Update app.py

app.py

@@ -0,0 +1,224 @@
+import gradio as gr
+import joblib
+import os
+import pandas as pd
+
+# 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'
+}
+
+# Example: Hard-coded dictionary of accuracies per model.
+model_accuracies = {
+    'AdaBoost': 97.13,
+    'Logistic Regression': 96.87,
+    'Gaussian Naive Bayes': 95.3,
+    'Gradient Boosting': 98.96,
+    'Support Vector Machine': 96.34,
+    'Decision Tree': 97.91,
+    'K-Nearest Neighbors': 97.13,
+    'Random Forest': 98.96,
+    'XGBoost': 98.43
+}
+
+# We assume:
+#   0 -> Cancer cannot recur
+#   1 -> Cancer can recur
+
+response_details = {
+    "0": """**Excellent Response**  
+Negative imaging studies and suppressed thyroglobulin levels  
+(below 0.2 ng/mL or stimulated Tg below 1 ng/mL).""",
+    "1": """**Indeterminate Response**  
+Nonspecific findings on imaging studies, making it difficult  
+to confidently classify as benign or malignant,  
+with potentially low thyroglobulin levels.""",
+    "2": """**Biochemical Incomplete**  
+Negative imaging but elevated thyroglobulin levels  
+(suppressed Tg above 1 ng/mL or stimulated Tg above 10 ng/mL)  
+or rising anti-Tg antibody levels.""",
+    "3": """**Structural Incomplete**  
+Presence of identifiable structural disease on imaging,  
+regardless of thyroglobulin level."""
+}
+
+def predict_cancer(age, gender, response, tumor_size, lymph_node_spread, focality):
+    """
+    Generates a Markdown table with emoji icons for each model's prediction,
+    including 'TumorSize', 'LymphNodeSpread', and 'Focality' as additional features.
+    """
+    # 1. Load your pre-fitted scaler from disk
+    scaler_file = "model/pjas-thyroid-scaler.pkl"
+    if not os.path.exists(scaler_file):
+        return "Error: Scaler file not found. Please check your path or name."
+
+    scaler = joblib.load(scaler_file)
+
+    # 2. Simple encodings for demonstration
+    gender_val = 0 if gender == "Female" else 1
+    response_val = int(response)
+    tumor_val = int(tumor_size)          # Convert Tumor Size dropdown value to int
+    lymph_val = int(lymph_node_spread)   # Convert Lymph Node Spread dropdown value to int
+    focality_val = int(focality)         # Convert Focality dropdown value to int
+
+    # 3. Create a DataFrame for the features
+    features = pd.DataFrame({
+        'Age': [age],
+        'Gender': [gender_val],
+        'T': [tumor_val],
+        'N': [lymph_val],
+        'Focality': [focality_val],
+        'Response': [response_val]
+    })
+
+    # 4. Remove any NaN
+    features = features.dropna()
+
+    # 5. Scale the 'Age' column
+    #    If you have more features to scale, include them here accordingly
+    features[['Age']] = scaler.transform(features[['Age']])
+
+    # 6. Sort models by accuracy (descending)
+    sorted_model_names = sorted(
+        model_paths.keys(),
+        key=lambda m: model_accuracies[m],
+        reverse=True
+    )
+
+    # 7. Build a Markdown table
+    table_header = (
+        "| **Model**                   | **Accuracy** | **Prediction**                |\n"
+        "|-----------------------------|--------------|--------------------------------|\n"
+    )
+    table_rows = []
+
+    # 8. Emojis for predictions
+    can_recur_emoji = "🔴"  # "Cancer can recur"
+    cannot_recur_emoji = "🟢"  # "Cancer cannot-recur"
+
+    # 9. Iterate through each model and make predictions
+    for model_name in sorted_model_names:
+        pickle_file = model_paths[model_name]
+        model_file_path = os.path.join("model", pickle_file) 
+
+        if not os.path.exists(model_file_path):
+            row = f"| {model_name} | N/A | **Error**: file not found |"
+            table_rows.append(row)
+            continue
+
+        model = joblib.load(model_file_path)
+        prediction = model.predict(features)  # e.g., [0] or [1]
+        pred_value = prediction[0]
+
+        # 10. Convert numeric prediction to icon + text
+        if pred_value == 1:
+            pred_text = f"{can_recur_emoji} Sorry, Your Cancer can recur"
+        else:
+            pred_text = f"{cannot_recur_emoji} Great News! Your Cancer cannot-recur"
+
+        accuracy = model_accuracies.get(model_name, "N/A")
+        row = f"| {model_name} | {accuracy}% | {pred_text} |"
+        table_rows.append(row)
+
+    # 11. Combine into a single Markdown table
+    md_table = table_header + "\n".join(table_rows)
+    return md_table
+
+def clear_md():
+    """Clears the Markdown output."""
+    return ""
+
+with gr.Blocks() as demo:
+    gr.Markdown("# Thyroid Cancer Recurrence Predictor")
+
+    # Existing inputs
+    age_slider = gr.Slider(
+        minimum=1,
+        maximum=100,
+        step=1,
+        label="Age",
+        value=44,
+        interactive=True
+    )
+    
+    gender_radio = gr.Radio(
+        choices=["Female", "Male"],
+        value="Female",
+        label="Gender",
+        interactive=True
+    )
+
+    # New Tumor Size dropdown with descriptive text
+    tumor_size_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",  # Default T1a
+        label="Tumor Size",
+        interactive=True
+    )
+
+    # New Lymph Node Spread dropdown with descriptive text
+    lymph_node_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",  # Default N0
+        label="Lymph Node Spread",
+        interactive=True
+    )
+
+    # New Focality of Differential Thyroid Cancer dropdown with descriptive text
+    focality_dropdown = gr.Dropdown(
+        choices=[
+            ("Uni-focal (Single focus of thyroid cancer)", "1"),
+            ("Multi-focal (Multiple foci of thyroid cancer)", "0")
+        ],
+        value="1",  # Default Uni-focal
+        label="Focality of Differential Thyroid Cancer",
+        interactive=True
+    )
+
+    response_dropdown = gr.Dropdown(
+        choices=[
+            ("✅ Excellent Response - Negative imaging studies and suppressed thyroglobulin levels (below 0.2 ng/mL or stimulated Tg below 1 ng/mL)", "0"),
+            ("❓ Indeterminate Response - Nonspecific findings on imaging studies with potentially low thyroglobulin levels", "1"),
+            ("⚠️ Biochemical Incomplete - Negative imaging but elevated thyroglobulin levels or rising anti-Tg antibody levels", "2"),
+            ("❌ Structural Incomplete - Presence of identifiable structural disease on imaging, regardless of thyroglobulin level", "3")
+        ],
+        value="0",  # Default
+        label="Response",
+        interactive=True
+    )
+
+    # Predict button
+    predict_button = gr.Button(
+        value="Predict",
+        variant="primary"
+    )
+    prediction_output = gr.Markdown(label="Prediction Results")
+
+    # Include all dropdowns as inputs for the predict function
+    predict_button.click(
+        fn=predict_cancer,
+        inputs=[age_slider, gender_radio, response_dropdown, tumor_size_dropdown, lymph_node_dropdown, focality_dropdown],
+        outputs=prediction_output
+    )
+
+demo.launch()