{"run app.py": "import os\r\nimport numpy as np\r\nimport matplotlib.pyplot as plt\r\nimport tensorflow as tf\r\nfrom tensorflow.keras.models import load_model\r\nfrom tensorflow.keras.preprocessing import image\r\nfrom tensorflow.keras.applications.resnet50 import ResNet50, preprocess_input as resnet_preprocess\r\nfrom tensorflow.keras.applications.vgg16 import VGG16, preprocess_input as vgg_preprocess\r\nfrom tensorflow.keras.applications.inception_v3 import InceptionV3, preprocess_input as inception_preprocess\r\nimport time\r\nimport streamlit as st\r\nfrom PIL import Image\r\nimport pandas as pd\r\nimport seaborn as sns\r\nimport requests\r\nimport io\r\nfrom sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score\r\n\r\n# Set page configuration\r\nst.set_page_config(\r\n page_title=\"Brain Disorder ML Analysis\",\r\n page_icon=\"\ud83e\udde0\",\r\n layout=\"wide\",\r\n initial_sidebar_state=\"expanded\"\r\n)\r\n\r\n# Add custom CSS\r\nst.markdown(\"\"\"\r\n\r\n\"\"\", unsafe_allow_html=True)\r\n\r\n# Title and Introduction\r\nst.markdown(\"

Comparative Analysis of ML Models on Brain Disorder Image Datasets

\", unsafe_allow_html=True)\r\n\r\nst.markdown(\"\"\"\r\nThis application allows you to analyze brain MRI images using different machine learning models \r\nand compare their performance across various brain disorders including:\r\n- Alzheimer's Disease (No/Very Mild/Mild/Moderate Impairment)\r\n- Brain Tumors (Glioma, Meningioma, No Tumor, Pituitary)\r\n- Parkinson's Disease\r\n\"\"\")\r\n\r\n# Initialize session state for storing results\r\nif 'results' not in st.session_state:\r\n st.session_state.results = {}\r\n\r\n# Create sidebar for model selection\r\nst.sidebar.title(\"Settings\")\r\n\r\n# Model selection\r\navailable_models = {\r\n \"AlexNet\": \"alexnet\",\r\n \"ResNet-50\": \"resnet50\",\r\n \"VGG-16\": \"vgg16\",\r\n \"Inception V3\": \"inceptionv3\"\r\n}\r\n\r\nselected_models = st.sidebar.multiselect(\r\n \"Select Models to Compare\",\r\n list(available_models.keys()),\r\n default=[\"ResNet-50\", \"VGG-16\"]\r\n)\r\n\r\n# Disorder selection\r\navailable_disorders = {\r\n \"Alzheimer's\": {\r\n \"classes\": [\"No Impairment\", \"Very Mild Impairment\", \"Mild Impairment\", \"Moderate Impairment\"],\r\n \"dataset_size\": \"5000+ images\"\r\n },\r\n \"Brain Tumor\": {\r\n \"classes\": [\"Glioma\", \"Meningioma\", \"No Tumor\", \"Pituitary\"],\r\n \"dataset_size\": \"7023 images\"\r\n },\r\n \"Parkinson's\": {\r\n \"classes\": [\"Parkinson's\", \"Normal\"],\r\n \"dataset_size\": \"830 images\"\r\n }\r\n}\r\n\r\nselected_disorders = st.sidebar.multiselect(\r\n \"Select Disorders to Analyze\",\r\n list(available_disorders.keys()),\r\n default=[\"Brain Tumor\"]\r\n)\r\n\r\n# Advanced settings\r\nst.sidebar.markdown(\"---\")\r\nst.sidebar.markdown(\"

Advanced Settings

\", unsafe_allow_html=True)\r\ndisplay_metrics = st.sidebar.checkbox(\"Show Detailed Metrics\", value=True)\r\nshow_comparison_charts = st.sidebar.checkbox(\"Show Comparison Charts\", value=True)\r\n\r\n# Main content area split into two columns\r\ncol1, col2 = st.columns([1, 2])\r\n\r\n# File uploader in the first column\r\nwith col1:\r\n st.markdown(\"

