Bone_Age_YOLO.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "authorship_tag": "ABX9TyNSygy1+2F9bf2esTT18xBk",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "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/Armin-Abdollahi/Bone-Age-Estimation/blob/main/Bone_Age_YOLO.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "AoQxbct7jWOa"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.layers import Input, Conv2D, MaxPooling2D, Flatten, Dense, Dropout, Reshape, Multiply\n",
        "from tensorflow.keras.models import Model"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Load the labels\n",
        "labels_path = '/content/drive/MyDrive/Digital_Hand_Atlas_Dataset/Total_Images/Bordered_XRay_Images/Label.csv'\n",
        "labels = pd.read_csv(labels_path)"
      ],
      "metadata": {
        "id": "UgEZLrFljgw4"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "labels['Case ID'] = labels['Case ID'].apply(lambda x: str(x)+'.jpg')\n",
        "\n",
        "labels.head()"
      ],
      "metadata": {
        "id": "qJWoiCSAjh-d"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Ensure the 'Case ID' column is of type string\n",
        "labels['Case ID'] = labels['Case ID'].astype(str)"
      ],
      "metadata": {
        "id": "xoQZSF68jjLh"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Create the image data generator\n",
        "datagen = ImageDataGenerator(rescale=1./255, validation_split=0.2)"
      ],
      "metadata": {
        "id": "wQXqP8NEjkXd"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Create the train and validation generators\n",
        "train_generator = datagen.flow_from_dataframe(\n",
        "    dataframe=labels,\n",
        "    directory='/content/drive/MyDrive/Digital_Hand_Atlas_Dataset/Total_Images/Bordered_XRay_Images',\n",
        "    x_col='Case ID',\n",
        "    y_col='Bone Age',\n",
        "    target_size=(256, 256),\n",
        "    color_mode='rgb',  # Change to 'rgb' for 3 channels\n",
        "    class_mode='raw',\n",
        "    batch_size=32,\n",
        "    subset='training'\n",
        ")\n",
        "\n",
        "validation_generator = datagen.flow_from_dataframe(\n",
        "    dataframe=labels,\n",
        "    directory='/content/drive/MyDrive/Digital_Hand_Atlas_Dataset/Total_Images/Bordered_XRay_Images',\n",
        "    x_col='Case ID',\n",
        "    y_col='Bone Age',\n",
        "    target_size=(256, 256),\n",
        "    color_mode='rgb',  # Change to 'rgb' for 3 channels\n",
        "    class_mode='raw',\n",
        "    batch_size=32,\n",
        "    subset='validation'\n",
        ")"
      ],
      "metadata": {
        "id": "j5QqjW6bjlbu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Define the YOLO model architecture for grayscale images\n",
        "def create_yolo_model(input_shape):\n",
        "    inputs = Input(input_shape)\n",
        "\n",
        "    # Your existing CNN layers adapted for single-channel input\n",
        "    x = Conv2D(32, (3,3), activation='relu')(inputs)\n",
        "    x = MaxPooling2D(2, 2)(x)\n",
        "    x = Conv2D(64, (3,3), activation='relu')(x)\n",
        "    x = MaxPooling2D(2,2)(x)\n",
        "    x = Conv2D(128, (3,3), activation='relu')(x)\n",
        "    x = MaxPooling2D(2,2)(x)\n",
        "\n",
        "    # Flatten and apply dropout\n",
        "    x = Flatten()(x)\n",
        "    x = Dense(512, activation='relu')(x)\n",
        "    x = Dropout(0.5)(x)\n",
        "\n",
        "    # Output layer for YOLO, reshaped to grid size\n",
        "    # Assuming 3x3 grid, 2 bounding boxes per grid cell, and 1 class (hand)\n",
        "    output = Dense(3 * 3 * 2 * (4 + 1))(x)  # 4 coordinates + 1 class confidence\n",
        "    output = Reshape((3, 3, 2, 5))(output)\n",
        "\n",
        "    return Model(inputs, output)"
      ],
      "metadata": {
        "id": "X6GcfrlyjmrU"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Create the YOLO model for grayscale images\n",
        "yolo_model = create_yolo_model((256, 256, 1))"
      ],
