app.py

import gradio as gr
import joblib
import os
import pandas as pd

# Path to the model scores file
model_scores_file = "report/model_summary_report_6_smote.csv"

# Load model performance metrics from the provided CSV file
if os.path.exists(model_scores_file):
    model_scores_df = pd.read_csv(model_scores_file)
    required_columns = {'Model Name', 'Model Sensitivity', 'Model Specificity'}
    if required_columns.issubset(model_scores_df.columns):
        model_performance = model_scores_df.set_index('Model Name')[['Model Sensitivity', 'Model Specificity']].T.to_dict()
    else:
        raise ValueError(f"The file '{model_scores_file}' must contain the columns: {required_columns}")
else:
    raise FileNotFoundError(f"The model scores file '{model_scores_file}' was not found. Please ensure it exists in the 'report/' directory.")

# Dictionary containing the model names and corresponding pickle file names
model_paths = {
    'AdaBoost': 'pjas-thyroid-AdaBoost.pkl',
    'Decision Tree': 'pjas-thyroid-Decision Tree.pkl',
    'Gaussian Naive Bayes': 'pjas-thyroid-Gaussian Naive Bayes.pkl',
    'Gradient Boosting': 'pjas-thyroid-Gradient Boosting.pkl',
    'K-Nearest Neighbors': 'pjas-thyroid-K-Nearest Neighbors.pkl',
    'Logistic Regression': 'pjas-thyroid-Logistic Regression.pkl',
    'Random Forest': 'pjas-thyroid-Random Forest.pkl',
    'Support Vector Machine': 'pjas-thyroid-Support Vector Machine.pkl',
    'XGBoost': 'pjas-thyroid-XGBoost.pkl'
}

# Preload all models at startup
loaded_models = {}
for model_name, pickle_file in model_paths.items():
    model_file_path = os.path.join("model", pickle_file)
    if os.path.exists(model_file_path):
        try:
            loaded_models[model_name] = joblib.load(model_file_path)
        except Exception as e:
            print(f"Error loading {model_name}: {e}")
    else:
        print(f"Model file for {model_name} not found.")

def predict_cancer(age, gender, T, N, Focality, Response):
    # Validate age
    if age is None or not (1 <= age <= 100):
        return "🔴 **Error:** Age must be a number between 1 and 100."
    
    # Validate gender
    if gender not in ["Female", "Male"]:
        return "🔴 **Error:** Please select a valid gender."

    # Validate T (Tumor Size)
    if T is None:
        return "🔴 **Error:** Please select a valid T (Tumor Size) option."

    # Validate N (Lymph Node Spread)
    if N is None:
        return "🔴 **Error:** Please select a valid N (Lymph Node Spread) option."

    # Validate Focality
    if Focality is None:
        return "🔴 **Error:** Please select a valid Focality option."

    # Validate Response
    if Response is None:
        return "🔴 **Error:** Please select a valid Response option."

    # Process gender and other fields
    gender_val = 0 if gender == "Female" else 1
    response_val = int(Response)
    T_val = int(T)
    N_val = int(N)
    Focality_val = int(Focality)

    # Prepare features
    features = pd.DataFrame({
        'Age': [age],
        'Gender': [gender_val],
        'T': [T_val],
        'N': [N_val],
        'Focality': [Focality_val],
        'Response': [response_val]
    })

    # Validate scaler file
    scaler_file = "model/pjas-thyroid-Scaler.pkl"
    if not os.path.exists(scaler_file):
        return "🔴 **Error:** Scaler file not found. Please contact the administrator."

    scaler = joblib.load(scaler_file)
    features[['Age']] = scaler.transform(features[['Age']])

    # Sort models based on sensitivity
    sorted_model_names = sorted(
        model_performance.keys(),
        key=lambda m: model_performance[m]['Model Sensitivity'],
        reverse=True
    )