Upload Brain MRI Image

\", unsafe_allow_html=True)\r\n uploaded_file = st.file_uploader(\"Choose an image...\", type=[\"jpg\", \"jpeg\", \"png\"])\r\n \r\n if uploaded_file is not None:\r\n image_pil = Image.open(uploaded_file)\r\n st.image(image_pil, caption=\"Uploaded MRI Image\", use_column_width=True)\r\n \r\n # Function to preprocess the image for different models\r\n def preprocess_image(img, model_name, target_size=(224, 224)):\r\n # Ensure the image is in RGB format\r\n if img.mode != 'RGB':\r\n img = img.convert('RGB')\r\n\r\n # Resize the image\r\n img_resized = img.resize(target_size)\r\n\r\n # Convert to array and add batch dimension\r\n img_array = image.img_to_array(img_resized)\r\n img_array = np.expand_dims(img_array, axis=0)\r\n\r\n # Apply model-specific preprocessing\r\n if model_name == \"resnet50\":\r\n return resnet_preprocess(img_array)\r\n elif model_name == \"vgg16\":\r\n return vgg_preprocess(img_array)\r\n elif model_name == \"inceptionv3\":\r\n return inception_preprocess(img_array)\r\n else: # Default: normalize to [0, 1]\r\n return img_array / 255.0\r\n\r\n# Second column for results and visualizations\r\nwith col2:\r\n if uploaded_file is not None and selected_models and selected_disorders:\r\n st.markdown(\"

Analysis Results

\", unsafe_allow_html=True)\r\n \r\n # Progress bar for analysis\r\n progress_bar = st.progress(0)\r\n status_text = st.empty()\r\n \r\n # Initialize results dictionary\r\n results = {}\r\n \r\n # Analyze image with selected models and disorders\r\n total_steps = len(selected_models) * len(selected_disorders)\r\n current_step = 0\r\n \r\n for model_name in selected_models:\r\n model_key = available_models[model_name]\r\n results[model_name] = {}\r\n \r\n for disorder in selected_disorders:\r\n current_step += 1\r\n progress = current_step / total_steps\r\n progress_bar.progress(progress)\r\n status_text.text(f\"Analyzing with {model_name} for {disorder}...\")\r\n \r\n # Simulate loading a pre-trained model\r\n time.sleep(1) # Simulating model loading time\r\n \r\n # Preprocess the image\r\n preprocessed_img = preprocess_image(image_pil, model_key)\r\n \r\n # Simulate prediction\r\n time_start = time.time()\r\n \r\n # In a real implementation, you would use an actual model:\r\n # model = load_your_model(model_key, disorder)\r\n # predictions = model.predict(preprocessed_img)\r\n \r\n # Simulating prediction results\r\n prediction_time = time.time() - time_start\r\n \r\n # Simulate prediction results based on the disorder\r\n if disorder == \"Alzheimer's\":\r\n class_names = available_disorders[disorder][\"classes\"]\r\n # Mock predictions - in real app, use actual model predictions\r\n confidences = np.random.dirichlet(np.ones(len(class_names)))\r\n predicted_class = class_names[np.argmax(confidences)]\r\n accuracy = 0.85 + np.random.random() * 0.1 # 85-95% accuracy\r\n \r\n elif disorder == \"Brain Tumor\":\r\n class_names = available_disorders[disorder][\"classes\"]\r\n # Mock predictions\r\n confidences = np.random.dirichlet(np.ones(len(class_names)))\r\n predicted_class = class_names[np.argmax(confidences)]\r\n accuracy = 0.88 + np.random.random() * 0.11 # 88-99% accuracy\r\n \r\n elif disorder == \"Parkinson's\":\r\n class_names = available_disorders[disorder][\"classes\"]\r\n # Mock predictions\r\n confidences = np.random.dirichlet(np.ones(len(class_names)))\r\n predicted_class = class_names[np.argmax(confidences)]\r\n accuracy = 0.82 + np.random.random() * 0.15 # 82-97% accuracy\r\n \r\n # Calculate additional metrics (simulated)\r\n precision = accuracy - 0.02 + np.random.random() * 0.04\r\n recall = accuracy - 0.03 + np.random.random() * 0.06\r\n f1 = 2 * (precision * recall) / (precision + recall)\r\n \r\n # Store results\r\n results[model_name][disorder] = {\r\n \"predicted_class\": predicted_class,\r\n \"confidences\": confidences,\r\n \"class_names\": class_names,\r\n \"accuracy\": accuracy,\r\n \"precision\": precision,\r\n \"recall\": recall,\r\n \"f1_score\": f1,\r\n \"prediction_time\": prediction_time\r\n }\r\n \r\n # Save results to session state\r\n st.session_state.results = results\r\n \r\n # Clear progress indicators\r\n progress_bar.empty()\r\n status_text.empty()\r\n \r\n # Display results\r\n st.markdown(\"

