File size: 23,466 Bytes
5cdf85c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 |
import logging
from .image_analyzer import XRayImageAnalyzer
from .text_analyzer import MedicalReportAnalyzer
class MultimodalFusion:
"""
A class for fusing insights from image analysis and text analysis of medical data.
This fusion approach combines the strengths of both modalities:
- Images provide visual evidence of abnormalities
- Text reports provide context, history and radiologist interpretations
The combined analysis provides a more comprehensive understanding than either modality alone.
"""
def __init__(self, image_model=None, text_model=None, device=None):
"""
Initialize the multimodal fusion module with image and text analyzers.
Args:
image_model (str, optional): Model to use for image analysis
text_model (str, optional): Model to use for text analysis
device (str, optional): Device to run models on ('cuda' or 'cpu')
"""
self.logger = logging.getLogger(__name__)
# Determine device
if device is None:
import torch
self.device = "cuda" if torch.cuda.is_available() else "cpu"
else:
self.device = device
self.logger.info(f"Using device: {self.device}")
# Initialize image analyzer
try:
self.image_analyzer = XRayImageAnalyzer(
model_name=image_model
if image_model
else "codewithdark/vit-chest-xray",
device=self.device,
)
self.logger.info("Successfully initialized image analyzer")
except Exception as e:
self.logger.error(f"Failed to initialize image analyzer: {e}")
self.image_analyzer = None
# Initialize text analyzer
try:
self.text_analyzer = MedicalReportAnalyzer(
classifier_model=text_model if text_model else "medicalai/ClinicalBERT",
device=self.device,
)
self.logger.info("Successfully initialized text analyzer")
except Exception as e:
self.logger.error(f"Failed to initialize text analyzer: {e}")
self.text_analyzer = None
def analyze_image(self, image_path):
"""
Analyze a medical image.
Args:
image_path (str): Path to the medical image
Returns:
dict: Image analysis results
"""
if not self.image_analyzer:
self.logger.warning("Image analyzer not available")
return {"error": "Image analyzer not available"}
try:
return self.image_analyzer.analyze(image_path)
except Exception as e:
self.logger.error(f"Error analyzing image: {e}")
return {"error": str(e)}
def analyze_text(self, text):
"""
Analyze medical report text.
Args:
text (str): Medical report text
Returns:
dict: Text analysis results
"""
if not self.text_analyzer:
self.logger.warning("Text analyzer not available")
return {"error": "Text analyzer not available"}
try:
return self.text_analyzer.analyze(text)
except Exception as e:
self.logger.error(f"Error analyzing text: {e}")
return {"error": str(e)}
def _calculate_agreement_score(self, image_results, text_results):
"""
Calculate agreement score between image and text analyses.
Args:
image_results (dict): Results from image analysis
text_results (dict): Results from text analysis
Returns:
float: Agreement score (0-1, where 1 is perfect agreement)
"""
try:
# Default to neutral agreement
agreement = 0.5
# Check if image detected abnormality
image_abnormal = image_results.get("has_abnormality", False)
# Check text severity
text_severity = text_results.get("severity", {}).get("level", "Unknown")
text_abnormal = text_severity not in ["Normal", "Unknown"]
# Basic agreement check
if image_abnormal == text_abnormal:
agreement += 0.25
else:
agreement -= 0.25
# Check if specific findings match
image_finding = image_results.get("primary_finding", "").lower()
# Extract problem entities from text
problems = text_results.get("entities", {}).get("problem", [])
problem_text = " ".join(problems).lower()
# Check for common keywords in both
common_conditions = [
"pneumonia",
"effusion",
"nodule",
"mass",
"cardiomegaly",
"opacity",
"fracture",
"tumor",
"edema",
]
matching_conditions = 0
total_mentioned = 0
for condition in common_conditions:
in_image = condition in image_finding
in_text = condition in problem_text
if in_image or in_text:
total_mentioned += 1
if in_image and in_text:
matching_conditions += 1
agreement += 0.05 # Boost agreement for each matching condition
# Calculate condition match ratio if any conditions were mentioned
if total_mentioned > 0:
match_ratio = matching_conditions / total_mentioned
agreement += match_ratio * 0.2
# Normalize agreement to 0-1 range
agreement = max(0, min(1, agreement))
return agreement
except Exception as e:
self.logger.error(f"Error calculating agreement score: {e}")
return 0.5 # Return neutral agreement on error
def _get_confidence_weighted_finding(self, image_results, text_results, agreement):
"""
Get the most confident finding weighted by modality confidence.
