@precisa-saude/fhir-ocr-utils 0.1.0

package/README.md

@@ -0,0 +1,85 @@
+# @precisa-saude/fhir-ocr-utils
+
+Utilitários de ancoragem OCR para extração de biomarcadores de PDFs de resultados laboratoriais.
+
+## Instalação
+
+```bash
+npm install @precisa-saude/fhir-ocr-utils
+```
+
+> **Nota:** Requer `@precisa-saude/fhir` como peer dependency.
+
+## Uso rápido
+
+### Encontrar biomarcadores em texto OCR
+
+```ts
+import { findBiomarkersInText, getMatchedCodes } from '@precisa-saude/fhir-ocr-utils';
+
+const ocrText = `
+  HEMOGRAMA COMPLETO
+  Hemoglobina: 14.2 g/dL
+  Glicose Jejum: 95 mg/dL
+  Colesterol Total: 195 mg/dL
+  HDL: 55 mg/dL
+  Triglicerídeos: 120 mg/dL
+`;
+
+const result = findBiomarkersInText(ocrText);
+
+console.log(result.matches);
+// [
+//   { code: 'Hgb', loinc: '718-7', matchedName: 'Hemoglobina', ... },
+//   { code: 'Glucose', loinc: '2345-7', matchedName: 'Glicose', ... },
+//   { code: 'Cholesterol', loinc: '2093-3', matchedName: 'Colesterol Total', ... },
+//   ...
+// ]
+
+const codes = getMatchedCodes(result);
+// ['Hgb', 'Glucose', 'Cholesterol', 'HDL', 'Triglycerides']
+
+console.log(result.filteredReference);
+// Referência LLM filtrada apenas com os biomarcadores encontrados
+```
+
+## Padrão anti-alucinação
+
+Este pacote implementa um padrão de **ancoragem antes do LLM** para prevenir alucinações na extração de dados laboratoriais:
+
+1. **Ancoragem (este pacote):** Escaneia o texto OCR bruto procurando nomes de biomarcadores conhecidos usando correspondência exata de strings contra as 183+ definições de `@precisa-saude/fhir`.
+
+2. **Filtragem:** Gera uma referência LLM filtrada (`filteredReference`) contendo apenas os biomarcadores que foram realmente encontrados no texto. O LLM só pode extrair valores para biomarcadores presentes nesta lista.
+
+3. **Extração (LLM):** O modelo de linguagem recebe o texto OCR junto com a referência filtrada, restringindo sua saída apenas aos biomarcadores ancorados.
+
+Este fluxo em dois estágios garante que o LLM não invente biomarcadores que não estão presentes no documento original.
+
+```
+PDF → OCR → findBiomarkersInText() → filteredReference → LLM → valores extraídos
+                    |                                                      |
+                    +── restringe quais biomarcadores ─────────────────────+
+                        o LLM pode extrair
+```
+
+## API
+
+### `findBiomarkersInText(ocrText: string): AnchorResult`
+
+Escaneia texto OCR e retorna todos os biomarcadores encontrados.
+
+- `result.matches` — Lista de biomarcadores encontrados com código, LOINC, nome e posição
+- `result.filteredReference` — Referência formatada para enviar ao LLM
+- `result.stats` — Estatísticas de execução (total de padrões, encontrados, tempo)
+
+### `getMatchedCodes(result: AnchorResult): string[]`
+
+Extrai a lista de códigos de biomarcadores de um resultado de ancoragem.
+
+## Aviso médico
+
+Este pacote fornece utilitários de processamento de texto para extração de dados laboratoriais. **Não substitui orientação médica profissional.** Consulte o [DISCLAIMER.md](../../DISCLAIMER.md) na raiz do repositório para detalhes completos.
