feat: pipeline T2A - anonymisation, extraction CIM-10 et intégration edsnlp

Pipeline complet de traitement de documents médicaux PDF : - Extraction texte (pdfplumber) et classification (Trackare/CRH) - Anonymisation multi-couche (regex + NER CamemBERT + sweep) - Extraction médicale CIM-10 hybride : edsnlp (AP-HP) enrichit les diagnostics, médicaments (codes ATC via Romedi) et négation, avec fallback regex pour les patterns spécifiques - Fix sentencepiece pinné à <0.2.0 pour compatibilité CamemBERT Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-10 15:24:12 +01:00
commit 4a12cd2676
25 changed files with 7592 additions and 0 deletions
--- a/tests/init.py
+++ b/tests/init.py
--- a/tests/test_anonymization.py
+++ b/tests/test_anonymization.py
@@ -0,0 +1,197 @@
+"""Tests pour le module d'anonymisation."""
+
+import pytest
+
+from src.anonymization.entity_registry import EntityRegistry
+from src.anonymization.regex_patterns import (
+    CRH_FOOTER_IPP_EPISODE,
+    CRH_FOOTER_PATIENT_PATTERN,
+    DATE_NAISSANCE_PATTERN,
+    DR_NAME_PATTERN,
+    EMAIL_PATTERN,
+    EPISODE_PATTERN,
+    FOOTER_PATIENT_PATTERN,
+    IPP_PATTERN,
+    NOTE_AUTHOR_PATTERN,
+    PHONE_PATTERN,
+    RPPS_PATTERN,
+)
+
+
+class TestRegexPatterns:
+    def test_ipp_with_colon(self):
+        m = IPP_PATTERN.search("IPP: 01306172")
+        assert m is not None
+        assert m.group(1) == "01306172"
+
+    def test_ipp_without_colon(self):
+        m = IPP_PATTERN.search("IPP 01306172")
+        assert m is not None
+        assert m.group(1) == "01306172"
+
+    def test_ipp_in_parentheses(self):
+        m = IPP_PATTERN.search("(01306172 )")
+        assert m is not None
+        assert m.group(2) == "01306172"
+
+    def test_episode_no(self):
+        m = EPISODE_PATTERN.search("Episode No: 23042753")
+        assert m is not None
+        assert m.group(1) == "23042753"
+
+    def test_episode_n_degree(self):
+        m = EPISODE_PATTERN.search("N° Episode 23042753")
+        assert m is not None
+        assert m.group(2) == "23042753"
+
+    def test_phone_dots(self):
+        m = PHONE_PATTERN.search("06.25.39.26.82")
+        assert m is not None
+        assert m.group(0) == "06.25.39.26.82"
+
+    def test_phone_spaces(self):
+        m = PHONE_PATTERN.search("05 59 44 35 35")
+        assert m is not None
+
+    def test_email(self):
+        m = EMAIL_PATTERN.search("faudemar@ch-cotebasque.fr")
+        assert m is not None
+        assert m.group(0) == "faudemar@ch-cotebasque.fr"
+
+    def test_rpps(self):
+        m = RPPS_PATTERN.search("RPPS : 10100532760")
+        assert m is not None
+        assert m.group(1) == "10100532760"
+
+    def test_date_naissance_nee_le(self):
+        m = DATE_NAISSANCE_PATTERN.search("née le 23/02/1980")
+        assert m is not None
+        assert m.group(1) == "23/02/1980"
+
+    def test_date_naissance_ne_e_le(self):
+        m = DATE_NAISSANCE_PATTERN.search("Né(e) le 23/02/1980")
+        assert m is not None
+        assert m.group(1) == "23/02/1980"
+
+    def test_date_naissance_field(self):
+        m = DATE_NAISSANCE_PATTERN.search("Date de naissance: 23/02/1980")
+        assert m is not None
+        assert m.group(1) == "23/02/1980"
+
+    def test_dr_name(self):
+        m = DR_NAME_PATTERN.search("Dr F. AUDEMAR")
+        assert m is not None
+        assert "AUDEMAR" in m.group(1)
+
+    def test_dr_name_docteur(self):
+        m = DR_NAME_PATTERN.search("Docteur AUDEMAR Franck")
