app.py

import os
from dotenv import load_dotenv
import gradio as gr
from PIL import Image
from transformers import BlipProcessor, BlipForConditionalGeneration
import openai

# Load environment variables from .env file
load_dotenv()

# Retrieve OpenAI credentials from environment
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
OPENAI_API_BASE = os.getenv("OPENAI_API_BASE", "https://models.inference.ai.azure.com")  # fallback

# Set OpenAI credentials
openai.api_key = OPENAI_API_KEY
openai.base_url = OPENAI_API_BASE

# Load the BLIP model and processor
processor = BlipProcessor.from_pretrained("nathansutton/generate-cxr")
model = BlipForConditionalGeneration.from_pretrained("nathansutton/generate-cxr")

def generate_report(image):
    """Generate a CXR report from the uploaded image."""
    inputs = processor(images=image, text="a chest x-ray", return_tensors="pt")
    output = model.generate(**inputs, max_length=512)
    report = processor.decode(output[0], skip_special_tokens=True)
    return report

def chat_with_openai(user_message, previous_report):
    """Chat with GPT-4o based on the generated report."""
    conversation = [
        {"role": "system", "content": "You are a helpful medical assistant."},
        {"role": "user", "content": f"Here is a medical report: {previous_report}. Now, {user_message}"}
    ]

    response = openai.ChatCompletion.create(
        model="gpt-4o",
        messages=conversation,
        temperature=1.0,
        top_p=1.0,
        max_tokens=1000,
    )

    return response['choices'][0]['message']['content']

def process_image_and_chat(image, user_message, chat_history):
    """Handle full process: generate report and chat."""
    if chat_history is None:
        chat_history = []

    # Step 1: Generate CXR report
    report = generate_report(image)
    chat_history.append({"role": "assistant", "content": report})

    # Step 2: Chat based on the report
    response = chat_with_openai(user_message, report)
    chat_history.append({"role": "user", "content": user_message})
    chat_history.append({"role": "assistant", "content": response})

    return chat_history, chat_history

# Gradio Interface
iface = gr.Interface(
    fn=process_image_and_chat,
    inputs=[
        gr.Image(type="pil", label="Upload Chest X-ray Image"),
        gr.Textbox(label="Your Question", placeholder="Ask a question about the report..."),
        gr.State(value=[]),  # Memory for chat history
    ],
    outputs=[
        gr.Chatbot(label="Medical Assistant Chat", type="messages"),
        gr.State(),  # Return updated history
    ],
    title="Chest X-ray Assistant",
    description="Upload a chest X-ray image and ask questions about it. The assistant will generate a radiology report and answer your questions using GPT-4o.",
)

if __name__ == "__main__":
    iface.launch()