scripts/Composite_4_class.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Composite_4_class.ipynb",
      "provenance": [],
      "toc_visible": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "V1XJ1h0ndT-3",
        "colab_type": "text"
      },
      "source": [
        "The 4-class task can be decomposed in two subtasks: mass-calcification classification and benign-malignant diagnosing. The idea is to train two models independently, each focusing on one subtask only; the two networks will then work in parallel, each producing one half of the full 4-class prediction. The rationale behind this *divide et impera* approach is that each network  specializes itself to solve a smaller and potentially easier subproblem, thus getting better results than addressing the whole problem in one fell swoop."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mJlLAXApFZZg",
        "colab_type": "text"
      },
      "source": [
        "# Environment setup"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_TOBgq9wEJx7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Connect to Google Drive\n",
        "\n",
        "from google.colab import drive\n",
        "drive.mount('/content/gdrive')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TnmYQR1TFYrj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Copy the dataset from Google Drive to local\n",
        "\n",
        "!cp \"/content/gdrive/My Drive/CBIS_DDSM.zip\" .\n",
        "!unzip -qq CBIS_DDSM.zip\n",
        "!rm CBIS_DDSM.zip\n",
        "cbis_path = 'CBIS_DDSM'"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "R2zXaceOFjaI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Import libraries\n",
        "\n",
        "%tensorflow_version 1.x\n",
        "import os\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "import itertools\n",
        "from sklearn.metrics import confusion_matrix, roc_curve, auc, classification_report\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras import layers\n",
        "from tensorflow.keras import models\n",
        "from tensorflow.keras import optimizers\n",
        "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint, LearningRateScheduler, Callback\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "from tensorflow.keras.optimizers import RMSprop\n",
        "from tensorflow.keras.regularizers import l2\n",
        "from tensorflow.keras.utils import plot_model\n",
        "from tensorflow.keras.utils import to_categorical\n",
        "from tensorflow.keras.applications import VGG16\n",
        "from tensorflow.keras.applications.vgg16 import preprocess_input"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rBz7pSbCFrCj",
        "colab_type": "text"
      },
      "source": [
        "# Data pre-processing"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9pIRh0p_FsU3",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def load_testing():\n",
        "    \"\"\"\n",
        "    Load the test set (abnormalities patches and labels, no baseline)\n",
        "    \"\"\"\n",
        "    images = np.load(os.path.join(cbis_path, 'numpy data', 'public_test_tensor.npy'))[1::2]\n",
        "    labels = np.load(os.path.join(cbis_path, 'numpy data', 'public_test_labels.npy'))[1::2]\n",
        "    return images, labels\n",
        "\n",
        "\n",
        "def remap_label_type(l):\n",
        "    \"\"\"\n",
        "    Remap the labels to:\n",
        "        0 -> mass \n",
        "        1 -> calcification\n",
        "    \"\"\"\n",
        "    if l == 1 or l == 2:\n",
        "        return 0\n",
        "    elif l == 3 or l == 4:\n",
        "        return 1\n",
        "    else:\n",
        "        print(\"[WARN] Unrecognized label (%d)\" % l)\n",
        "        return None\n",
        "\n",
        "\n",
        "def remap_label_severity(l):\n",
        "    \"\"\"\n",
        "    Remap the labels to 0->benign 1->malignant\n",
        "    \"\"\"\n",
        "    if l == 1 or l == 3:\n",
        "        return 0\n",
        "    elif l == 2 or l == 4:\n",
        "        return 1\n",
        "    else:\n",
        "        print(\"[WARN] Unrecognized label (%d)\" % l)\n",
        "        return None\n",
        "\n",
        "\n",
        "def combine_labels_type_severity(ltype, lsev):\n",
        "    if ltype == 0 and lsev == 0:\n",
        "        return 1\n",
        "    elif ltype == 0 and lsev == 1:\n",
        "        return 2\n",
        "    elif ltype == 1 and lsev == 0:\n",
        "        return 3\n",
        "    elif ltype == 1 and lsev == 1:\n",
        "        return 4\n",
        "    else:\n",
        "        print(\"[WARN] Unrecognized labels (%d %d)\" % (ltype, lsev))\n",
        "        return None"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jvJXn-wLHUbb",
        "colab_type": "code",
        "outputId": "ec3b749b-ef78-42e7-ca2a-30b89e009dc8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "# Load training and test images (abnormalities only, no baseline)\n",
        "test_images, test_labels= load_testing()\n",
        "test_images_vgg, _ = load_testing()\n",
        "\n",
        "# Number of images\n",
        "n_test_img = test_images.shape[0]\n",
        "print(\"Test size: %d\" % (n_test_img))\n",
        "\n",
        "# Compute width and height of images\n",
        "img_w = test_images.shape[1]\n",
        "img_h = test_images.shape[2]\n",
        "print(\"Image size: %dx%d\" % (img_w, img_h))\n",
        "\n",
        "test_labels_raw = test_labels.copy()\n",
        "\n",
        "# Remap labels for mass-calc classification\n",
        "test_labels_type = np.array([remap_label_type(l) for l in test_labels])\n",