+        assert m is not None
+        assert "AUDEMAR" in m.group(1)
+
+    def test_note_author(self):
+        m = NOTE_AUTHOR_PATTERN.search("Note IDE Annie GUIRESSE Non algique")
+        assert m is not None
+        assert m.group(1) == "Annie GUIRESSE"
+
+    def test_footer_patient_trackare(self):
+        m = FOOTER_PATIENT_PATTERN.search(
+            "Patient: CLIER NARBAIS AUDREY - Date de naissance: 23/02/1980"
+        )
+        assert m is not None
+        assert "CLIER" in m.group(1)
+
+    def test_crh_footer_patient(self):
+        m = CRH_FOOTER_PATIENT_PATTERN.search(
+            "Patient(e) : CLIER AUDREY NARBAIS Né(e) le 23/02/1980"
+        )
+        assert m is not None
+        assert "CLIER" in m.group(1)
+
+    def test_crh_footer_ipp_episode(self):
+        m = CRH_FOOTER_IPP_EPISODE.search(
+            "IPP 01306172 / N° Episode 23042753 (MEDECINE GASTRO B2 HC)"
+        )
+        assert m is not None
+        assert m.group(1) == "01306172"
+        assert m.group(2) == "23042753"
+
+
+class TestEntityRegistry:
+    def test_register_returns_pseudo(self):
+        reg = EntityRegistry()
+        pseudo = reg.register("Jean Dupont", "patient")
+        assert pseudo == "[PATIENT_1]"
+
+    def test_register_same_entity_returns_same(self):
+        reg = EntityRegistry()
+        p1 = reg.register("Jean Dupont", "patient")
+        p2 = reg.register("Jean Dupont", "patient")
+        assert p1 == p2
+
+    def test_register_case_insensitive(self):
+        reg = EntityRegistry()
+        p1 = reg.register("Jean DUPONT", "patient")
+        p2 = reg.register("jean dupont", "patient")
+        assert p1 == p2
+
+    def test_register_different_categories(self):
+        reg = EntityRegistry()
+        p1 = reg.register("Dupont", "patient")
+        p2 = reg.register("Martin", "medecin")
+        assert p1 == "[PATIENT_1]"
+        assert p2 == "[MEDECIN_1]"
+
+    def test_get_replacement(self):
+        reg = EntityRegistry()
+        reg.register("Jean Dupont", "patient")
+        assert reg.get_replacement("jean dupont") == "[PATIENT_1]"
+        assert reg.get_replacement("inconnu") is None
+
+
+class TestAnonymizer:
+    def test_anonymize_basic(self):
+        from src.anonymization.anonymizer import Anonymizer
+
+        parsed = {
+            "patient": {"nom_prenom": "DUPONT Jean", "nom_naissance": "DUPONT"},
+            "medecins": ["MARTIN Pierre"],
+            "contacts": [],
+        }
+        anonymizer = Anonymizer(parsed_data=parsed)
+        text = "Le patient DUPONT Jean a été vu par Dr MARTIN Pierre."
+        result = anonymizer.anonymize(text)
+
+        assert "DUPONT" not in result
+        assert "MARTIN" not in result
+        assert "[PATIENT" in result or "[MEDECIN" in result
+
+    def test_preserves_medical_content(self):
+        from src.anonymization.anonymizer import Anonymizer
+
+        anonymizer = Anonymizer(parsed_data={"patient": {}, "medecins": [], "contacts": []})
+        text = "Pancréatite aiguë biliaire. Cholécystectomie par cœlioscopie. IMC 34.37."
+        result = anonymizer.anonymize(text)
+
+        assert "Pancréatite" in result
+        assert "Cholécystectomie" in result
+        assert "IMC" in result
+
+    def test_anonymize_phone(self):
+        from src.anonymization.anonymizer import Anonymizer
+
+        anonymizer = Anonymizer(parsed_data={"patient": {}, "medecins": [], "contacts": []})
+        text = "Appeler le 06.25.39.26.82 pour le rendez-vous."