Args:
image_results (dict): Results from image analysis
text_results (dict): Results from text analysis
agreement (float): Agreement score between modalities
Returns:
str: Most confident finding
"""
try:
image_finding = image_results.get("primary_finding", "")
image_confidence = image_results.get("confidence", 0.5)
# For text, use the most severe problem as primary finding
problems = text_results.get("entities", {}).get("problem", [])
text_confidence = text_results.get("severity", {}).get("confidence", 0.5)
if not problems:
# No problems identified in text
if image_confidence > 0.7:
return image_finding
else:
return "No significant findings"
# Simple confidence-weighted selection
if image_confidence > text_confidence + 0.2:
return image_finding
elif problems and text_confidence > image_confidence + 0.2:
return (
problems[0]
if isinstance(problems, list) and problems
else "Unknown finding"
)
else:
# Similar confidence, check agreement
if agreement > 0.7:
# High agreement, try to find the specific condition mentioned in both
for problem in problems:
if problem.lower() in image_finding.lower():
return problem
# Default to image finding if high confidence
if image_confidence > 0.6:
return image_finding
elif problems:
return problems[0]
else:
return image_finding
else:
# Low agreement, include both perspectives
if image_finding and problems:
return f"{image_finding} (image) / {problems[0]} (report)"
elif image_finding:
return image_finding
elif problems:
return problems[0]
else:
return "Findings unclear - review recommended"
except Exception as e:
self.logger.error(f"Error getting weighted finding: {e}")
return "Unable to determine primary finding"
def _merge_followup_recommendations(self, image_results, text_results):
"""
Merge follow-up recommendations from both modalities.
Args:
image_results (dict): Results from image analysis
text_results (dict): Results from text analysis
Returns:
list: Combined follow-up recommendations
"""
try:
# Get text-based recommendations
text_recommendations = text_results.get("followup_recommendations", [])
# Create image-based recommendations based on findings
image_recommendations = []
if image_results.get("has_abnormality", False):
primary = image_results.get("primary_finding", "")
confidence = image_results.get("confidence", 0)
if (
"nodule" in primary.lower()
or "mass" in primary.lower()
or "tumor" in primary.lower()
):
image_recommendations.append(
f"Follow-up imaging recommended to further evaluate {primary}."
)
elif "pneumonia" in primary.lower():
image_recommendations.append(
"Clinical correlation and follow-up imaging recommended."
)
elif confidence > 0.8:
image_recommendations.append(
f"Consider follow-up imaging to monitor {primary}."
)
elif confidence > 0.5:
image_recommendations.append(
"Consider clinical correlation and potential follow-up."
)
# Combine recommendations, removing duplicates
all_recommendations = text_recommendations + image_recommendations
# Remove near-duplicates (similar recommendations)
unique_recommendations = []
for rec in all_recommendations:
if not any(
self._is_similar_recommendation(rec, existing)
for existing in unique_recommendations
):
unique_recommendations.append(rec)
return unique_recommendations
except Exception as e:
self.logger.error(f"Error merging follow-up recommendations: {e}")
return ["Follow-up recommended based on findings."]
def _is_similar_recommendation(self, rec1, rec2):
"""Check if two recommendations are semantically similar."""