      "metadata": {
        "id": "2qt3-rObjoEM"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Define the attention mechanism\n",
        "def attention_mechanism(feature_map):\n",
        "    # Apply attention to the feature map\n",
        "    attention_map = Conv2D(1, (1, 1), activation='sigmoid')(feature_map)\n",
        "    attended_features = Multiply()([feature_map, attention_map])\n",
        "    return attended_features"
      ],
      "metadata": {
        "id": "f1HAl86bjpg6"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Add the attention mechanism to the model\n",
        "inputs = Input((256, 256, 1))\n",
        "features = yolo_model(inputs)\n",
        "attended_features = attention_mechanism(features)"
      ],
      "metadata": {
        "id": "_H9S8E2LjqfE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Final model with attention\n",
        "final_model = Model(inputs, attended_features)"
      ],
      "metadata": {
        "id": "vLUGDhIljrgu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Compile the model\n",
        "final_model.compile(optimizer=Adam(), loss='mean_squared_error', metrics=['mae', 'accuracy'])"
      ],
      "metadata": {
        "id": "86F3UwLijss3"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Train the model\n",
        "history = final_model.fit(\n",
        "    train_generator,\n",
        "    steps_per_epoch=train_generator.samples // 32,\n",
        "    validation_data=validation_generator,\n",
        "    validation_steps=validation_generator.samples // 32,\n",
        "    epochs=10\n",
        ")"
      ],
      "metadata": {
        "id": "G3Qlo-5bjtwx"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Plot training & validation MAE values\n",
        "plt.figure(figsize=(12, 4))\n",
        "plt.subplot(1, 2, 1)\n",
        "plt.plot(history.history['mae'])\n",
        "plt.plot(history.history['val_mae'])\n",
        "plt.title('Model MAE')\n",
        "plt.ylabel('MAE')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Validation'], loc='upper left')\n",
        "\n",
        "# Plot training & validation loss values\n",
        "plt.subplot(1, 2, 2)\n",
        "plt.plot(history.history['loss'])\n",
        "plt.plot(history.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Validation'], loc='upper left')\n",
        "\n",
        "plt.show()"
      ],
      "metadata": {
        "id": "JpdQFCX3jvZP"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Train the model\n",
        "history = final_model.fit(\n",
        "    train_generator,\n",
        "    steps_per_epoch=train_generator.samples // 32,\n",
        "    validation_data=validation_generator,\n",
        "    validation_steps=validation_generator.samples // 32,\n",
        "    epochs=18\n",
        ")"
      ],
      "metadata": {
        "id": "f9uSOQO8jwiC"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Plot training & validation MAE values\n",
        "plt.figure(figsize=(18, 4))\n",
        "\n",
        "plt.subplot(1, 3, 1)\n",
        "plt.plot(history.history['mae'])\n",
        "plt.plot(history.history['val_mae'])\n",
        "plt.title('Model MAE')\n",
        "plt.ylabel('MAE')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Validation'], loc='upper left')\n",
        "\n",
        "# Plot training & validation loss values\n",
        "plt.subplot(1, 3, 2)\n",
        "plt.plot(history.history['loss'])\n",
        "plt.plot(history.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Validation'], loc='upper left')\n",
        "\n",
        "# Plot training & validation accuracy values\n",
        "plt.subplot(1, 3, 3)\n",
        "plt.plot(history.history['accuracy'])\n",
        "plt.plot(history.history['val_accuracy'])\n",
        "plt.title('Model Accuracy')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Validation'], loc='upper left')\n",
        "\n",
        "plt.tight_layout()\n",
        "plt.show()"
      ],
      "metadata": {
        "id": "-Kjp3Xigjxnh"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}