+        result = anonymizer.anonymize(text)
+
+        assert "06.25.39.26.82" not in result
+        assert "[TEL" in result
+
+    def test_anonymize_email(self):
+        from src.anonymization.anonymizer import Anonymizer
+
+        anonymizer = Anonymizer(parsed_data={"patient": {}, "medecins": [], "contacts": []})
+        text = "Contact: faudemar@ch-cotebasque.fr"
+        result = anonymizer.anonymize(text)
+
+        assert "faudemar@ch-cotebasque.fr" not in result
+        assert "[EMAIL" in result
--- a/tests/test_extraction.py
+++ b/tests/test_extraction.py
@@ -0,0 +1,126 @@
+"""Tests pour le module d'extraction."""
+
+import pytest
+
+from src.extraction.document_classifier import classify
+from src.extraction.crh_parser import parse_crh
+from src.extraction.trackare_parser import parse_trackare, _clean_person_name
+
+
+class TestDocumentClassifier:
+    def test_classify_trackare(self):
+        text = """CENTRE HOSPITALIER COTE BASQUE
+Dossier Patient
+Détails des patients
+Nom de naissance: CLIER IPP: 01306172
+Détails épisode
+Episode No: 23042753
+Signes Vitaux"""
+        assert classify(text) == "trackare"
+
+    def test_classify_crh(self):
+        text = """N° Finess CENTRE HOSPITALIER COTE BASQUE
+Pôle Spécialités Médicales
+Service de Gastro-Entérologie
+Mon cher confrère,
+Votre patiente a été hospitalisée"""
+        assert classify(text) == "crh"
+
+    def test_classify_trackare_by_ipp(self):
+        text = "IPP: 12345678 Episode No: 87654321"
+        assert classify(text) == "trackare"
+
+
+class TestCRHParser:
+    def test_parse_patient_info(self):
+        text = """MME NARBAIS AUDREY
+MAISON IRREXELAIA
+64430 ST ETIENNE DE BAIGORRY
+
+Mon cher confrère,
+Votre patiente NARBAIS Audrey née le 23/02/1980 a été hospitalisée
+du 25/02/2023 au 03/03/2023 pour le motif suivant:
+Pancréatite aiguë lithiasique"""
+        result = parse_crh(text)
+
+        assert result["patient"]["nom_complet"] == "NARBAIS AUDREY"
+        assert result["patient"]["sexe"] == "F"
+        assert result["patient"]["date_naissance"] == "23/02/1980"
+
+    def test_parse_sejour(self):
+        text = """Votre patiente née le 23/02/1980 a été hospitalisée
+du 25/02/2023 au 03/03/2023 pour le motif suivant:
+Pancréatite aiguë"""
+        result = parse_crh(text)
+
+        assert result["sejour"]["date_entree"] == "25/02/2023"
+        assert result["sejour"]["date_sortie"] == "03/03/2023"
+
+    def test_parse_medecins(self):
+        text = "Dr PUJOS. Dr F. AUDEMAR. Docteur DUTREY Sarah."
