Amarthya7's picture
Upload 21 files
86a74e6 verified
import logging
import os
import sys
import tempfile
from pathlib import Path
import gradio as gr
import matplotlib.pyplot as plt
from PIL import Image
# Add parent directory to path
parent_dir = os.path.dirname(os.path.abspath(__file__))
sys.path.append(parent_dir)
# Import our modules
from models.multimodal_fusion import MultimodalFusion
from utils.preprocessing import enhance_xray_image, normalize_report_text
from utils.visualization import (
plot_image_prediction,
plot_multimodal_results,
plot_report_entities,
)
# Set up logging
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
handlers=[logging.StreamHandler(), logging.FileHandler("mediSync.log")],
)
logger = logging.getLogger(__name__)
# Create temporary directory for sample data if it doesn't exist
os.makedirs(os.path.join(parent_dir, "data", "sample"), exist_ok=True)
class MediSyncApp:
"""
Main application class for the MediSync multi-modal medical analysis system.
"""
def __init__(self):
"""Initialize the application and load models."""
self.logger = logging.getLogger(__name__)
self.logger.info("Initializing MediSync application")
# Initialize models with None for lazy loading
self.fusion_model = None
self.image_model = None
self.text_model = None
def load_models(self):
"""
Load models if not already loaded.
Returns:
bool: True if models loaded successfully, False otherwise
"""
try:
if self.fusion_model is None:
self.logger.info("Loading models...")
self.fusion_model = MultimodalFusion()
self.image_model = self.fusion_model.image_analyzer
self.text_model = self.fusion_model.text_analyzer
self.logger.info("Models loaded successfully")
return True
except Exception as e:
self.logger.error(f"Error loading models: {e}")
return False
def analyze_image(self, image):
"""
Analyze a medical image.
Args:
image: Image file uploaded through Gradio
Returns:
tuple: (image, image_results_html, plot_as_html)
"""
try:
# Ensure models are loaded
if not self.load_models() or self.image_model is None:
return image, "Error: Models not loaded properly.", None
# Save uploaded image to a temporary file
temp_dir = tempfile.mkdtemp()
temp_path = os.path.join(temp_dir, "upload.png")
if isinstance(image, str):
# Copy the file if it's a path
from shutil import copyfile
copyfile(image, temp_path)
else:
# Save if it's a Gradio UploadButton image
image.save(temp_path)
# Run image analysis
self.logger.info(f"Analyzing image: {temp_path}")
results = self.image_model.analyze(temp_path)
# Create visualization
fig = plot_image_prediction(
image,
results.get("predictions", []),
f"Primary Finding: {results.get('primary_finding', 'Unknown')}",
)
# Convert to HTML for display
plot_html = self.fig_to_html(fig)
# Format results as HTML
html_result = f"""
<h2>X-ray Analysis Results</h2>
<p><strong>Primary Finding:</strong> {results.get("primary_finding", "Unknown")}</p>
<p><strong>Confidence:</strong> {results.get("confidence", 0):.1%}</p>
<p><strong>Abnormality Detected:</strong> {"Yes" if results.get("has_abnormality", False) else "No"}</p>
<h3>Top Predictions:</h3>
<ul>
"""
# Add top 5 predictions
for label, prob in results.get("predictions", [])[:5]:
html_result += f"<li>{label}: {prob:.1%}</li>"
html_result += "</ul>"
# Add explanation
explanation = self.image_model.get_explanation(results)
html_result += f"<h3>Analysis Explanation:</h3><p>{explanation}</p>"
return image, html_result, plot_html
except Exception as e:
self.logger.error(f"Error in image analysis: {e}")
return image, f"Error analyzing image: {str(e)}", None
def analyze_text(self, text):
"""
Analyze a medical report text.