# Convert to lowercase for comparison
rec1_lower = rec1.lower()
rec2_lower = rec2.lower()
# Check for significant overlap
words1 = set(rec1_lower.split())
words2 = set(rec2_lower.split())
# Calculate Jaccard similarity
intersection = words1.intersection(words2)
union = words1.union(words2)
similarity = len(intersection) / len(union) if union else 0
# Consider similar if more than 60% overlap
return similarity > 0.6
def _get_final_severity(self, image_results, text_results, agreement):
"""
Determine final severity based on both modalities.
Args:
image_results (dict): Results from image analysis
text_results (dict): Results from text analysis
agreement (float): Agreement score between modalities
Returns:
dict: Final severity assessment
"""
try:
# Get text-based severity
text_severity = text_results.get("severity", {})
text_level = text_severity.get("level", "Unknown")
text_score = text_severity.get("score", 0)
text_confidence = text_severity.get("confidence", 0.5)
# Convert image findings to severity
image_abnormal = image_results.get("has_abnormality", False)
image_confidence = image_results.get("confidence", 0.5)
# Default severity mapping from image
image_severity = "Normal" if not image_abnormal else "Moderate"
image_score = 0 if not image_abnormal else 2.0
# Adjust image severity based on specific findings
primary_finding = image_results.get("primary_finding", "").lower()
# Map certain conditions to severity levels
severity_mapping = {
"pneumonia": ("Moderate", 2.5),
"pneumothorax": ("Severe", 3.0),
"effusion": ("Moderate", 2.0),
"pulmonary edema": ("Moderate", 2.5),
"nodule": ("Mild", 1.5),
"mass": ("Moderate", 2.5),
"tumor": ("Severe", 3.0),
"cardiomegaly": ("Mild", 1.5),
"fracture": ("Moderate", 2.0),
"consolidation": ("Moderate", 2.0),
}
# Check if any key terms are in the primary finding
for key, (severity, score) in severity_mapping.items():
if key in primary_finding:
image_severity = severity
image_score = score
break
# Weight based on confidence and agreement
if agreement > 0.7:
# High agreement - weight equally
final_score = (image_score + text_score) / 2
else:
# Lower agreement - weight by confidence
total_confidence = image_confidence + text_confidence
if total_confidence > 0:
image_weight = image_confidence / total_confidence
text_weight = text_confidence / total_confidence
final_score = (image_score * image_weight) + (
text_score * text_weight
)
else:
final_score = (image_score + text_score) / 2
# Map score to severity level
severity_levels = {
0: "Normal",
1: "Mild",
2: "Moderate",
3: "Severe",
4: "Critical",
}
# Round to nearest level
level_index = round(min(4, max(0, final_score)))
final_level = severity_levels[level_index]
return {
"level": final_level,
"score": round(final_score, 1),
"confidence": round((image_confidence + text_confidence) / 2, 2),
}
except Exception as e:
self.logger.error(f"Error determining final severity: {e}")
return {"level": "Unknown", "score": 0, "confidence": 0}
def fuse_analyses(self, image_results, text_results):
"""
Fuse the results from image and text analyses.
Args:
image_results (dict): Results from image analysis
text_results (dict): Results from text analysis
Returns:
dict: Fused analysis results
"""
try:
# Calculate agreement between modalities
agreement = self._calculate_agreement_score(image_results, text_results)
self.logger.info(f"Agreement score between modalities: {agreement:.2f}")
# Get confidence-weighted primary finding
primary_finding = self._get_confidence_weighted_finding(
image_results, text_results, agreement
)
# Merge follow-up recommendations
followup = self._merge_followup_recommendations(image_results, text_results)
# Get final severity assessment
severity = self._get_final_severity(image_results, text_results, agreement)
# Create comprehensive findings list
findings = []
# Add text-extracted findings
text_findings = text_results.get("findings", [])
if text_findings:
findings.extend(text_findings)
# Add primary image finding if not already included
image_finding = image_results.get("primary_finding", "")
if image_finding and not any(
image_finding.lower() in f.lower() for f in findings
):
findings.append(f"Image finding: {image_finding}")
# Create fused result
fused_result = {
"agreement_score": round(agreement, 2),
"primary_finding": primary_finding,
"severity": severity,
"findings": findings,
"followup_recommendations": followup,
"modality_results": {"image": image_results, "text": text_results},
}
return fused_result
except Exception as e:
self.logger.error(f"Error fusing analyses: {e}")
return {
"error": str(e),
"modality_results": {"image": image_results, "text": text_results},
}
def analyze(self, image_path, report_text):
"""
Perform multimodal analysis of medical image and report.
