app.py

import streamlit as st
import tensorflow as tf
import numpy as np
from PIL import Image
from huggingface_hub import hf_hub_download
import os

# Suppress warnings
import warnings
warnings.filterwarnings("ignore")

# Set page configuration
st.set_page_config(
    page_title="ChestAI - Pneumonia Detection",
    page_icon="🫁",
    initial_sidebar_state="auto",
)

# Hide Streamlit style
hide_streamlit_style = """
<style>
#MainMenu {visibility: hidden;}
footer {visibility: hidden;}
</style>
"""
st.markdown(hide_streamlit_style, unsafe_allow_html=True)

# Function to load the model
@st.cache_resource(show_spinner=False)
def load_model():
    try:
        # Download the model directory
        model_dir = hf_hub_download(repo_id="ryefoxlime/PneumoniaDetection", repo_type="model", library="tf", cache_dir="/home/user/.cache/huggingface/hub")

        # Load the model using tf.saved_model.load
        model = tf.saved_model.load(model_dir)
        return model
    except Exception as e:
        st.error(f"Error loading model: {str(e)}")
        return None

# Load the model
with st.spinner("Model is being loaded..."):
    model = load_model()

if model is None:
    st.error("Failed to load model. Please try again.")
    st.stop()

# Sidebar for app information
with st.sidebar:
    st.title("ChestAI")
    st.markdown("""
        ### About
        ChestAI uses advanced deep learning to detect pneumonia in chest X-rays.
                
        ### How to use
        1. Upload a chest X-ray image (JPG/PNG)
        2. Wait for the analysis
        3. View the results and confidence score
                
        ### Note
        This tool is for educational purposes only. Always consult healthcare professionals for medical advice.
    """)

st.set_option("deprecation.showfileUploaderEncoding", False)

# File uploader for image input
file = st.file_uploader("Upload a chest X-ray image", type=["jpg", "png"])

def import_and_predict(image_data, model):
    img_array = tf.keras.preprocessing.image.img_to_array(image_data)
    img_array = np.expand_dims(img_array, axis=0)  # Add batch dimension
    img_array = img_array / 255.0  # Normalize the image

    # Perform prediction
    predictions = model(img_array)  # Call the model for prediction
    return predictions

# Class names for prediction results
class_names = ["Normal", "PNEUMONIA"]

if file is None:
    st.text("Please upload an image file")
else:
    try:
        image = tf.keras.preprocessing.image.load_img(file, target_size=(224, 224), color_mode='rgb')
        st.image(image, caption="Uploaded Image.", use_column_width=True)
        
        predictions = import_and_predict(image, model)
        predicted_class = np.argmax(predictions)  # Get the index of the highest prediction
        confidence = float(predictions[0][predicted_class] * 100)  # Confidence percentage

        # Display the results
        st.info(f"Confidence: {confidence:.2f}%")
        
        if class_names[predicted_class] == "Normal":
            st.balloons()
            st.success(f"Result: {class_names[predicted_class]}")
        else:
            st.warning(f"Result: {class_names[predicted_class]}")

    except Exception as e:
        st.error(f"Error processing image: {str(e)}")