Args:
text: Report text input through Gradio
Returns:
tuple: (text, text_results_html, entities_plot_html)
"""
try:
# Ensure models are loaded
if not self.load_models() or self.text_model is None:
return text, "Error: Models not loaded properly.", None
# Check for empty text
if not text or len(text.strip()) < 10:
return (
text,
"Error: Please enter a valid medical report text (at least 10 characters).",
None,
)
# Normalize text
normalized_text = normalize_report_text(text)
# Run text analysis
self.logger.info("Analyzing medical report text")
results = self.text_model.analyze(normalized_text)
# Get entities and create visualization
entities = results.get("entities", {})
fig = plot_report_entities(normalized_text, entities)
# Convert to HTML for display
entities_plot_html = self.fig_to_html(fig)
# Format results as HTML
html_result = f"""
<h2>Medical Report Analysis Results</h2>
<p><strong>Severity Level:</strong> {results.get("severity", {}).get("level", "Unknown")}</p>
<p><strong>Severity Score:</strong> {results.get("severity", {}).get("score", 0)}/4</p>
<p><strong>Confidence:</strong> {results.get("severity", {}).get("confidence", 0):.1%}</p>
<h3>Key Findings:</h3>
<ul>
"""
# Add findings
findings = results.get("findings", [])
if findings:
for finding in findings:
html_result += f"<li>{finding}</li>"
else:
html_result += "<li>No specific findings detailed.</li>"
html_result += "</ul>"
# Add entities
html_result += "<h3>Extracted Medical Entities:</h3>"
for category, items in entities.items():
if items:
html_result += f"<p><strong>{category.capitalize()}:</strong> {', '.join(items)}</p>"
# Add follow-up recommendations
html_result += "<h3>Follow-up Recommendations:</h3><ul>"
followups = results.get("followup_recommendations", [])
if followups:
for rec in followups:
html_result += f"<li>{rec}</li>"
else:
html_result += "<li>No specific follow-up recommendations.</li>"
html_result += "</ul>"
return text, html_result, entities_plot_html
except Exception as e:
self.logger.error(f"Error in text analysis: {e}")
return text, f"Error analyzing text: {str(e)}", None
def analyze_multimodal(self, image, text):
"""
Perform multimodal analysis of image and text.
Args:
image: Image file uploaded through Gradio
text: Report text input through Gradio
Returns:
tuple: (results_html, multimodal_plot_html)
"""
try:
# Ensure models are loaded
if not self.load_models() or self.fusion_model is None:
return "Error: Models not loaded properly.", None
# Check for empty inputs
if image is None:
return "Error: Please upload an X-ray image for analysis.", None
if not text or len(text.strip()) < 10:
return (
"Error: Please enter a valid medical report text (at least 10 characters).",
None,
)
# Save uploaded image to a temporary file
temp_dir = tempfile.mkdtemp()
temp_path = os.path.join(temp_dir, "upload.png")
if isinstance(image, str):
# Copy the file if it's a path
from shutil import copyfile
copyfile(image, temp_path)
else:
# Save if it's a Gradio UploadButton image
image.save(temp_path)
# Normalize text
normalized_text = normalize_report_text(text)
# Run multimodal analysis
self.logger.info("Performing multimodal analysis")
results = self.fusion_model.analyze(temp_path, normalized_text)
# Create visualization
fig = plot_multimodal_results(results, image, text)
# Convert to HTML for display
plot_html = self.fig_to_html(fig)
# Generate explanation
explanation = self.fusion_model.get_explanation(results)
# Format results as HTML
html_result = f"""
<h2>Multimodal Medical Analysis Results</h2>
<h3>Overview</h3>
<p><strong>Primary Finding:</strong> {results.get("primary_finding", "Unknown")}</p>
<p><strong>Severity Level:</strong> {results.get("severity", {}).get("level", "Unknown")}</p>
<p><strong>Severity Score:</strong> {results.get("severity", {}).get("score", 0)}/4</p>
<p><strong>Agreement Score:</strong> {results.get("agreement_score", 0):.0%}</p>
<h3>Detailed Findings</h3>
<ul>
"""
# Add findings
findings = results.get("findings", [])
if findings:
for finding in findings:
html_result += f"<li>{finding}</li>"
else:
html_result += "<li>No specific findings detailed.</li>"
html_result += "</ul>"
# Add follow-up recommendations
html_result += "<h3>Recommended Follow-up</h3><ul>"
followups = results.get("followup_recommendations", [])
if followups:
for rec in followups:
html_result += f"<li>{rec}</li>"
else:
html_result += (
"<li>No specific follow-up recommendations provided.</li>"
)
html_result += "</ul>"
# Add confidence note
confidence = results.get("severity", {}).get("confidence", 0)
html_result += f"""
<p><em>Note: This analysis has a confidence level of {confidence:.0%}.
