"""Asociación automática heurística de documentos a paciente/caso.""" from __future__ import annotations from difflib import SequenceMatcher import re from typing import Any from app.batch_processing.domain.models import ( ASSOCIATION_SOURCE_AUTO, AssociationDecision, AssociationResult, ExtractedSignals, FILE_STATUS_ASOCIADO, FILE_STATUS_PENDIENTE_VALIDACION, ) from app.services.patient_name_extraction import ( extract_patient_name_from_text, sanitize_patient_name_candidate, ) _DATE_PATTERN = re.compile(r"\b(\d{1,2}[/-]\d{1,2}[/-]\d{2,4})\b") _DOCUMENT_NUMBER_PATTERN = re.compile( r"(?:^|[\s:.-])(?:cc|ti|ce|dni|nit)?\s*[-:]?\s*(\d{5,12})(?:\b|$)", re.IGNORECASE, ) _CASE_INLINE_PATTERNS = ( re.compile( r"(?:caso\s*(?:no\.?)?|no\.?\s*de\s*caso|n[°o]\s*caso)\s*[:#-]?\s*(\d{4,12})", re.IGNORECASE, ), re.compile( r"\b(\d{4,12})\s*(?:no\.?\s*de\s*caso|n[°o]\s*caso|caso\s*(?:no\.?)?|caso)\b", re.IGNORECASE, ), ) _CASE_LABEL_PATTERN = re.compile( r"(?:caso\s*(?:no\.?)?|no\.?\s*de\s*caso|n[°o]\s*caso)$", re.IGNORECASE, ) _CASE_HEADER_LABEL_PATTERN = re.compile( r"(?:caso\s*(?:no\.?)?|no\.?\s*de\s*caso|n[°o]\s*caso|caso)\b", re.IGNORECASE, ) _PROCEDURE_LINE_PATTERN = re.compile( r"(?:procedimiento|cirug[ií]a|intervenci[oó]n)\s*[:#]?\s*([^\n\r]{5,120})", re.IGNORECASE, ) _PATIENT_ID_LABEL_TOKENS = ( "identificación", "identificacion", "tipo y n", "tipo y no", "tipo y nro", "documento", "cc", "cédula", "cedula", ) _PROCEDURE_DESCRIPTION_LABELS = ( "procedimientos realizados", "descripción del procedimiento", "procedimiento realizado", ) AUTO_ASSOCIATION_MIN_SCORE = 0.85 AUTO_ASSOCIATION_MIN_GAP = 0.20 REQUIRED_ANCHOR_SIGNALS = {"identificacion", "numero_caso"} _FILENAME_IGNORE_TOKENS = { "historia", "clinica", "clínica", "factura", "cirugia", "cirugía", "quirurgico", "quirúrgico", "quirurgica", "quirúrgica", "documento", "doc", "archivo", "tipo", "egreso", "fecha", "telefono", "teléfono", "paciente", "usuario", "caso", "ingreso", "hc", } def _normalize_text(value: str) -> str: value = str(value or "").strip().lower() value = re.sub(r"\s+", " ", value) return value def _normalize_name(value: str) -> str: normalized = _normalize_text(value) normalized = re.sub(r"[^a-z0-9áéíóúñ ]+", " ", normalized) normalized = re.sub(r"\s+", " ", normalized).strip() return normalized def _slug(value: str) -> str: normalized = _normalize_name(value) return re.sub(r"[^a-z0-9]+", "-", normalized).strip("-") class HeuristicCaseAssociationService: """ Extrae señales mínimas por documento y construye clústeres de caso por score. La heurística prioriza identificación y número de caso; el nombre del paciente y datos secundarios solo refinan el resultado. """ def extract_signals(self, filename: str, text: str, detected_type: str) -> ExtractedSignals: evidence: list[str] = [] date_match = _DATE_PATTERN.search(text) patient_name = self._pick_patient_name(filename, text) if patient_name: evidence.append("nombre_paciente") patient_id = self._extract_patient_id(text) if patient_id: evidence.append("identificacion") case_number = self._extract_case_number(text) if case_number: evidence.append("numero_caso") service_date = date_match.group(1) if date_match else "" if service_date: evidence.append("fecha_atencion") procedure_code = "" procedure_description = self._extract_procedure_description(text) if procedure_description: evidence.append("descripcion_procedimiento") if filename: evidence.append("nombre_archivo") return ExtractedSignals( patient_name=patient_name, patient_id=patient_id, case_number=case_number, service_date=service_date, procedure_code=procedure_code, procedure_description=procedure_description, evidence=evidence, ) def associate(self, file_records: list[dict[str, Any]]) -> AssociationResult: """Asocia documentos a clústeres de caso o los deja pendientes de validación.""" clusters: list[dict[str, Any]] = [] decisions: list[AssociationDecision] = [] ordered_files = sorted( file_records, key=lambda item: self._signal_strength(item), reverse=True, ) for file_record in ordered_files: candidate_scores: list[dict[str, Any]] = [] for cluster in clusters: score, evidence, breakdown = self._score(file_record, cluster) candidate_scores.append( { "cluster": cluster, "score": score, "evidence": evidence, "score_breakdown": breakdown, } ) candidate_scores.sort(key=lambda item: item["score"], reverse=True) best_candidate = candidate_scores[0] if candidate_scores else None best_cluster = best_candidate["cluster"] if best_candidate else None best_score = float(best_candidate["score"]) if best_candidate else 0.0 second_score = ( float(candidate_scores[1]["score"]) if len(candidate_scores) > 1 else 0.0 ) best_evidence = list(best_candidate["evidence"]) if best_candidate else [] best_breakdown = ( dict(best_candidate["score_breakdown"]) if best_candidate else {} ) top_candidates = [ self._candidate_payload(item) for item in candidate_scores[:2] ] if ( best_cluster and self._can_auto_associate( best_score=best_score, second_score=second_score, evidence=best_evidence, ) ): self._merge_into_cluster(best_cluster, file_record) decision = AssociationDecision( file_id=file_record["_id"], status=FILE_STATUS_ASOCIADO, case_key=best_cluster["case_key"], confidence=self._clamp_score(best_score), evidence=best_evidence, patient_name=file_record.get("patient_name", "") or best_cluster.get("patient_name", ""), patient_id=file_record.get("patient_id", "") or best_cluster.get("patient_id", ""), case_number=file_record.get("case_number", "") or best_cluster.get("case_number", ""), procedure_code=file_record.get("procedure_code", "") or best_cluster.get("procedure_code", ""), procedure_description=file_record.get("procedure_description", "") or best_cluster.get("procedure_description", ""), associated_user=file_record.get("associated_user", "") or best_cluster.get("associated_user", ""), association_source=ASSOCIATION_SOURCE_AUTO, score_breakdown=best_breakdown, top_candidates=top_candidates, ) decisions.append(decision) best_cluster["file_ids"].append(file_record["_id"]) continue if self._should_seed_cluster( file_record=file_record, best_cluster=best_cluster, evidence=best_evidence, ): cluster = self._build_cluster(file_record) seed_breakdown = self._seed_score_breakdown(file_record) seed_score = self._score_from_breakdown(seed_breakdown) seed_evidence = self._seed_evidence(file_record) clusters.append(cluster) decisions.append( AssociationDecision( file_id=file_record["_id"], status=FILE_STATUS_ASOCIADO, case_key=cluster["case_key"], confidence=seed_score, evidence=seed_evidence, patient_name=file_record.get("patient_name", ""), patient_id=file_record.get("patient_id", ""), case_number=file_record.get("case_number", ""), procedure_code=file_record.get("procedure_code", ""), procedure_description=file_record.get( "procedure_description", "" ), associated_user=self._build_associated_user( file_record.get("patient_id", ""), file_record.get("patient_name", ""), ), association_source=ASSOCIATION_SOURCE_AUTO, score_breakdown=seed_breakdown, top_candidates=[ { "case_key": cluster["case_key"], "score": seed_score, "evidence": seed_evidence, "score_breakdown": seed_breakdown, } ], ) ) cluster["file_ids"].append(file_record["_id"]) continue decisions.append( AssociationDecision( file_id=file_record["_id"], status=FILE_STATUS_PENDIENTE_VALIDACION, confidence=self._clamp_score(best_score), evidence=best_evidence or file_record.get("evidence", []) or ["nombre_archivo"], patient_name=file_record.get("patient_name", ""), patient_id=file_record.get("patient_id", ""), case_number=file_record.get("case_number", ""), procedure_code=file_record.get("procedure_code", ""), procedure_description=file_record.get("procedure_description", ""), associated_user=self._build_associated_user( file_record.get("patient_id", ""), file_record.get("patient_name", ""), ), association_source=ASSOCIATION_SOURCE_AUTO, score_breakdown=best_breakdown, top_candidates=top_candidates, ) ) cases = [] for cluster in clusters: cases.append( { "batch_id": cluster["batch_id"], "case_key": cluster["case_key"], "patient_name": cluster.get("patient_name", ""), "patient_id": cluster.get("patient_id", ""), "case_number": cluster.get("case_number", ""), "service_date": cluster.get("service_date", ""), "procedure_code": cluster.get("procedure_code", ""), "procedure_description": cluster.get( "procedure_description", "" ), "associated_user": cluster.get("associated_user", ""), "file_ids": cluster.get("file_ids", []), } ) return AssociationResult(decisions=decisions, cases=cases) def _pick_patient_name(self, filename: str, text: str) -> str: extracted_name = extract_patient_name_from_text(text) sanitized_extracted = sanitize_patient_name_candidate(extracted_name) if sanitized_extracted: return sanitized_extracted from_filename = self._name_candidate_from_filename(filename) if from_filename: return from_filename return "" def _signal_strength(self, file_record: dict[str, Any]) -> int: return sum( 1 for key in ( "patient_id", "case_number", "patient_name", "service_date", ) if file_record.get(key) ) def _can_seed_cluster(self, file_record: dict[str, Any]) -> bool: return bool(file_record.get("patient_id") or file_record.get("case_number")) def _seed_evidence(self, file_record: dict[str, Any]) -> list[str]: evidence = [] if file_record.get("patient_id"): evidence.append("identificacion") if file_record.get("case_number"): evidence.append("numero_caso") if file_record.get("patient_name"): evidence.append("nombre_paciente") return evidence or ["nombre_archivo"] def _build_cluster(self, file_record: dict[str, Any]) -> dict[str, Any]: case_key = self._build_case_key(file_record) return { "batch_id": file_record["batch_id"], "case_key": case_key, "patient_name": file_record.get("patient_name", ""), "patient_id": file_record.get("patient_id", ""), "case_number": file_record.get("case_number", ""), "service_date": file_record.get("service_date", ""), "procedure_code": "", "procedure_description": file_record.get("procedure_description", ""), "associated_user": self._build_associated_user( file_record.get("patient_id", ""), file_record.get("patient_name", ""), ), "file_ids": [], } def _merge_into_cluster( self, cluster: dict[str, Any], file_record: dict[str, Any], ) -> None: for key in ( "patient_name", "patient_id", "case_number", "service_date", "procedure_description", ): if not cluster.get(key) and file_record.get(key): cluster[key] = file_record.get(key) if not cluster.get("associated_user"): cluster["associated_user"] = self._build_associated_user( cluster.get("patient_id", "") or file_record.get("patient_id", ""), cluster.get("patient_name", "") or file_record.get("patient_name", ""), ) def _build_case_key(self, file_record: dict[str, Any]) -> str: parts = [ _slug(file_record.get("patient_id", "")), _slug(file_record.get("case_number", "")), _slug(file_record.get("patient_name", "")), ] parts = [part for part in parts if part] return "-".join(parts[:3]) or _slug(file_record.get("original_name", "archivo")) def _score( self, file_record: dict[str, Any], cluster: dict[str, Any], ) -> tuple[float, list[str], dict[str, float]]: """Calcula el score acordado para decidir asociación automática segura.""" evidence: list[str] = [] breakdown: dict[str, float] = {} if ( file_record.get("patient_id") and cluster.get("patient_id") and _normalize_text(file_record["patient_id"]) == _normalize_text(cluster["patient_id"]) ): evidence.append("identificacion") breakdown["identificacion"] = 0.55 if ( file_record.get("case_number") and cluster.get("case_number") and _normalize_text(file_record["case_number"]) == _normalize_text(cluster["case_number"]) ): evidence.append("numero_caso") breakdown["numero_caso"] = 0.25 name_ratio = self._name_similarity( file_record.get("patient_name", ""), cluster.get("patient_name", ""), ) if name_ratio >= 0.92: evidence.append("nombre_paciente") breakdown["nombre_paciente"] = 0.20 if ( file_record.get("service_date") and cluster.get("service_date") and file_record["service_date"] == cluster["service_date"] ): evidence.append("fecha_atencion") breakdown["fecha_atencion"] = 0.05 normalized_breakdown = self._normalize_breakdown(breakdown) return self._score_from_breakdown(normalized_breakdown), evidence, normalized_breakdown def _seed_score_breakdown(self, file_record: dict[str, Any]) -> dict[str, float]: breakdown: dict[str, float] = {} if file_record.get("patient_id"): breakdown["identificacion"] = 0.55 if file_record.get("case_number"): breakdown["numero_caso"] = 0.25 if file_record.get("patient_name"): breakdown["nombre_paciente"] = 0.20 return self._normalize_breakdown(breakdown) def _candidate_payload(self, candidate: dict[str, Any]) -> dict[str, Any]: cluster = candidate["cluster"] return { "case_key": cluster.get("case_key", ""), "score": self._clamp_score(candidate.get("score", 0.0)), "evidence": list(candidate.get("evidence", [])), "score_breakdown": dict(candidate.get("score_breakdown", {})), "patient_name": cluster.get("patient_name", ""), "patient_id": cluster.get("patient_id", ""), "procedure_code": "", } def _build_associated_user(self, patient_id: str, patient_name: str) -> str: if patient_id: return str(patient_id).strip() if patient_name: return re.sub(r"\s+", " ", str(patient_name).strip()) return "" def _name_similarity(self, left: str, right: str) -> float: left_norm = _normalize_name(left) right_norm = _normalize_name(right) if not left_norm or not right_norm: return 0.0 if left_norm == right_norm: return 1.0 return SequenceMatcher(a=left_norm, b=right_norm).ratio() def _extract_patient_id(self, text: str) -> str: lines = self._clean_lines(text) for index, line in enumerate(lines): line_lower = line.lower() if "paciente:" in line_lower: candidate = self._extract_document_number_from_fragment(line) if candidate: return candidate if not self._line_has_patient_id_label(line_lower): continue candidate = self._extract_document_number_from_fragment(line) if candidate: return candidate for probe in range(index + 1, min(index + 4, len(lines))): next_line = lines[probe] if self._line_has_phone_context(next_line): continue candidate = self._extract_document_number_from_fragment(next_line) if candidate: return candidate return "" def _extract_case_number(self, text: str) -> str: lines = self._clean_lines(text) for line in lines[:12]: candidate = self._extract_case_number_from_fragment(line) if candidate: return candidate for pattern in _CASE_INLINE_PATTERNS: match = pattern.search(text) if match: return match.group(1).strip() for index, line in enumerate(lines): if _CASE_LABEL_PATTERN.search(line): for probe in range(index + 1, min(index + 3, len(lines))): number_match = re.search(r"\b(\d{4,12})\b", lines[probe]) if number_match: return number_match.group(1) return "" def _extract_procedure_description(self, text: str) -> str: lines = self._clean_lines(text) for index, line in enumerate(lines): line_lower = line.lower() if any(label in line_lower for label in _PROCEDURE_DESCRIPTION_LABELS): inline_match = _PROCEDURE_LINE_PATTERN.search(line) if inline_match and ":" in line: candidate = self._sanitize_procedure_description( inline_match.group(1) ) if candidate: return candidate for probe in range(index + 1, min(index + 4, len(lines))): candidate = self._sanitize_procedure_description(lines[probe]) if candidate: return candidate return "" def _name_candidate_from_filename(self, filename: str) -> str: basename = re.sub(r"\.pdf$", "", str(filename or ""), flags=re.IGNORECASE) tokens = re.findall(r"[A-Za-zÁÉÍÓÚÑáéíóúñ]+", basename) cleaned_tokens = [] for token in tokens: normalized = _normalize_name(token) if len(normalized) < 2: continue if normalized