+        result = parse_crh(text)
+
+        assert any("PUJOS" in m for m in result["medecins"])
+        assert any("AUDEMAR" in m for m in result["medecins"])
+
+
+class TestTrackareParser:
+    def test_parse_patient_info(self):
+        text = """Nom de naissance: CLIER IPP: 01306172
+Nom et Prénom: NARBAIS AUDREY Date de naissance: 23/02/1980
+Sexe: Féminin Lieu de naissance: CHAMPIGNY SUR MARNE
+Adresse: MAISON IRREXELAIA QUARTIER AUZO TTIPI Ville de résidence: ST ETIENNE DE BAIGORRY
+Code Postal: 64430
+Episode No: 23042753
+Date d'admission: 25/02/2023 Heure d'admission: 03:07
+Date de sortie: 03/03/2023
+Taille: 162 cm - Poids: 90.2 kg - IMC: 34.370"""
+        result = parse_trackare(text)
+
+        assert result["patient"]["nom_naissance"] == "CLIER"
+        assert result["patient"]["nom_prenom"] == "NARBAIS AUDREY"
+        assert result["patient"]["ipp"] == "01306172"
+        assert result["patient"]["sexe"] == "F"
+        assert result["patient"]["date_naissance"] == "23/02/1980"
+        assert result["patient"]["imc"] == 34.370
+        assert result["sejour"]["episode"] == "23042753"
+        assert result["sejour"]["date_entree"] == "25/02/2023"
+
+    def test_parse_diagnostics(self):
+        text = """Diagnostic aux urgences
+Type Etat Code Date
+Principal actif K80.5 Calcul des canaux biliaires (sans angiocholite ni cholécystite) [CMA2] 25/02/2023 05:27"""
+        result = parse_trackare(text)
+
+        assert len(result["diagnostics"]) >= 1
+        assert result["diagnostics"][0]["code_cim10"] == "K80.5"
+        assert result["diagnostics"][0]["type"] == "Principal"
+
+    def test_parse_vitals(self):
+        text = """Poids/Taille
+Taille [cm] 162,00
+Poids [kg] 90,20
+Indice
+de masse 34.370"""
+        result = parse_trackare(text)
+
+        assert result["signes_vitaux"]["taille_cm"] == 162.0
+        assert result["signes_vitaux"]["poids_kg"] >= 90.0
+        assert result["signes_vitaux"]["imc"] == 34.370
+
+
+class TestCleanPersonName:
+    def test_clean_simple(self):
+        assert _clean_person_name("Sarah DUTREY") == "Sarah DUTREY"
+
+    def test_clean_with_noise(self):
+        assert _clean_person_name("Sarah DUTREY une complication") == "Sarah DUTREY"
+
+    def test_clean_multiline(self):
+        assert _clean_person_name("Sarah\nDUTREY") == "Sarah DUTREY"
+
+    def test_clean_medical_term(self):
+        assert _clean_person_name("Bilirubine") == ""
+
+    def test_clean_empty(self):
+        assert _clean_person_name("") == ""
--- a/tests/test_integration.py
+++ b/tests/test_integration.py
@@ -0,0 +1,124 @@
+"""Tests d'intégration end-to-end sur les PDFs réels."""
+
+import json
+from pathlib import Path
+
+import pytest
+
+from src.config import INPUT_DIR, ANONYMIZED_DIR, STRUCTURED_DIR
+from src.main import process_pdf
+
+
+# Skip si les PDFs ne sont pas disponibles
+CRH_PDF = INPUT_DIR / "CRH 23042753.pdf"
+TRACKARE_PDF = INPUT_DIR / "trackare-01306172-23042753_01306172_23042753.pdf"
+
+needs_pdfs = pytest.mark.skipif(
+    not CRH_PDF.exists() or not TRACKARE_PDF.exists(),
+    reason="PDFs de test non disponibles dans input/",
+)
+
+
+# Données personnelles connues à vérifier
+PATIENT_PII = [
+    "NARBAIS", "CLIER", "AUDREY", "01306172", "23042753",
+    "23/02/1980", "IRREXELAIA", "BAIGORRY", "06.25.39.26.82",
+]
+
+SOIGNANT_NAMES = [
+    "DUTREY", "MENDIBOURE", "PUJOS", "AUDEMAR", "MELLIN",
+    "GUIRESSE", "GOYTINO", "SERRE", "NOVION",
+]
+
+
+@needs_pdfs
+class TestCRHIntegration:
+    @pytest.fixture(autouse=True)
+    def setup(self):
+        self.anonymized, self.dossier, self.report = process_pdf(CRH_PDF)
+
+    def test_no_patient_pii(self):
+        text_upper = self.anonymized.upper()
+        for pii in PATIENT_PII:
+            assert pii.upper() not in text_upper, f"PII trouvé : {pii}"
+
+    def test_medical_content_preserved(self):
+        text_lower = self.anonymized.lower()
+        for term in ["pancréatite", "cholécystectomie", "cholangiographie", "lithiase"]:
+            assert term in text_lower, f"Terme médical manquant : {term}"
+
+    def test_diagnostic_principal(self):
+        dp = self.dossier.diagnostic_principal
+        assert dp is not None
+        assert dp.cim10_suggestion == "K85.1"
+
+    def test_diagnostics_associes(self):
+        codes = {d.cim10_suggestion for d in self.dossier.diagnostics_associes}
+        assert "K80.5" in codes or "K80.2" in codes
+
+    def test_sejour(self):
+        s = self.dossier.sejour
+        assert s.sexe == "F"
+        assert s.age == 43
+        assert s.date_entree == "25/02/2023"
+        assert s.date_sortie == "03/03/2023"
+        assert s.duree_sejour == 6
+
+    def test_traitements_have_optional_atc(self):
+        """Vérifie que les traitements ont le champ code_atc (peut être None)."""
