app.py

import gradio as gr
import pandas as pd
from datasets import load_dataset
from openai import OpenAI
from PIL import Image
import io
import base64
import logging

# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# App version
APP_VERSION = "1.0.0"
logger.info(f"Starting Radiology Teaching App v{APP_VERSION}")

try:
    # Load only 10 rows from the dataset
    logger.info("Loading MIMIC-CXR dataset...")
    dataset = load_dataset("itsanmolgupta/mimic-cxr-dataset", split="train").select(range(10))
    df = pd.DataFrame(dataset)
    logger.info(f"Successfully loaded {len(df)} cases")
except Exception as e:
    logger.error(f"Error loading dataset: {str(e)}")
    raise

def encode_image_to_base64(image_bytes):
    return base64.b64encode(image_bytes).decode('utf-8')

def analyze_report(user_findings, ground_truth_findings, ground_truth_impression, api_key):
    if not api_key:
        return "Please provide a DeepSeek API key to analyze the report."
    
    try:
        client = OpenAI(api_key=api_key, base_url="https://api.deepseek.com")
        logger.info("Analyzing report with DeepSeek...")
        
        prompt = f"""You are an expert radiologist reviewing a trainee's chest X-ray report. 
        
        Trainee's Findings:
        {user_findings}
        
        Ground Truth Findings:
        {ground_truth_findings}
        
        Ground Truth Impression:
        {ground_truth_impression}
        
        Please provide:
        1. Number of important findings missed by the trainee (list them)
        2. Quality assessment of the trainee's report (structure, completeness, accuracy)
        3. Constructive feedback for improvement
        
        Format your response in clear sections."""

        response = client.chat.completions.create(
            model="deepseek-chat",
            messages=[
                {"role": "system", "content": "You are an expert radiologist providing constructive feedback."},
                {"role": "user", "content": prompt}
            ],
            stream=False
        )
        
        return response.choices[0].message.content
    except Exception as e:
        logger.error(f"Error in report analysis: {str(e)}")
        return f"Error analyzing report: {str(e)}"

def load_random_case(hide_ground_truth):
    try:
        # Randomly select a case from our dataset
        random_case = df.sample(n=1).iloc[0]
        logger.info("Loading random case...")
        
        # Get the image, findings, and impression
        image = random_case['image']
        findings = "" if hide_ground_truth else random_case['findings']
        impression = "" if hide_ground_truth else random_case['impression']
        
        return image, findings, impression
    except Exception as e:
        logger.error(f"Error loading random case: {str(e)}")
        return None, "Error loading case", "Error loading case"

def process_case(image, user_findings, hide_ground_truth, api_key, current_findings="", current_impression=""):
    if hide_ground_truth:
        return "", "", ""
    else:
        analysis = analyze_report(user_findings, current_findings, current_impression, api_key)
        return current_findings, current_impression, analysis

# Create the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown(f"# Radiology Report Training System v{APP_VERSION}")
    gr.Markdown("### Practice your chest X-ray reading and reporting skills")
    
    with gr.Row():
        with gr.Column():
            image_display = gr.Image(label="Chest X-ray Image", type="pil")
            api_key_input = gr.Textbox(label="DeepSeek API Key", type="password")
            hide_truth = gr.Checkbox(label="Hide Ground Truth", value=False)
            load_btn = gr.Button("Load Random Case")
        
        with gr.Column():
            user_findings_input = gr.Textbox(label="Your Findings", lines=10, placeholder="Type or dictate your findings here...")
            ground_truth_findings = gr.Textbox(label="Ground Truth Findings", lines=5, interactive=False)
            ground_truth_impression = gr.Textbox(label="Ground Truth Impression", lines=5, interactive=False)
            analysis_output = gr.Textbox(label="Analysis and Feedback", lines=10, interactive=False)
            submit_btn = gr.Button("Submit Report")

    # Event handlers
    load_btn.click(
        fn=load_random_case,
        inputs=[hide_truth],
        outputs=[image_display, ground_truth_findings, ground_truth_impression]
    )
    
    submit_btn.click(
        fn=process_case,
        inputs=[
            image_display,
            user_findings_input,
            hide_truth,
            api_key_input,
            ground_truth_findings,
            ground_truth_impression
        ],
        outputs=[
            ground_truth_findings,
            ground_truth_impression,
            analysis_output
        ]
    )

    hide_truth.change(
        fn=lambda x: ("", "", "") if x else (ground_truth_findings.value, ground_truth_impression.value, ""),
        inputs=[hide_truth],
        outputs=[ground_truth_findings, ground_truth_impression, analysis_output]
    )

logger.info("Starting Gradio interface...")
demo.launch()