Results Summary

\", unsafe_allow_html=True)\r\n \r\n # Create tabs for each disorder\r\n tabs = st.tabs(selected_disorders)\r\n \r\n for i, disorder in enumerate(selected_disorders):\r\n with tabs[i]:\r\n st.write(f\"### {disorder} Analysis\")\r\n \r\n # Create a dataframe for all models' results for this disorder\r\n model_results = []\r\n for model_name in selected_models:\r\n res = results[model_name][disorder]\r\n model_results.append({\r\n \"Model\": model_name,\r\n \"Predicted Class\": res[\"predicted_class\"],\r\n \"Confidence\": f\"{np.max(res['confidences']):.2%}\",\r\n \"Accuracy\": f\"{res['accuracy']:.2%}\",\r\n \"Precision\": f\"{res['precision']:.2%}\",\r\n \"Recall\": f\"{res['recall']:.2%}\",\r\n \"F1 Score\": f\"{res['f1_score']:.2%}\",\r\n \"Analysis Time\": f\"{res['prediction_time']:.4f} sec\"\r\n })\r\n \r\n # Display results as a table\r\n df_results = pd.DataFrame(model_results)\r\n st.dataframe(df_results)\r\n \r\n # If detailed metrics are requested\r\n if display_metrics:\r\n st.write(\"#### Detailed Class Probabilities\")\r\n \r\n for model_name in selected_models:\r\n res = results[model_name][disorder]\r\n st.write(f\"**{model_name}**\")\r\n \r\n # Create a dataframe for class probabilities\r\n probs_df = pd.DataFrame({\r\n 'Class': res[\"class_names\"],\r\n 'Probability': res[\"confidences\"]\r\n })\r\n \r\n # Plot bar chart of probabilities\r\n fig, ax = plt.subplots(figsize=(8, 3))\r\n sns.barplot(x='Probability', y='Class', data=probs_df, ax=ax)\r\n ax.set_xlim(0, 1)\r\n ax.set_title(f\"{model_name} - {disorder} Class Probabilities\")\r\n st.pyplot(fig)\r\n \r\n # Show comparison charts if requested\r\n if show_comparison_charts and len(selected_models) > 1:\r\n st.write(\"#### Model Comparison\")\r\n \r\n # Extract metrics for comparison\r\n metrics_df = pd.DataFrame([\r\n {\r\n \"Model\": model_name,\r\n \"Accuracy\": results[model_name][disorder][\"accuracy\"],\r\n \"Precision\": results[model_name][disorder][\"precision\"],\r\n \"Recall\": results[model_name][disorder][\"recall\"],\r\n \"F1 Score\": results[model_name][disorder][\"f1_score\"],\r\n \"Analysis Time (sec)\": results[model_name][disorder][\"prediction_time\"]\r\n }\r\n for model_name in selected_models\r\n ])\r\n \r\n # Create metrics comparison chart\r\n fig, ax = plt.subplots(figsize=(10, 5))\r\n metrics_df_melt = pd.melt(\r\n metrics_df, \r\n id_vars=['Model'], \r\n value_vars=['Accuracy', 'Precision', 'Recall', 'F1 Score'],\r\n var_name='Metric', \r\n value_name='Value'\r\n )\r\n sns.barplot(x='Model', y='Value', hue='Metric', data=metrics_df_melt, ax=ax)\r\n ax.set_title(f\"Performance Metrics Comparison - {disorder}\")\r\n ax.set_ylim(0, 1)\r\n st.pyplot(fig)\r\n \r\n # Analysis time comparison\r\n fig, ax = plt.subplots(figsize=(8, 3))\r\n sns.barplot(x='Analysis Time (sec)', y='Model', data=metrics_df, ax=ax)\r\n ax.set_title(f\"Analysis Time Comparison - {disorder}\")\r\n st.pyplot(fig)\r\n\r\n # Overall comparison if multiple disorders are selected\r\n if len(selected_disorders) > 1:\r\n st.markdown(\"---\")\r\n st.markdown(\"