Please consult with healthcare professionals for official diagnosis.</em></p>
"""
return html_result, plot_html
except Exception as e:
self.logger.error(f"Error in multimodal analysis: {e}")
return f"Error in multimodal analysis: {str(e)}", None
def enhance_image(self, image):
"""
Enhance X-ray image contrast.
Args:
image: Image file uploaded through Gradio
Returns:
PIL.Image: Enhanced image
"""
try:
if image is None:
return None
# Save uploaded image to a temporary file
temp_dir = tempfile.mkdtemp()
temp_path = os.path.join(temp_dir, "upload.png")
if isinstance(image, str):
# Copy the file if it's a path
from shutil import copyfile
copyfile(image, temp_path)
else:
# Save if it's a Gradio UploadButton image
image.save(temp_path)
# Enhance image
self.logger.info(f"Enhancing image: {temp_path}")
output_path = os.path.join(temp_dir, "enhanced.png")
enhance_xray_image(temp_path, output_path)
# Load enhanced image
enhanced = Image.open(output_path)
return enhanced
except Exception as e:
self.logger.error(f"Error enhancing image: {e}")
return image # Return original image on error
def fig_to_html(self, fig):
"""Convert matplotlib figure to HTML for display in Gradio."""
try:
import base64
import io
buf = io.BytesIO()
fig.savefig(buf, format="png", bbox_inches="tight")
buf.seek(0)
img_str = base64.b64encode(buf.read()).decode("utf-8")
plt.close(fig)
return f'<img src="data:image/png;base64,{img_str}" alt="Analysis Plot">'
except Exception as e:
self.logger.error(f"Error converting figure to HTML: {e}")
return "<p>Error displaying visualization.</p>"
def create_interface():
"""Create and launch the Gradio interface."""
app = MediSyncApp()
# Example medical report for demo
example_report = """
CHEST X-RAY EXAMINATION
CLINICAL HISTORY: 55-year-old male with cough and fever.
FINDINGS: The heart size is at the upper limits of normal. The lungs are clear without focal consolidation,
effusion, or pneumothorax. There is mild prominence of the pulmonary vasculature. No pleural effusion is seen.
There is a small nodular opacity noted in the right lower lobe measuring approximately 8mm, which is suspicious
and warrants further investigation. The mediastinum is unremarkable. The visualized bony structures show no acute abnormalities.
IMPRESSION:
1. Mild cardiomegaly.
2. 8mm nodular opacity in the right lower lobe, recommend follow-up CT for further evaluation.
3. No acute pulmonary parenchymal abnormality.
RECOMMENDATIONS: Follow-up chest CT to further characterize the nodular opacity in the right lower lobe.
"""
# Get sample image path if available
sample_images_dir = Path(parent_dir) / "data" / "sample"
sample_images = list(sample_images_dir.glob("*.png")) + list(
sample_images_dir.glob("*.jpg")
)
sample_image_path = None
if sample_images:
sample_image_path = str(sample_images[0])
# Define interface
with gr.Blocks(
title="MediSync: Multi-Modal Medical Analysis System", theme=gr.themes.Soft()
) as interface:
gr.Markdown("""
# MediSync: Multi-Modal Medical Analysis System
This AI-powered healthcare solution combines X-ray image analysis with patient report text processing
to provide comprehensive medical insights.
