File size: 16,852 Bytes
86a74e6 |
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 |
import logging
import re
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline
class MedicalReportAnalyzer:
"""
A class for analyzing medical text reports using pre-trained NLP models from Hugging Face.
This analyzer can:
1. Extract medical entities (conditions, treatments, tests)
2. Classify report severity
3. Extract key findings
4. Identify suggested follow-up actions
"""
def __init__(
self,
ner_model="samrawal/bert-base-uncased_medical-ner",
classifier_model="medicalai/ClinicalBERT",
device=None,
):
"""
Initialize the text analyzer with specific pre-trained models.
Args:
ner_model (str): Model for named entity recognition
classifier_model (str): Model for text classification
device (str, optional): Device to run models on ('cuda' or 'cpu')
"""
self.logger = logging.getLogger(__name__)
# Determine device
if device is None:
self.device = "cuda" if torch.cuda.is_available() else "cpu"
else:
self.device = device
self.logger.info(f"Using device: {self.device}")
# Load NER model for entity extraction
try:
self.ner_pipeline = pipeline(
"token-classification",
model=ner_model,
aggregation_strategy="simple",
device=0 if self.device == "cuda" else -1,
)
self.logger.info(f"Successfully loaded NER model: {ner_model}")
except Exception as e:
self.logger.error(f"Failed to load NER model: {e}")
self.ner_pipeline = None
# Load classifier model for severity assessment
try:
self.tokenizer = AutoTokenizer.from_pretrained(classifier_model)
self.classifier = AutoModelForSequenceClassification.from_pretrained(
classifier_model
)
self.classifier.to(self.device)
self.classifier.eval()
self.logger.info(
f"Successfully loaded classifier model: {classifier_model}"
)
except Exception as e:
self.logger.error(f"Failed to load classifier model: {e}")
self.classifier = None
# Severity levels mapping
self.severity_levels = {
0: "Normal",
1: "Mild",
2: "Moderate",
3: "Severe",
4: "Critical",
}
# Common medical findings and their severity levels
self.finding_severity = {
"pneumonia": 3,
"fracture": 3,
"tumor": 4,
"nodule": 2,
"mass": 3,
"edema": 2,
"effusion": 2,
"hemorrhage": 3,
"opacity": 1,
"atelectasis": 2,
"pneumothorax": 3,
"consolidation": 2,
"cardiomegaly": 2,
}
def extract_entities(self, text):
"""
Extract medical entities from the report text.
Args:
text (str): Medical report text
Returns:
dict: Dictionary of entity lists by category
"""
if not self.ner_pipeline:
self.logger.warning("NER model not available")
return {}
try:
# Run NER
entities = self.ner_pipeline(text)
# Group entities by type
grouped_entities = {
"problem": [], # Medical conditions
"test": [], # Tests/procedures
"treatment": [], # Treatments/medications
"anatomy": [], # Anatomical locations
}
for entity in entities:
entity_type = entity.get("entity_group", "").lower()
# Map entity types to our categories
if entity_type in ["problem", "disease", "condition", "diagnosis"]:
category = "problem"
elif entity_type in ["test", "procedure", "examination"]:
category = "test"
elif entity_type in ["treatment", "medication", "drug"]:
category = "treatment"
elif entity_type in ["body_part", "anatomy", "organ"]:
category = "anatomy"
else:
continue # Skip other entity types
word = entity.get("word", "")
score = entity.get("score", 0)
# Only include if confidence is reasonable
if score > 0.7 and word not in grouped_entities[category]:
grouped_entities[category].append(word)
return grouped_entities
except Exception as e:
self.logger.error(f"Error extracting entities: {e}")
return {}
def assess_severity(self, text):
"""
Assess the severity level of the medical report.
Args:
text (str): Medical report text
Returns:
dict: Severity assessment including level and confidence
"""
if not self.classifier:
self.logger.warning("Classifier model not available")
return {"level": "Unknown", "score": 0.0}
try:
# Use rule-based approach along with model
severity_score = 0
confidence = 0.5 # Start with neutral confidence
# Check for severe keywords
severe_keywords = [
"severe",
"critical",
"urgent",
"emergency",
"immediate attention",
]
moderate_keywords = ["moderate", "concerning", "follow-up", "monitor"]
mild_keywords = ["mild", "minimal", "slight", "minor"]
normal_keywords = [
"normal",
"unremarkable",
"no abnormalities",
"within normal limits",
]
# Count keyword occurrences
text_lower = text.lower()
severe_count = sum(text_lower.count(word) for word in severe_keywords)
moderate_count = sum(text_lower.count(word) for word in moderate_keywords)
mild_count = sum(text_lower.count(word) for word in mild_keywords)
normal_count = sum(text_lower.count(word) for word in normal_keywords)
# Adjust severity based on keyword counts
if severe_count > 0:
severity_score += min(severe_count, 2) * 1.5
confidence += 0.1
if moderate_count > 0:
severity_score += min(moderate_count, 3) * 0.75
confidence += 0.05
if mild_count > 0:
severity_score += min(mild_count, 3) * 0.25
confidence += 0.05
if normal_count > 0:
severity_score -= min(normal_count, 3) * 0.75
confidence += 0.1
# Check for specific medical findings
for finding, level in self.finding_severity.items():
if finding in text_lower:
severity_score += level * 0.5
confidence += 0.05
# Normalize severity score to 0-4 range
severity_score = max(0, min(4, severity_score))
severity_level = int(round(severity_score))
# Map to severity level
severity = self.severity_levels.get(severity_level, "Moderate")
# Cap confidence at 0.95
confidence = min(0.95, confidence)
return {
"level": severity,
"score": round(severity_score, 1),
"confidence": round(confidence, 2),
}
except Exception as e:
self.logger.error(f"Error assessing severity: {e}")
return {"level": "Unknown", "score": 0.0, "confidence": 0.0}
def extract_findings(self, text):
"""
Extract key clinical findings from the report.