+
+## Licença
+
+[Apache-2.0](../../LICENSE)

package/dist/index.cjs

@@ -0,0 +1,126 @@
+"use strict";Object.defineProperty(exports, "__esModule", {value: true});// src/anchor.ts
+
+
+
+var _fhir = require('@precisa-saude/fhir');
+function normalize(text) {
+  return text.normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase().replace(/\s+/g, " ");
+}
+var UNAMBIGUOUS_SHORT_NAMES = /* @__PURE__ */ new Set([
+  "hdl",
+  "ldl",
+  "lh",
+  "tsh",
+  "crp",
+  "pcr",
+  "ggt",
+  "alt",
+  "ast",
+  "bun",
+  "wbc",
+  "rbc",
+  "mcv",
+  "mch",
+  "rdw",
+  "mpv",
+  "psa",
+  "fsh",
+  "hba1c",
+  "egfr",
+  "acr",
+  "esr",
+  "vhs",
+  "bmc",
+  "bmd",
+  "vat",
+  "dxa",
+  "dmo",
+  "cmo",
+  "ffm",
+  "lbm",
+  "mlg",
+  "tav"
+]);
+var cachedPatterns = null;
+var cachedNormalized = null;
+function getPatterns() {
+  if (!cachedPatterns) {
+    cachedPatterns = _fhir.getAllSearchPatterns.call(void 0, );
+  }
+  return cachedPatterns;
+}
+function getNormalizedPatterns() {
+  if (!cachedNormalized) {
+    const patterns = getPatterns();
+    const map = /* @__PURE__ */ new Map();
+    for (const pattern of patterns) {
+      for (const name of pattern.names) {
+        const normalized = normalize(name);
+        const existing = map.get(normalized) || [];
+        existing.push({
+          code: pattern.code,
+          ...pattern.loinc && { loinc: pattern.loinc },
+          original: name
+        });
+        map.set(normalized, existing);
+      }
+    }
+    cachedNormalized = map;
+  }
+  return cachedNormalized;
+}
+function findBiomarkersInText(ocrText) {
+  const startTime = Date.now();
+  const normalizedText = normalize(ocrText);
+  const matchedCodes = /* @__PURE__ */ new Set();
+  const matches = [];
+  const normalizedPatterns = getNormalizedPatterns();
+  for (const [normalizedName, entries] of normalizedPatterns) {
+    if (normalizedName.length < 3 && !UNAMBIGUOUS_SHORT_NAMES.has(normalizedName)) {
+      continue;
+    }
+    let position = -1;
+    if (normalizedName.length <= 4) {
+      const regex = new RegExp(`\\b${normalizedName}\\b`);
+      const match = regex.exec(normalizedText);
+      if (match) {
+        position = match.index;
+      }
+    } else {
+      position = normalizedText.indexOf(normalizedName);
+    }
+    if (position !== -1) {
+      for (const entry of entries) {
+        if (!matchedCodes.has(entry.code)) {
+          matchedCodes.add(entry.code);
+          matches.push({
+            code: entry.code,
+            confidence: 1,
+            loinc: entry.loinc,
+            matchedName: entry.original,
+            position
+          });
+        }
+      }
+    }
+  }
+  const scanTimeMs = Date.now() - startTime;
+  const matchedCodesArray = Array.from(matchedCodes);
+  return {
+    filteredReference: _fhir.generateFilteredLLMReference.call(void 0, matchedCodesArray),
+    matches,
+    stats: {
+      matchedCount: matches.length,
+      scanTimeMs,
+      totalPatterns: getPatterns().length
+    }
+  };
+}
+function getMatchedCodes(result) {
+  return result.matches.map((m) => m.code);
+}
+
+
+