in _FILENAME_IGNORE_TOKENS: continue cleaned_tokens.append(token) if len(cleaned_tokens) < 2: return "" candidate = " ".join(cleaned_tokens[:4]) return sanitize_patient_name_candidate(candidate) def _can_auto_associate( self, *, best_score: float, second_score: float, evidence: list[str], ) -> bool: score = self._clamp_score(best_score) gap = score - self._clamp_score(second_score) evidence_set = set(evidence) has_anchor_signal = any(signal in REQUIRED_ANCHOR_SIGNALS for signal in evidence_set) has_required_anchor_match = REQUIRED_ANCHOR_SIGNALS.issubset(evidence_set) if has_required_anchor_match and gap >= AUTO_ASSOCIATION_MIN_GAP: return True return ( has_anchor_signal and score >= AUTO_ASSOCIATION_MIN_SCORE and gap >= AUTO_ASSOCIATION_MIN_GAP ) def _should_seed_cluster( self, *, file_record: dict[str, Any], best_cluster: dict[str, Any] | None, evidence: list[str], ) -> bool: if not self._can_seed_cluster(file_record): return False if not best_cluster: return True anchor_evidence = { signal for signal in evidence if signal in REQUIRED_ANCHOR_SIGNALS } return not anchor_evidence def _normalize_breakdown(self, breakdown: dict[str, float]) -> dict[str, float]: positive = { key: float(value) for key, value in breakdown.items() if float(value) > 0.0 } if not positive: return {} total = sum(positive.values()) if total <= 1.0: return {key: round(value, 4) for key, value in positive.items()} factor = 1.0 / total normalized = { key: round(value * factor, 4) for key, value in positive.items() } normalized_total = sum(normalized.values()) if normalized_total > 1.0: first_key = next(iter(normalized)) normalized[first_key] = round( max(0.0, normalized[first_key] - (normalized_total - 1.0)), 4, ) return normalized def _score_from_breakdown(self, breakdown: dict[str, float]) -> float: return self._clamp_score(sum(breakdown.values())) def _clamp_score(self, value: float | int) -> float: try: numeric = float(value) except (TypeError, ValueError): return 0.0 if numeric < 0.0: return 0.0 if numeric > 1.0: return 1.0 return round(numeric, 4) def _clean_lines(self, text: str) -> list[str]: return [ re.sub(r"\s+", " ", line).strip() for line in text.splitlines() if re.sub(r"\s+", " ", line).strip() ] def _extract_document_number_from_fragment(self, fragment: str) -> str: if self._line_has_phone_context(fragment): return "" explicit_match = re.search( r"(?:cc|ti|ce|dni|nit)\s*[-:]?\s*(\d{5,12})", fragment, re.IGNORECASE, ) if explicit_match: return explicit_match.group(1).strip() label_match = re.search( r"(?:identificaci[oó]n|documento)\D*(\d{5,12})", fragment, re.IGNORECASE, ) if label_match: return label_match.group(1).strip() match = _DOCUMENT_NUMBER_PATTERN.search(fragment) if not match: return "" return match.group(1).strip() def _extract_case_number_from_fragment(self, fragment: str) -> str: normalized = re.sub(r"\s+", " ", str(fragment or "")).strip() if not normalized: return "" if not _CASE_HEADER_LABEL_PATTERN.search(normalized): return "" for pattern in _CASE_INLINE_PATTERNS: match = pattern.search(normalized) if match: return match.group(1).strip() return "" def _line_has_patient_id_label(self, line_lower: str) -> bool: if any(token in line_lower for token in ("identificación", "identificacion", "documento")): return True return bool(re.search(r"\bcc\b|\bc[ée]dula\b", line_lower)) def _line_has_phone_context(self, value: str) -> bool: lowered = value.lower() return any( token in lowered for token in ("tel", "telefono", "teléfono", "cel", "fax", "contacto") ) def _sanitize_procedure_description(self, value: str) -> str: candidate = re.sub(r"\s+", " ", str(value or "")).strip(" :-") if len(candidate) < 8: return "" lowered = candidate.lower() if lowered in {label.lower() for label in _PROCEDURE_DESCRIPTION_LABELS}: return "" if any( token in lowered for token in ("paciente quien", "estado salida", "registro de muestras") ): return "" return candidate