_values\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0m_calculate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred_\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0my_pred_\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mzip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_pred\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    154\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0maverage\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0mAverage\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mNONE\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    155\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mauc_values\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/monai/metrics/rocauc.py\u001b[0m in \u001b[0;36m_calculate\u001b[0;34m(y_pred, y)\u001b[0m\n\u001b[1;32m     67\u001b[0m         \u001b[0;32mraise\u001b[0m \u001b[0mAssertionError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"y and y_pred must be 1 dimension data with same length.\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     68\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munique\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mequal\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtensor\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdtype\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdtype\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdevice\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdevice\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 69\u001b[0;31m         \u001b[0;32mraise\u001b[0m \u001b[0mAssertionError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"y values must be 0 or 1, can not be all 0 or all 1.\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     70\u001b[0m     \u001b[0mn\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     71\u001b[0m     \u001b[0mindices\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0my_pred\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0margsort\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mAssertionError\u001b[0m: y values must be 0 or 1, can not be all 0 or all 1."
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZIsk3v3gKTrB"
      },
      "source": [
        "**Evaluation Metrics**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-LyP68KcHELg"
      },
      "source": [
        "from enum import Enum\n",
        "from sklearn.metrics import (\n",
        "    classification_report, confusion_matrix,\n",
        "    ConfusionMatrixDisplay\n",
        ")\n",
        "class Diagnosis(Enum):\n",
        "    Liver_Cirrhosis = 0\n",
        "    Lymphoma = 1\n",
        "    \n",
        "\n",
        "%matplotlib inline\n",
        "# Load best model\n",
        "model.load_state_dict(torch.load(\"/content/drive/My Drive/Spleen_AI/best_metric_model_classification3d_array.pth\"))\n",
        "model.to(device)\n",
        "model.eval()\n",
        "\n",
        "with torch.no_grad():\n",
        "    y_pred = torch.tensor([], dtype=torch.float32, device=device)\n",
        "    y = torch.tensor([], dtype=torch.long, device=device)\n",
        "\n",
        "    for val_data in val_loader:\n",
        "        val_images, val_labels = val_data[0].to(\n",
        "            device), val_data[1].to(device),\n",
        "\n",
        "        outputs = model(val_images)\n",
        "        y_pred = torch.cat([y_pred, outputs.argmax(dim=1)], dim=0)\n",
        "        y = torch.cat([y, val_labels], dim=0)\n",
        "\n",
        "print(classification_report(\n",
        "    y.cpu().numpy(),\n",
        "    y_pred.cpu().numpy(),\n",
        "    target_names=[d.name for d in Diagnosis]))\n",
        "\n",
        "cm = confusion_matrix (\n",
        "    y.cpu().numpy(),\n",
        "    y_pred.cpu().numpy(),\n",
        "    normalize='true',\n",
        ")\n",
        "cmm = ConfusionMatrixMetric (include_background=True, metric_name=[\"sensitivity\",\"specificity\",\"negative predictive value\"])\n",
        "\n",
        "disp = ConfusionMatrixDisplay(\n",
        "    confusion_matrix=cm,\n",
        "    display_labels=[d.name for d in Diagnosis],\n",
        ")\n",
        "disp.plot(ax=plt.subplots(1, 1, facecolor='white')[1])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Nm0LXsvlgp_N"
      },
      "source": [
        "files = [{\"img\": img} for img in images]\n",
        "post_transforms = Compose([\n",