+        for t in self.dossier.traitements_sortie:
+            assert hasattr(t, "code_atc")
+
+
+@needs_pdfs
+class TestTrackareIntegration:
+    @pytest.fixture(autouse=True)
+    def setup(self):
+        self.anonymized, self.dossier, self.report = process_pdf(TRACKARE_PDF)
+
+    def test_no_patient_pii(self):
+        text_upper = self.anonymized.upper()
+        for pii in PATIENT_PII:
+            assert pii.upper() not in text_upper, f"PII trouvé : {pii}"
+
+    def test_no_soignant_names(self):
+        # Ignorer "selles" qui contient "SELLE"
+        text = self.anonymized
+        for name in SOIGNANT_NAMES:
+            # Chercher le nom comme mot complet
+            import re
+            pattern = re.compile(r"\b" + re.escape(name) + r"\b", re.IGNORECASE)
+            matches = pattern.findall(text)
+            assert len(matches) == 0, f"Nom soignant trouvé : {name} ({len(matches)} occurrences)"
+
+    def test_medical_content_preserved(self):
+        text_lower = self.anonymized.lower()
+        for term in ["pancréatite", "cholécystectomie", "morphine", "paracétamol"]:
+            assert term in text_lower, f"Terme médical manquant : {term}"
+
+    def test_diagnostic_principal(self):
+        dp = self.dossier.diagnostic_principal
+        assert dp is not None
+        assert dp.cim10_suggestion in ("K80.5", "K85.1")
+
+    def test_sejour_with_vitals(self):
+        s = self.dossier.sejour
+        assert s.sexe == "F"
+        assert s.age == 43
+        assert s.imc is not None
+        assert s.imc > 30
+        assert s.poids is not None
+        assert s.taille is not None
+
+    def test_biologie(self):
+        tests = {b.test for b in self.dossier.biologie_cle}
+        assert "Lipasémie" in tests
+        assert "CRP" in tests
+        # Lipasémie doit être anormale
+        for b in self.dossier.biologie_cle:
+            if b.test == "Lipasémie":
+                assert b.anomalie is True
+
+    def test_report_counts(self):
+        assert self.report.total_replacements > 100
+        assert self.report.regex_replacements > 50
--- a/tests/test_medical.py
+++ b/tests/test_medical.py
@@ -0,0 +1,238 @@
+"""Tests pour le module d'extraction médicale CIM-10."""