    # Generate HTML table
    table_header = """
    <table>
        <thead>
            <tr>
                <th>Model</th>
                <th style="color:#FBCEB1;">Recurrence Accuracy (%)</th>
                <th style="color:green;">Non-Recurrence Accuracy (%)</th>
                <th>Prediction</th>
            </tr>
        </thead>
        <tbody>
    """
    table_rows = []
    can_recur_emoji = "🔴"
    cannot_recur_emoji = "🟢"

    for model_name in sorted_model_names:
        model = loaded_models.get(model_name)
        if not model:
            row = f"<tr><td>{model_name}</td><td>N/A</td><td>N/A</td><td>Error: Model not loaded</td></tr>"
            table_rows.append(row)
            continue

        try:
            prediction = model.predict(features)
            pred_value = prediction[0]
            pred_text = f"{can_recur_emoji} Can recur" if pred_value == 1 else f"{cannot_recur_emoji} Cannot-recur"

            sensitivity = model_performance[model_name]['Model Sensitivity']
            specificity = model_performance[model_name]['Model Specificity']

            row = f"<tr><td>{model_name}</td><td>{sensitivity:.2f}%</td><td>{specificity:.2f}%</td><td>{pred_text}</td></tr>"
            table_rows.append(row)
        except Exception as e:
            row = f"<tr><td>{model_name}</td><td>N/A</td><td>N/A</td><td>Error: {str(e)}</td></tr>"
            table_rows.append(row)

    table_footer = "</tbody></table>"
    html_table = table_header + "".join(table_rows) + table_footer
    success_message = "<br><br>✅ <strong>Prediction completed successfully.</strong>"
    return html_table + success_message

def clear_md():
    return ""

# UI Layout
with gr.Blocks(theme=gr.themes.Ocean()) as demo:
    gr.Markdown("# Thyroid Cancer Recurrence Predictor")

    with gr.Row():
        age_slider = gr.Number(
            label="Age",
            value=44,
            interactive=True,
            elem_id="age-box",
            step=1
        )
        gender_radio = gr.Radio(
            choices=["Female", "Male"],
            value="Female",
            label="Gender",
            interactive=True
        )

    with gr.Row():
        T_dropdown = gr.Dropdown(
            choices=[
                ("T1a (≤1 cm, confined to the thyroid)", "0"),
                ("T1b (>1 cm and ≤2 cm, confined to the thyroid)", "1"),
                ("T2 (>2 cm and ≤4 cm, confined to the thyroid)", "2"),
                ("T3a (>4 cm, confined to the thyroid)", "3"),
                ("T3b (Minimal extrathyroidal extension)", "4"),
                ("T4a (Moderate extrathyroidal extension, operable)", "5"),
                ("T4b (Extensive extrathyroidal extension, inoperable)", "6")
            ],
            value="0",
            label="T (Tumor Size)",
            interactive=True
        )

    with gr.Row():
        N_dropdown = gr.Dropdown(
            choices=[
                ("N0 (No spread to nearby lymph nodes)", "0"),
                ("N1a (Spread to lymph nodes in the neck close to the thyroid)", "1"),
                ("N1b (Spread to lymph nodes in the neck farther from the thyroid or upper chest)", "2")
            ],
            value="0",
            label="N (Lymph Node Spread)",
            interactive=True
        )

    with gr.Row():
        focality_dropdown = gr.Dropdown(
            choices=[
                ("Uni-focal (Single focus of thyroid cancer)", "1"),
                ("Multi-focal (Multiple foci of thyroid cancer)", "0")
            ],
            value="1",
            label="Focality",
            interactive=True
        )

    with gr.Row():
        response_dropdown = gr.Dropdown(
            choices=[
                ("✅ Excellent Response - Negative imaging studies and Tg < 0.2 ng/mL or stimulated Tg < 1 ng/mL", "0"),
                ("❓ Indeterminate Response - Nonspecific findings; Tg potentially low", "1"),
                ("⚠️ Biochemical Incomplete - Tg > 1 ng/mL or rising anti-Tg antibody levels", "2"),
                ("❌ Structural Incomplete - Identifiable structural disease on imaging", "3")
            ],
            value="0",
            label="Response",
            interactive=True
        )

    predict_button = gr.Button(value="Predict", variant="primary")
    prediction_output = gr.HTML(label="Prediction Results")

    predict_button.click(fn=clear_md, outputs=prediction_output)
    predict_button.click(
        fn=predict_cancer,
        inputs=[age_slider, gender_radio, T_dropdown, N_dropdown, focality_dropdown, response_dropdown],
        outputs=prediction_output
    )

demo.launch()