Cross-Disorder Comparison

\", unsafe_allow_html=True)\r\n \r\n for model_name in selected_models:\r\n st.write(f\"### {model_name} Performance Across Disorders\")\r\n \r\n # Create dataframe for cross-disorder comparison\r\n cross_disorder_df = pd.DataFrame([\r\n {\r\n \"Disorder\": disorder,\r\n \"Predicted Class\": results[model_name][disorder][\"predicted_class\"],\r\n \"Accuracy\": results[model_name][disorder][\"accuracy\"],\r\n \"Analysis Time (sec)\": results[model_name][disorder][\"prediction_time\"]\r\n }\r\n for disorder in selected_disorders\r\n ])\r\n \r\n st.dataframe(cross_disorder_df)\r\n \r\n # Plot accuracy comparison across disorders\r\n fig, ax = plt.subplots(figsize=(8, 4))\r\n sns.barplot(x='Disorder', y='Accuracy', data=cross_disorder_df, ax=ax)\r\n ax.set_title(f\"{model_name} - Accuracy Across Disorders\")\r\n ax.set_ylim(0, 1)\r\n st.pyplot(fig)\r\n\r\n elif uploaded_file is not None:\r\n st.warning(\"Please select at least one model and one disorder type to analyze.\")\r\n else:\r\n st.info(\"Please upload an MRI image to begin analysis.\")\r\n\r\n# Additional information about the application\r\nst.markdown(\"---\")\r\nst.markdown(\"

About This Application

\", unsafe_allow_html=True)\r\nst.write(\"\"\"\r\nThis application demonstrates the comparative analysis of different machine learning models\r\nfor brain disorder detection from MRI images. The current implementation uses simulated results,\r\nbut in a production environment, it would load pre-trained models for each disorder type.\r\n\r\n**Supported Model Architectures:**\r\n- AlexNet\r\n- ResNet-50\r\n- VGG-16\r\n- Inception V3\r\n\r\n**Supported Brain Disorders:**\r\n- Alzheimer's Disease (4 classes)\r\n- Brain Tumors (4 classes)\r\n- Parkinson's Disease (2 classes)\r\n\r\n**Metrics Evaluated:**\r\n- Accuracy\r\n- Precision\r\n- Recall\r\n- F1 Score\r\n- Analysis Time\r\n\"\"\")\r\n\r\n# Footer\r\nst.markdown(\"---\")\r\nst.markdown(\"\"\"\r\n

\r\n Developed by Aman Singh
\r\n Under the Supervision of Mr. Ashis Datta and Dr. Palash Ghosal\r\n

\r\n\"\"\", unsafe_allow_html=True)\r\n\r\n# Function to run the app\r\ndef run_app():\r\n st.empty()\r\n\r\nif __name__ == \"__main__\":\r\n run_app()", "requirement.txt": "tensorflow>=2.6.0\r\nnumpy>=1.19.5\r\npandas>=1.3.3\r\nmatplotlib>=3.4.3\r\nseaborn>=0.11.2\r\nstreamlit>=1.8.0\r\nPillow>=8.3.2\r\nscikit-learn>=1.0.0"}