        "\n",
        "# Remap labels for benign-malignant classification\n",
        "test_labels_severity = np.array([remap_label_severity(l) for l in test_labels])\n",
        "\n",
        "# Create a new dimension for color in the images arrays\n",
        "test_images = test_images.reshape((n_test_img, img_w, img_h, 1))\n",
        "test_images_vgg = test_images_vgg.reshape((n_test_img, img_w, img_h, 1))\n",
        "\n",
        "# Rescale values to (0-1), or (0-255) for VGG\n",
        "test_images = test_images.astype('uint16') / 65535\n",
        "test_images_vgg = test_images_vgg.astype('uint16') / 256\n",
        "\n",
        "# Replicate the only channel 3 times for VGG16\n",
        "test_images_vgg = np.repeat(test_images_vgg, 3, axis=3)\n",
        "\n",
        "# Normalize VGG images\n",
        "preprocess_input(test_images_vgg);"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Test size: 336\n",
            "Image size: 150x150\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hQFE8jqsenTA",
        "colab_type": "text"
      },
      "source": [
        "# Composite model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NFir0kYRLK0r",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!cp \"/content/gdrive/My Drive/models/vgg16_ft_2layer_2cl_best.h5\" .\n",
        "!cp \"/content/gdrive/My Drive/models/model_3_benmal_best.h5\" .\n",
        "\n",
        "# Model for mass-calcification classification\n",
        "model_type = models.load_model('vgg16_ft_2layer_2cl_best.h5');\n",
        "# Model for benign-malignant classification\n",
        "model_severity = models.load_model('model_3_benmal_best.h5');"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hvUdYIqhNqSc",
        "colab_type": "code",
        "outputId": "980bfc55-0923-4e5e-d93e-d22db88a50ea",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "# Accuracy of the individual models on their respective tasks\n",
        "_, acc_type = model_type.evaluate(test_images_vgg, test_labels_type, verbose=False)\n",
        "_, acc_severity = model_severity.evaluate(test_images, test_labels_severity, verbose=False)\n",
        "\n",
        "pred_type = np.squeeze(model_type.predict_classes(test_images_vgg)).tolist()\n",
        "pred_severity = np.squeeze(model_severity.predict_classes(test_images)).tolist()\n",
        "\n",
        "print(\"Standalone type accuracy: %.4f\" % acc_type)\n",
        "print(\"Standalone severity accuracy: %.4f\" % acc_severity)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Standalone type accuracy: 0.9137\n",
            "Standalone severity accuracy: 0.6905\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "j-4FvmxFQsF7",
        "colab_type": "code",
        "outputId": "d3398114-e6aa-41ea-a6df-293b0213e8ef",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 221
        }
      },
      "source": [
        "# Get a 4-class prediction by combining two 2-class predictions\n",
        "pred_combined = [combine_labels_type_severity(a, b) for a,b in zip(pred_type, pred_severity)]\n",
        "\n",
        "combined_right = sum([p == l for p, l in zip(pred_combined, test_labels_raw)])\n",
        "combined_acc = combined_right / n_test_img\n",
        "\n",
        "print(\"Accuracy: %.4f\" % combined_acc)\n",
        "print(classification_report(test_labels_raw, pred_combined, digits=4))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracy: 0.6250\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           1     0.6961    0.6283    0.6605       113\n",
            "           2     0.5135    0.5758    0.5429        66\n",
            "           3     0.6975    0.7830    0.7378       106\n",
            "           4     0.4390    0.3529    0.3913        51\n",
            "\n",
            "    accuracy                         0.6250       336\n",
            "   macro avg     0.5865    0.5850    0.5831       336\n",
            "weighted avg     0.6216    0.6250    0.6209       336\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VLo86FRoRea7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def plot_confusion_matrix(cm, classes, normalize=False, title='Confusion matrix', cmap=plt.cm.Blues):\n",
        "\n",
        "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
        "    plt.title(title)\n",
        "    plt.colorbar()\n",
        "    tick_marks = np.arange(len(classes))\n",
        "    plt.xticks(tick_marks, classes, rotation=45)\n",
        "    plt.yticks(tick_marks, classes)\n",
        "\n",
        "    if normalize:\n",
        "        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
        "\n",
        "    thresh = cm.max() / 2.\n",
        "    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n",
        "        plt.text(j, i, cm[i, j],\n",
        "                 horizontalalignment=\"center\",\n",
        "                 color=\"white\" if cm[i, j] > thresh else \"black\")\n",
        "\n",
        "    plt.tight_layout()\n",
        "    plt.ylabel('True label')\n",
        "    plt.xlabel('Predicted label')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6gQkqqoFRerw",
        "colab_type": "code",
        "outputId": "067ffb01-1b78-46c2-a070-cc39594eff89",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 311
        }
      },
      "source": [
        "confusion_mtx = confusion_matrix(test_labels_raw, pred_combined) \n",
        "plot_confusion_matrix(confusion_mtx, classes=range(4), title='Composite model confusion matrix')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUUAAAEmCAYAAAD1FIKpAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}