app.py

import gradio as gr
import pandas as pd
import numpy as np
import joblib
from skimage.measure import shannon_entropy
from skimage.color import rgb2hsv
from scipy.ndimage import generic_filter
import cv2
from PIL import Image
from sklearn.preprocessing import LabelEncoder

# Initialize LabelEncoder for gender encoding
gender_encoder = LabelEncoder()
gender_encoder.fit(['Female', 'Male'])  # Ensure that the labels are ordered

# Function to extract features from the image
def extract_features(image):
    # Convert PIL image to NumPy array
    image = np.array(image) 

    # Extract RGB means
    meanr = np.mean(image[:, :, 0])  # Red channel
    meang = np.mean(image[:, :, 1])  # Green channel
    meanb = np.mean(image[:, :, 2])  # Blue channel

    # Convert to HSI and compute HHR
    hsv_image = rgb2hsv(image)
    hue = hsv_image[:, :, 0]
    high_hue_pixels = np.sum(hue > 0.95)
    total_pixels = hue.size
    HHR = high_hue_pixels / total_pixels

    # Convert to Grayscale
    gray_image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)

    # Compute Entropy
    Ent = shannon_entropy(gray_image)

    # Compute Brightness
    B = np.mean(gray_image)

    # Sliding window for gray-level features
    def g1_filter(window):
        return window[4] - np.min(window)

    def g2_filter(window):
        return np.max(window) - window[4]

    def g3_filter(window):
        return window[4] - np.mean(window)

    def g4_filter(window):
        return np.std(window)

    def g5_filter(window):
        return window[4]

    # Apply filters with 3x3 window
    g1 = generic_filter(gray_image, g1_filter, size=3).mean()
    g2 = generic_filter(gray_image, g2_filter, size=3).mean()
    g3 = generic_filter(gray_image, g3_filter, size=3).mean()
    g4 = generic_filter(gray_image, g4_filter, size=3).mean()
    g5 = generic_filter(gray_image, g5_filter, size=3).mean()

    # Return features
    return {
        "meanr": meanr,
        "meang": meang,
        "meanb": meanb,
        "HHR": HHR,
        "Ent": Ent,
        "B": B,
        "g1": g1,
        "g2": g2,
        "g3": g3,
        "g4": g4,
        "g5": g5,
    }

# Function to check if the image is a valid file format
def check_image_format(filepath):
    try:
        # Try opening the image using PIL
        with Image.open(filepath) as img:
            img.verify()  # Verify if it's a valid image file
        return True
    except Exception as e:
        print(f"Error opening image: {e}")
        return False

# Function to predict hemoglobin value with label encoding for gender
def predict_hemoglobin(age, gender, image):
    try:
        # Ensure the image is not None
        if image is None:
            return "Error: No image uploaded. Please upload an image."

        # Check if the image is valid
        if not isinstance(image, Image.Image):
            return "Error: Invalid image format. Please upload a valid image file."

        # Extract features from the image
        features = extract_features(image)

        # Use LabelEncoder to convert gender to numerical value
        features['Gender'] = gender_encoder.transform([gender])[0]
        features['Age'] = age

        # Create a DataFrame for features (do not include Hgb, as it's the predicted value)
        features_df = pd.DataFrame([features])

        # Load the trained model, scaler, and label encoder
        svr_model = joblib.load('svr_model (1).pkl')  # Replace with the actual path to your model file
        scaler = joblib.load('minmax_scaler.pkl')     # Replace with the actual path to your scaler file

        # Ensure that features_df matches the expected training feature set (without 'Hgb')
        expected_columns = ['meanr', 'meang', 'meanb', 'HHR', 'Ent', 'B', 'g1', 'g2', 'g3', 'g4', 'g5', 'Age', 'Gender']

        for col in expected_columns:
            if col not in features_df:
                features_df[col] = 0  # Or some default value to match the expected columns.

        features_df = features_df[expected_columns]  # Ensure the correct order of columns

        # Apply scaling (do not include 'Hgb' as it is the target)
        features_df_scaled = scaler.transform(features_df)

        # Predict hemoglobin using the trained SVR model
        hemoglobin = svr_model.predict(features_df_scaled)[0]

        return f"Predicted Hemoglobin Value: {hemoglobin:.2f}"

    except Exception as e:
        print(f"An error occurred during prediction: {e}")
        return "An error occurred during prediction. Please try again."

# Gradio interface
with gr.Blocks() as anemia_detection_app:
    gr.Markdown("# Hemoglobin Prediction App")
    
    with gr.Row():
        age_input = gr.Number(label="Age", value=25)
        gender_input = gr.Radio(label="Gender", choices=["Male", "Female"], value="Male")
    
    image_input = gr.Image(label="Upload Retinal Image", type="pil")
    output_text = gr.Textbox(label="Predicted Hemoglobin Value")
    
    predict_button = gr.Button("Predict")
    
    predict_button.click(
        fn=predict_hemoglobin,
        inputs=[age_input, gender_input, image_input],
        outputs=output_text
    )

# Run the app
if __name__ == "__main__":
    anemia_detection_app.launch()