## How to Use
1. Upload a chest X-ray image
2. Enter the corresponding medical report text
3. Choose the analysis type: image-only, text-only, or multimodal (combined)
""")
with gr.Tab("Multimodal Analysis"):
with gr.Row():
with gr.Column():
multi_img_input = gr.Image(label="Upload X-ray Image", type="pil")
multi_img_enhance = gr.Button("Enhance Image")
multi_text_input = gr.Textbox(
label="Enter Medical Report Text",
placeholder="Enter the radiologist's report text here...",
lines=10,
value=example_report if sample_image_path is None else None,
)
multi_analyze_btn = gr.Button(
"Analyze Image & Text", variant="primary"
)
with gr.Column():
multi_results = gr.HTML(label="Analysis Results")
multi_plot = gr.HTML(label="Visualization")
# Set up examples if sample image exists
if sample_image_path:
gr.Examples(
examples=[[sample_image_path, example_report]],
inputs=[multi_img_input, multi_text_input],
label="Example X-ray and Report",
)
with gr.Tab("Image Analysis"):
with gr.Row():
with gr.Column():
img_input = gr.Image(label="Upload X-ray Image", type="pil")
img_enhance = gr.Button("Enhance Image")
img_analyze_btn = gr.Button("Analyze Image", variant="primary")
with gr.Column():
img_output = gr.Image(label="Processed Image")
img_results = gr.HTML(label="Analysis Results")
img_plot = gr.HTML(label="Visualization")
# Set up example if sample image exists
if sample_image_path:
gr.Examples(
examples=[[sample_image_path]],
inputs=[img_input],
label="Example X-ray Image",
)
with gr.Tab("Text Analysis"):
with gr.Row():
with gr.Column():
text_input = gr.Textbox(
label="Enter Medical Report Text",
placeholder="Enter the radiologist's report text here...",
lines=10,
value=example_report,
)
text_analyze_btn = gr.Button("Analyze Text", variant="primary")
with gr.Column():
text_output = gr.Textbox(label="Processed Text")
text_results = gr.HTML(label="Analysis Results")
text_plot = gr.HTML(label="Entity Visualization")
# Set up example
gr.Examples(
examples=[[example_report]],
inputs=[text_input],
label="Example Medical Report",
)
with gr.Tab("About"):
gr.Markdown("""
## About MediSync
MediSync is an AI-powered healthcare solution that uses multi-modal analysis to provide comprehensive insights from medical images and reports.
### Key Features
- **X-ray Image Analysis**: Detects abnormalities in chest X-rays using pre-trained vision models
- **Medical Report Processing**: Extracts key information from patient reports using NLP models
- **Multi-modal Integration**: Combines insights from both image and text data for more accurate analysis
### Models Used
- **X-ray Analysis**: facebook/deit-base-patch16-224-medical-cxr
- **Medical Text Analysis**: medicalai/ClinicalBERT
### Important Disclaimer
This tool is for educational and research purposes only. It is not intended to provide medical advice or replace professional healthcare. Always consult with qualified healthcare providers for medical decisions.
""")
# Set up event handlers
multi_img_enhance.click(
app.enhance_image, inputs=multi_img_input, outputs=multi_img_input
)
multi_analyze_btn.click(
app.analyze_multimodal,
inputs=[multi_img_input, multi_text_input],
outputs=[multi_results, multi_plot],
)
img_enhance.click(app.enhance_image, inputs=img_input, outputs=img_output)
img_analyze_btn.click(
app.analyze_image,
inputs=img_input,
outputs=[img_output, img_results, img_plot],
)
text_analyze_btn.click(
app.analyze_text,
inputs=text_input,
outputs=[text_output, text_results, text_plot],
)
# Run the interface
interface.launch()
if __name__ == "__main__":
create_interface()