+
+import pytest
+
+from src.config import DossierMedical, Diagnostic
+from src.medical.cim10_extractor import (
+    extract_medical_info,
+    _lookup_cim10,
+    _is_abnormal,
+)
+
+
+class TestCIM10Lookup:
+    def test_pancréatite_lithiasique(self):
+        assert _lookup_cim10("pancréatite aiguë lithiasique") == "K85.1"
+
+    def test_lithiase_choledoque(self):
+        assert _lookup_cim10("lithiase du cholédoque") == "K80.5"
+
+    def test_eruption_medicamenteuse(self):
+        assert _lookup_cim10("éruption médicamenteuse") == "L27.0"
+
+    def test_obesite(self):
+        assert _lookup_cim10("obésité") == "E66.0"
+
+    def test_unknown(self):
+        assert _lookup_cim10("grippe") is None
+
+
+class TestIsAbnormal:
+    def test_lipasemie_high(self):
+        assert _is_abnormal("Lipasémie", "6000") is True
+
+    def test_crp_normal(self):
+        assert _is_abnormal("CRP", "3") is False
+
+    def test_crp_high(self):
+        assert _is_abnormal("CRP", "12") is True
+
+    def test_troponine_negative(self):
+        assert _is_abnormal("Troponine", "négative") is False
+
+    def test_unknown_test(self):
+        assert _is_abnormal("TestInconnu", "42") is None
+
+
+class TestExtractMedicalInfo:
+    def test_extract_from_trackare(self):
+        parsed = {
+            "type": "trackare",
+            "patient": {
+                "sexe": "F",
+                "date_naissance": "23/02/1980",
+                "imc": 34.37,
+                "poids_kg": 90.2,
+                "taille_cm": 162,
+            },
+            "sejour": {
+                "date_entree": "25/02/2023",
+                "date_sortie": "03/03/2023",
+            },
+            "urgences": {"mode_entree": "Urgences"},
+            "diagnostics": [
+                {
+                    "type": "Principal",
+                    "statut": "actif",
+                    "code_cim10": "K80.5",
+                    "libelle": "Calcul des canaux biliaires",
+                }
+            ],
+            "signes_vitaux": {"imc": 34.37, "poids_kg": 90.2, "taille_cm": 162},
+        }
+        text = """Pancréatite aiguë lithiasique.
+Cholécystectomie par cœlioscopie le 01/03.
+Cholangiographie retrouvant une lithiase du bas cholédoque.
+TDM à J3 retrouve : Absence de signe de gravité. Score de Balthazar à 0.
+Éruption cutanée érythémateuse. Réaction au tramadol.
+IMC: 34.370
+TTT de sortie :
+Paracétamol et Acupan si besoin
+Cétirizine
+
+Devenir : sortie le 03/03."""
+
+        dossier = extract_medical_info(parsed, text)
+
+        # Séjour
+        assert dossier.sejour.sexe == "F"
+        assert dossier.sejour.age == 43
+        assert dossier.sejour.duree_sejour == 6
+        assert dossier.sejour.imc == 34.37
+
+        # DP
+        assert dossier.diagnostic_principal is not None
+        assert dossier.diagnostic_principal.cim10_suggestion == "K80.5"
+
+        # DAS
+        codes = {d.cim10_suggestion for d in dossier.diagnostics_associes}
+        assert "L27.0" in codes  # Éruption médicamenteuse
+        assert "E66.0" in codes  # Obésité
+
+        # Actes
+        acte_codes = {a.code_ccam_suggestion for a in dossier.actes_ccam}
+        assert "HMFC004" in acte_codes  # Cholécystectomie
+        assert "ZCQK002" in acte_codes  # TDM
+
+        # Traitements
+        meds = [t.medicament for t in dossier.traitements_sortie]
+        assert any("Paracétamol" in m for m in meds)
+        assert any("Cétirizine" in m for m in meds)
+
+        # Bio
+        tests = {b.test for b in dossier.biologie_cle}
+        assert "Troponine" not in tests  # pas dans ce texte minimal
+
+        # Imagerie
+        assert len(dossier.imagerie) >= 1
+        assert any("Balthazar" in (i.score or "") for i in dossier.imagerie)
+
+        # Complications
+        assert any("cutanée" in c.lower() for c in dossier.complications)
+
+    def test_extract_without_edsnlp(self):
+        """Vérifie que l'extraction fonctionne sans résultat edsnlp."""
+        parsed = {
+            "type": "crh",
+            "patient": {"sexe": "M"},
+            "sejour": {},
+            "diagnostics": [],
+        }
+        text = "Pancréatite aiguë biliaire.\nTTT de sortie :\nParacétamol 1g matin et soir\n\nDevenir : retour."
+
+        dossier = extract_medical_info(parsed, text, edsnlp_result=None)
+        assert dossier.diagnostic_principal is not None
+        assert dossier.diagnostic_principal.cim10_suggestion == "K85.1"
+        assert len(dossier.traitements_sortie) >= 1
+
+    def test_extract_with_edsnlp_result(self):
+        """Vérifie que les résultats edsnlp enrichissent les diagnostics."""