        "        #Activationsd(keys=\"pred\", sigmoid=True),\n",
        "        #AsDiscreted(keys=\"pred\", threshold_values=True),\n",
        "        Invertd(\n",
        "            keys=\"y_pred\",  # invert the `pred` data field, also support multiple fields\n",
        "            transform=train_transforms,\n",
        "            loader=val_loader,\n",
        "            orig_keys=\"files\",  # get the previously applied pre_transforms information on the `img` data field,\n",
        "                              # then invert `pred` based on this information. we can use same info\n",
        "                              # for multiple fields, also support different orig_keys for different fields\n",
        "            #meta_keys=\"pred_meta_dict\",  # key field to save inverted meta data, every item maps to `keys`\n",
        "            #orig_meta_keys=\"img_meta_dict\",  # get the meta data from `img_meta_dict` field when inverting,\n",
        "                                             # for example, may need the `affine` to invert `Spacingd` transform,\n",
        "                                             # multiple fields can use the same meta data to invert\n",
        "            #meta_key_postfix=\"meta_dict\",  # if `meta_keys=None`, use \"{keys}_{meta_key_postfix}\" as the meta key,\n",
        "                                           # if `orig_meta_keys=None`, use \"{orig_keys}_{meta_key_postfix}\",\n",
        "                                           # otherwise, no need this arg during inverting\n",
        "            nearest_interp=True,  # change to use \"nearest\" mode in interpolation when inverting\n",
        "            to_tensor=True,  # convert to PyTorch Tensor after inverting\n",
        "        ),\n",
        "        #SaveImaged(keys=\"pred\", meta_keys=\"pred_meta_dict\", output_dir=\"./out\", output_postfix=\"seg\", resample=False),\n",
        "    ])\n",
        "\n",
        "test_ds = ImageDataset(\n",
        "    image_files=files, labels=labels, transform=post_transforms)\n",
        "test_loader = DataLoader(test_ds, batch_size=1, num_workers=4)\n",
        "itera = iter(test_loader)\n",
        "\n",
        "\n",
        "def get_next_im():\n",
        "    test_data = next(itera)\n",
        "    return test_data[0].to(device), test_data[1].unsqueeze(0).to(device)\n",
        "\n",
        "\n",
        "def plot_occlusion_heatmap(im, heatmap):\n",
        "    plt.subplots(1, 2)\n",
        "    plt.subplot(1, 2, 1)\n",
        "    plt.imshow(np.squeeze(im.cpu()))\n",
        "    plt.colorbar()\n",
        "    plt.subplot(1, 2, 2)\n",
        "    plt.imshow(heatmap)\n",
        "    plt.colorbar()\n",
        "    plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vfFYizJOg1TG"
      },
      "source": [
        "# Get a random image and its corresponding label\n",
        "img, label = get_next_im()\n",
        "\n",
        "# Get the occlusion sensitivity map\n",
        "occ_sens = monai.visualize.OcclusionSensitivity(\n",
        "    nn_module=model, mask_size=12, n_batch=10, stride=12)\n",
        "# Get the CAM\n",
        "cam = monai.visualize.GradCAM(\n",
        "    nn_module=model, target_layers=\"class_layers.relu\"\n",
        ")\n",
        "# Only get a single slice to save time.\n",
        "# For the other dimensions (channel, width, height), use\n",
        "# -1 to use 0 and img.shape[x]-1 for min and max, respectively\n",
        "depth_slice = img.shape[2] // 2\n",
        "\n",
        "occ_sens_b_box = [-1, -1, depth_slice, depth_slice, -1, -1, -1, -1]\n",
        "\n",
        "occ_result, _ = occ_sens(x=img, b_box=occ_sens_b_box)\n",
        "occ_result = occ_result[..., label.item()]\n",
        "cam_result = cam(x=img, class_idx=None)\n",
        "cam_result = cam_result[..., depth_slice]\n",
        "\n",
        "\n",
        "fig, axes = plt.subplots(1, 3, figsize=(25, 15), facecolor='white')\n",
        "\n",
        "for i, im in enumerate([img[:, :, depth_slice, ...], occ_result, cam_result]):\n",
        "    cmap = 'gray' if i == 0 else 'jet'\n",
        "    ax = axes[i]\n",
        "    if isinstance(im, torch.Tensor):\n",
        "        im = im.cpu().detach()    \n",
        "    #im_show = ax.imshow(im[0][0], cmap=cmap)\n",
        "    im_show = ax.imshow(np.squeeze(im[0][0]), cmap=cmap)\n",
        "    ax.axis('off')\n",
        "    fig.colorbar(im_show, ax=ax)\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}