models/text_analyzer.py

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline


class TextAnalyzer:
    def __init__(self):
        # Load the medical text analysis model
        try:
            model_name = "medicalai/ClinicalBERT"
            self.tokenizer = AutoTokenizer.from_pretrained(model_name)
            self.model = AutoModelForSequenceClassification.from_pretrained(model_name)
            self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
            self.model.to(self.device)

            # NER pipeline for medical entities
            self.ner_pipeline = pipeline(
                "ner", model="samrawal/bert-base-uncased_medical-ner"
            )
            print(f"Text model loaded on {self.device}")
        except Exception as e:
            print(f"Error loading text model: {e}")
            # Fallback to dummy functionality
            self.model = None
            self.tokenizer = None
            self.ner_pipeline = None

    def analyze(self, text):
        """Analyze medical report text and extract key insights"""
        if text.strip() == "":
            return {"Insights": "No text provided"}

        if self.model is None or self.tokenizer is None:
            # Dummy analysis
            return {
                "Entities": ["fever", "cough"],
                "Sentiment": "Concerning",
                "Key findings": "Patient shows symptoms of respiratory illness",
            }

        try:
            # Extract medical entities
            if self.ner_pipeline:
                entities = self.ner_pipeline(text)
                unique_entities = list(set([entity["word"] for entity in entities]))
            else:
                unique_entities = []

            # Simple text classification (in real app, would be more sophisticated)
            inputs = self.tokenizer(
                text, return_tensors="pt", padding=True, truncation=True
            ).to(self.device)
            with torch.no_grad():
                outputs = self.model(**inputs)

            # This is a placeholder - in reality would depend on the actual model output
            sentiment = (
                "Concerning" if torch.sigmoid(outputs.logits).item() > 0.5 else "Normal"
            )

            # Generate key findings (simplified)
            key_findings = f"Report indicates {'abnormal' if sentiment == 'Concerning' else 'normal'} findings"
            if unique_entities:
                key_findings += f" with mentions of {', '.join(unique_entities[:5])}"

            return {
                "Entities": unique_entities[:10],
                "Sentiment": sentiment,
                "Key findings": key_findings,
            }
        except Exception as e:
            print(f"Error during text analysis: {e}")
            return {"Error": "Could not analyze text"}