+        from src.medical.edsnlp_pipeline import EdsnlpResult, CIM10Entity, DrugEntity
+
+        parsed = {
+            "type": "crh",
+            "patient": {"sexe": "M"},
+            "sejour": {},
+            "diagnostics": [],
+        }
+        text = "Patient admis pour douleur abdominale."
+
+        edsnlp_result = EdsnlpResult(
+            cim10_entities=[
+                CIM10Entity(texte="douleur abdominale", code="R10.4", negation=False),
+            ],
+            drug_entities=[],
+        )
+
+        dossier = extract_medical_info(parsed, text, edsnlp_result=edsnlp_result)
+        # Le DP devrait être trouvé via edsnlp
+        assert dossier.diagnostic_principal is not None
+        assert dossier.diagnostic_principal.cim10_suggestion == "R10.4"
+
+    def test_negated_edsnlp_entities_ignored(self):
+        """Vérifie que les entités niées par edsnlp ne sont pas retenues."""
+        from src.medical.edsnlp_pipeline import EdsnlpResult, CIM10Entity
+
+        parsed = {
+            "type": "crh",
+            "patient": {"sexe": "M"},
+            "sejour": {},
+            "diagnostics": [],
+        }
+        text = "Pas de fièvre. Patient en bon état."
+
+        edsnlp_result = EdsnlpResult(
+            cim10_entities=[
+                CIM10Entity(texte="fièvre", code="R50.9", negation=True),
+            ],
+        )
+
+        dossier = extract_medical_info(parsed, text, edsnlp_result=edsnlp_result)
+        # L'entité niée ne doit pas apparaître comme diagnostic
+        all_codes = set()
+        if dossier.diagnostic_principal:
+            all_codes.add(dossier.diagnostic_principal.cim10_suggestion)
+        for d in dossier.diagnostics_associes:
+            all_codes.add(d.cim10_suggestion)
+        assert "R50.9" not in all_codes
+
+    def test_drug_atc_enrichment(self):
+        """Vérifie que les codes ATC edsnlp sont ajoutés aux traitements."""
+        from src.medical.edsnlp_pipeline import EdsnlpResult, DrugEntity
+
+        parsed = {
+            "type": "crh",
+            "patient": {"sexe": "M"},
+            "sejour": {},
+            "diagnostics": [],
+        }
+        text = "TTT de sortie :\nParacétamol 1g matin\n\nDevenir : retour."
+
+        edsnlp_result = EdsnlpResult(
+            drug_entities=[
+                DrugEntity(texte="Paracétamol", code_atc="N02BE01", negation=False),
+            ],
+        )
+
+        dossier = extract_medical_info(parsed, text, edsnlp_result=edsnlp_result)
+        assert len(dossier.traitements_sortie) >= 1
+        paracetamol = next(
+            (t for t in dossier.traitements_sortie if "Paracétamol" in t.medicament), None
+        )
+        assert paracetamol is not None
+        assert paracetamol.code_atc == "N02BE01"
+
+    def test_edsnlp_negation_for_complications(self):
+        """Vérifie que la négation edsnlp filtre les complications."""
+        from src.medical.edsnlp_pipeline import EdsnlpResult, CIM10Entity
+
+        parsed = {
+            "type": "crh",
+            "patient": {"sexe": "M"},
+            "sejour": {},
+            "diagnostics": [],
+        }
+        text = "Pas de fièvre ni d'infection. Bonne évolution."
+
+        edsnlp_result = EdsnlpResult(
+            cim10_entities=[
+                CIM10Entity(texte="fièvre", code="R50.9", negation=True),
+                CIM10Entity(texte="infection", code="A49.9", negation=True),
+            ],
+        )
+
+        dossier = extract_medical_info(parsed, text, edsnlp_result=edsnlp_result)
+        # Fièvre et infection sont niées, ne doivent pas apparaître dans complications
+        complication_terms = [c.lower() for c in dossier.complications]
+        assert "fièvre" not in complication_terms
+        assert "infection" not in complication_terms