Args:
image_path (str): Path to the medical image
report_text (str): Medical report text
Returns:
dict: Fused analysis results
"""
try:
# Analyze image
image_results = self.analyze_image(image_path)
# Analyze text
text_results = self.analyze_text(report_text)
# Fuse the analyses
return self.fuse_analyses(image_results, text_results)
except Exception as e:
self.logger.error(f"Error in multimodal analysis: {e}")
return {"error": str(e)}
def get_explanation(self, fused_results):
"""
Generate a human-readable explanation of the fused analysis.
Args:
fused_results (dict): Results from the fused analysis
Returns:
str: A text explanation of the fused analysis
"""
try:
explanation = []
# Add overview section
primary_finding = fused_results.get("primary_finding", "Unknown")
severity = fused_results.get("severity", {}).get("level", "Unknown")
explanation.append("# Medical Analysis Summary\n")
explanation.append("## Overview\n")
explanation.append(f"Primary finding: **{primary_finding}**\n")
explanation.append(f"Severity level: **{severity}**\n")
# Add agreement information
agreement = fused_results.get("agreement_score", 0)
agreement_text = (
"High" if agreement > 0.7 else "Moderate" if agreement > 0.4 else "Low"
)
explanation.append(
f"Image and text analysis agreement: **{agreement_text}** ({agreement:.0%})\n"
)
# Add findings section
explanation.append("\n## Detailed Findings\n")
findings = fused_results.get("findings", [])
if findings:
for finding in findings:
explanation.append(f"- {finding}\n")
else:
explanation.append("No specific findings detailed.\n")
# Add follow-up section
explanation.append("\n## Recommended Follow-up\n")
followups = fused_results.get("followup_recommendations", [])
if followups:
for followup in followups:
explanation.append(f"- {followup}\n")
else:
explanation.append("No specific follow-up recommendations provided.\n")
# Add confidence note
confidence = fused_results.get("severity", {}).get("confidence", 0)
explanation.append(
f"\n*Note: This analysis has a confidence level of {confidence:.0%}. "
f"Please consult with healthcare professionals for official diagnosis.*"
)
return "\n".join(explanation)
except Exception as e:
self.logger.error(f"Error generating explanation: {e}")
return "Error generating analysis explanation."
# Example usage
if __name__ == "__main__":
# Set up logging
logging.basicConfig(level=logging.INFO)
# Test on sample data if available
import os
fusion = MultimodalFusion()
# Sample text report
sample_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.
"""
# Check if sample data directory exists and contains images
sample_dir = "../data/sample"
if os.path.exists(sample_dir) and os.listdir(sample_dir):
sample_image = os.path.join(sample_dir, os.listdir(sample_dir)[0])
print(f"Analyzing sample image: {sample_image}")
# Perform multimodal analysis
fused_results = fusion.analyze(sample_image, sample_report)
explanation = fusion.get_explanation(fused_results)
print("\nFused Analysis Results:")
print(explanation)
else:
print("No sample images found. Only analyzing text report.")
# Analyze just the text
text_results = fusion.analyze_text(sample_report)
print("\nText Analysis Results:")
print(
f"Severity: {text_results['severity']['level']} (Score: {text_results['severity']['score']})"
)
print("\nKey Findings:")
for finding in text_results["findings"]:
print(f"- {finding}")
print("\nEntities:")
for category, items in text_results["entities"].items():
if items:
print(f"- {category.capitalize()}: {', '.join(items)}")
print("\nFollow-up Recommendations:")
for rec in text_results["followup_recommendations"]:
print(f"- {rec}")
|