Args:
text (str): Medical report text
Returns:
list: List of key findings
"""
try:
# Split text into sentences
sentences = re.split(r"[.!?]\s+", text)
findings = []
# Key phrases that often introduce findings
finding_markers = [
"finding",
"observed",
"noted",
"shows",
"reveals",
"demonstrates",
"indicates",
"evident",
"apparent",
"consistent with",
"suggestive of",
]
# Negative markers
negation_markers = ["no", "not", "none", "negative", "without", "denies"]
for sentence in sentences:
# Skip very short sentences
if len(sentence.split()) < 3:
continue
sentence = sentence.strip()
# Check if this sentence likely contains a finding
contains_finding_marker = any(
marker in sentence.lower() for marker in finding_markers
)
# Check for negation
contains_negation = any(
marker in sentence.lower().split() for marker in negation_markers
)
# Only include positive findings or explicitly negated findings that are important
if contains_finding_marker or (
contains_negation
and any(
term in sentence.lower()
for term in self.finding_severity.keys()
)
):
findings.append(sentence)
return findings
except Exception as e:
self.logger.error(f"Error extracting findings: {e}")
return []
def suggest_followup(self, text, entities, severity):
"""
Suggest follow-up actions based on report analysis.
Args:
text (str): Medical report text
entities (dict): Extracted entities
severity (dict): Severity assessment
Returns:
list: Suggested follow-up actions
"""
try:
followups = []
# Base recommendations on severity
severity_level = severity.get("level", "Unknown")
severity_score = severity.get("score", 0)
# Extract problems from entities
problems = entities.get("problem", [])
# Check if follow-up is already mentioned in the text
followup_mentioned = any(
phrase in text.lower()
for phrase in [
"follow up",
"follow-up",
"followup",
"return",
"refer",
"consult",
]
)
# Default recommendations based on severity
if severity_level == "Critical":
followups.append("Immediate specialist consultation recommended.")
elif severity_level == "Severe":
followups.append("Prompt follow-up with specialist is recommended.")
# Add specific recommendations for common severe conditions
for problem in problems:
if "pneumonia" in problem.lower():
followups.append(
"Consider antibiotic therapy and close monitoring."
)
elif "fracture" in problem.lower():
followups.append(
"Orthopedic consultation for treatment planning."
)
elif "mass" in problem.lower() or "tumor" in problem.lower():
followups.append(
"Further imaging and possible biopsy recommended."
)
elif severity_level == "Moderate":
followups.append("Follow-up with primary care physician recommended.")
if not followup_mentioned and problems:
followups.append(
"Consider additional imaging or tests for further evaluation."
)
elif severity_level == "Mild":
if problems:
followups.append(
"Routine follow-up with primary care physician as needed."
)
else:
followups.append("No immediate follow-up required.")
else: # Normal
followups.append(
"No specific follow-up indicated based on this report."
)
# Check for specific findings that always need follow-up
for critical_term in ["mass", "tumor", "nodule", "opacity"]:
if (
critical_term in text.lower()
and "follow-up" not in " ".join(followups).lower()
):
followups.append(
f"Follow-up imaging recommended to monitor {critical_term}."
)
break
return followups
except Exception as e:
self.logger.error(f"Error suggesting follow-up: {e}")
return ["Unable to generate follow-up recommendations."]
def analyze(self, text):
"""
Perform comprehensive analysis of medical report text.
Args:
text (str): Medical report text
Returns:
dict: Complete analysis results
"""
try:
# Extract entities
entities = self.extract_entities(text)
# Assess severity
severity = self.assess_severity(text)
# Extract key findings
findings = self.extract_findings(text)
# Generate follow-up suggestions
followups = self.suggest_followup(text, entities, severity)
# Create detailed report
report = {
"entities": entities,
"severity": severity,
"findings": findings,
"followup_recommendations": followups,
}
return report
except Exception as e:
self.logger.error(f"Error analyzing report: {e}")
return {"error": str(e)}
# Example usage
if __name__ == "__main__":
# Set up logging
logging.basicConfig(level=logging.INFO)
# Test on a sample report
analyzer = MedicalReportAnalyzer()
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.
"""
results = analyzer.analyze(sample_report)
print("\nMedical Report Analysis:")
print(
f"\nSeverity: {results['severity']['level']} (Score: {results['severity']['score']})"
)
print("\nKey Findings:")
for finding in results["findings"]:
print(f"- {finding}")
print("\nEntities:")
for category, items in results["entities"].items():
if items:
print(f"- {category.capitalize()}: {', '.join(items)}")
print("\nFollow-up Recommendations:")
for rec in results["followup_recommendations"]:
print(f"- {rec}")
|