+exports.findBiomarkersInText = findBiomarkersInText; exports.getMatchedCodes = getMatchedCodes;
+//# sourceMappingURL=index.cjs.map

package/dist/index.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/Users/rafael/Github/precisa-saude/fhir-brasil/packages/ocr-utils/dist/index.cjs","../src/anchor.ts"],"names":[],"mappings":"AAAA;ACQA;AAEE;AACA;AAAA,2CACK;AA0BP,SAAS,SAAA,CAAU,IAAA,EAAsB;AACvC,EAAA,OAAO,IAAA,CACJ,SAAA,CAAU,KAAK,CAAA,CACf,OAAA,CAAQ,kBAAA,EAAoB,EAAE,CAAA,CAC9B,WAAA,CAAY,CAAA,CACZ,OAAA,CAAQ,MAAA,EAAQ,GAAG,CAAA;AACxB;AAEA,IAAM,wBAAA,kBAA0B,IAAI,GAAA,CAAI;AAAA,EACtC,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,IAAA;AAAA,EAAM,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAC9D,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,OAAA;AAAA,EAAS,MAAA;AAAA,EACjE,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO,KAAA;AAAA,EAC/D,KAAA;AAAA,EAAO,KAAA;AAAA,EAAO;AAChB,CAAC,CAAA;AAID,IAAI,eAAA,EAAkD,IAAA;AACtD,IAAI,iBAAA,EAAuD,IAAA;AAE3D,SAAS,WAAA,CAAA,EAAwC;AAC/C,EAAA,GAAA,CAAI,CAAC,cAAA,EAAgB;AACnB,IAAA,eAAA,EAAiB,wCAAA,CAAqB;AAAA,EACxC;AACA,EAAA,OAAO,cAAA;AACT;AAEA,SAAS,qBAAA,CAAA,EAAqD;AAC5D,EAAA,GAAA,CAAI,CAAC,gBAAA,EAAkB;AACrB,IAAA,MAAM,SAAA,EAAW,WAAA,CAAY,CAAA;AAC7B,IAAA,MAAM,IAAA,kBAAM,IAAI,GAAA,CAA4B,CAAA;AAC5C,IAAA,IAAA,CAAA,MAAW,QAAA,GAAW,QAAA,EAAU;AAC9B,MAAA,IAAA,CAAA,MAAW,KAAA,GAAQ,OAAA,CAAQ,KAAA,EAAO;AAChC,QAAA,MAAM,WAAA,EAAa,SAAA,CAAU,IAAI,CAAA;AACjC,QAAA,MAAM,SAAA,EAAW,GAAA,CAAI,GAAA,CAAI,UAAU,EAAA,GAAK,CAAC,CAAA;AACzC,QAAA,QAAA,CAAS,IAAA,CAAK;AAAA,UACZ,IAAA,EAAM,OAAA,CAAQ,IAAA;AAAA,UACd,GAAI,OAAA,CAAQ,MAAA,GAAS,EAAE,KAAA,EAAO,OAAA,CAAQ,MAAM,CAAA;AAAA,UAC5C,QAAA,EAAU;AAAA,QACZ,CAAC,CAAA;AACD,QAAA,GAAA,CAAI,GAAA,CAAI,UAAA,EAAY,QAAQ,CAAA;AAAA,MAC9B;AAAA,IACF;AACA,IAAA,iBAAA,EAAmB,GAAA;AAAA,EACrB;AACA,EAAA,OAAO,gBAAA;AACT;AAOO,SAAS,oBAAA,CAAqB,OAAA,EAA+B;AAClE,EAAA,MAAM,UAAA,EAAY,IAAA,CAAK,GAAA,CAAI,CAAA;AAC3B,EAAA,MAAM,eAAA,EAAiB,SAAA,CAAU,OAAO,CAAA;AACxC,EAAA,MAAM,aAAA,kBAAe,IAAI,GAAA,CAAY,CAAA;AACrC,EAAA,MAAM,QAAA,EAAyB,CAAC,CAAA;AAChC,EAAA,MAAM,mBAAA,EAAqB,qBAAA,CAAsB,CAAA;AAEjD,EAAA,IAAA,CAAA,MAAW,CAAC,cAAA,EAAgB,OAAO,EAAA,GAAK,kBAAA,EAAoB;AAC1D,IAAA,GAAA,CAAI,cAAA,CAAe,OAAA,EAAS,EAAA,GAAK,CAAC,uBAAA,CAAwB,GAAA,CAAI,cAAc,CAAA,EAAG;AAC7E,MAAA,QAAA;AAAA,IACF;AAEA,IAAA,IAAI,SAAA,EAAW,CAAA,CAAA;AACf,IAAA,GAAA,CAAI,cAAA,CAAe,OAAA,GAAU,CAAA,EAAG;AAC9B,MAAA,MAAM,MAAA,EAAQ,IAAI,MAAA,CAAO,CAAA,GAAA,EAAM,cAAc,CAAA,GAAA,CAAK,CAAA;AAClD,MAAA,MAAM,MAAA,EAAQ,KAAA,CAAM,IAAA,CAAK,cAAc,CAAA;AACvC,MAAA,GAAA,CAAI,KAAA,EAAO;AACT,QAAA,SAAA,EAAW,KAAA,CAAM,KAAA;AAAA,MACnB;AAAA,IACF,EAAA,KAAO;AACL,MAAA,SAAA,EAAW,cAAA,CAAe,OAAA,CAAQ,cAAc,CAAA;AAAA,IAClD;AAEA,IAAA,GAAA,CAAI,SAAA,IAAa,CAAA,CAAA,EAAI;AACnB,MAAA,IAAA,CAAA,MAAW,MAAA,GAAS,OAAA,EAAS;AAC3B,QAAA,GAAA,CAAI,CAAC,YAAA,CAAa,GAAA,CAAI,KAAA,CAAM,IAAI,CAAA,EAAG;AACjC,UAAA,YAAA,CAAa,GAAA,CAAI,KAAA,CAAM,IAAI,CAAA;AAC3B,UAAA,OAAA,CAAQ,IAAA,CAAK;AAAA,YACX,IAAA,EAAM,KAAA,CAAM,IAAA;AAAA,YACZ,UAAA,EAAY,CAAA;AAAA,YACZ,KAAA,EAAO,KAAA,CAAM,KAAA;AAAA,YACb,WAAA,EAAa,KAAA,CAAM,QAAA;AAAA,YACnB;AAAA,UACF,CAAC,CAAA;AAAA,QACH;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,EAAA,MAAM,WAAA,EAAa,IAAA,CAAK,GAAA,CAAI,EAAA,EAAI,SAAA;AAChC,EAAA,MAAM,kBAAA,EAAoB,KAAA,CAAM,IAAA,CAAK,YAAY,CAAA;AAEjD,EAAA,OAAO;AAAA,IACL,iBAAA,EAAmB,gDAAA,iBAA8C,CAAA;AAAA,IACjE,OAAA;AAAA,IACA,KAAA,EAAO;AAAA,MACL,YAAA,EAAc,OAAA,CAAQ,MAAA;AAAA,MACtB,UAAA;AAAA,MACA,aAAA,EAAe,WAAA,CAAY,CAAA,CAAE;AAAA,IAC/B;AAAA,EACF,CAAA;AACF;AAKO,SAAS,eAAA,CAAgB,MAAA,EAAgC;AAC9D,EAAA,OAAO,MAAA,CAAO,OAAA,CAAQ,GAAA,CAAI,CAAC,CAAA,EAAA,GAAM,CAAA,CAAE,IAAI,CAAA;AACzC;AD5BA;AACE;AACA;AACF,+FAAC","file":"/Users/rafael/Github/precisa-saude/fhir-brasil/packages/ocr-utils/dist/index.cjs","sourcesContent":[null,"/**\n * OCR Anchor — Biomarker text anchoring\n *\n * Scans OCR text for biomarker names BEFORE sending to LLM.\n * This prevents hallucination by constraining what biomarkers\n * the LLM is allowed to extract.\n */\n\nimport {\n  type BiomarkerSearchPattern,\n  generateFilteredLLMReference,\n  getAllSearchPatterns,\n} from '@precisa-saude/fhir';\n\nexport interface AnchorMatch {\n  code: string;\n  confidence: number;\n  loinc?: string;\n  matchedName: string;\n  position: number;\n}\n\nexport interface AnchorResult {\n  filteredReference: string;\n  matches: AnchorMatch[];\n  stats: {\n    totalPatterns: number;\n    matchedCount: number;\n    scanTimeMs: number;\n  };\n}\n\n/**\n * Normalize text for comparison:\n * - Removes diacritics (ã→a, ç→c, é→e)\n * - Converts to lowercase\n * - Normalizes whitespace\n */\nfunction normalize(text: string): string {\n  return text\n    .normalize('NFD')\n    .replace(/[\\u0300-\\u036f]/g, '')\n    .toLowerCase()\n    .replace(/\\s+/g, ' ');\n}\n\nconst UNAMBIGUOUS_SHORT_NAMES = new Set([\n  'hdl', 'ldl', 'lh', 'tsh', 'crp', 'pcr', 'ggt', 'alt', 'ast', 'bun',\n  'wbc', 'rbc', 'mcv', 'mch', 'rdw', 'mpv', 'psa', 'fsh', 'hba1c', 'egfr',\n  'acr', 'esr', 'vhs', 'bmc', 'bmd', 'vat', 'dxa', 'dmo', 'cmo', 'ffm',\n  'lbm', 'mlg', 'tav',\n]);\n\ntype PatternEntry = { original: string; code: string; loinc?: string };\n\nlet cachedPatterns: BiomarkerSearchPattern[] | null = null;\nlet cachedNormalized: Map<string, PatternEntry[]> | null = null;\n\nfunction getPatterns(): BiomarkerSearchPattern[] {\n  if (!cachedPatterns) {\n    cachedPatterns = getAllSearchPatterns();\n  }\n  return cachedPatterns;\n}\n\nfunction getNormalizedPatterns(): Map<string, PatternEntry[]> {\n  if (!cachedNormalized) {\n    const patterns = getPatterns();\n    const map = new Map<string, PatternEntry[]>();\n    for (const pattern of patterns) {\n      for (const name of pattern.names) {\n        const normalized = normalize(name);\n        const existing = map.get(normalized) || [];\n        existing.push({\n          code: pattern.code,\n          ...(pattern.loinc && { loinc: pattern.loinc }),\n          original: name,\n        });\n        map.set(normalized, existing);\n      }\n    }\n    cachedNormalized = map;\n  }\n  return cachedNormalized;\n}\n\n/**\n * Find all biomarker names present in OCR text.\n * Uses exact string matching on normalized text.\n * Returns unique matches (same biomarker won't be matched twice).\n */\nexport function findBiomarkersInText(ocrText: string): AnchorResult {\n  const startTime = Date.now();\n  const normalizedText = normalize(ocrText);\n  const matchedCodes = new Set<string>();\n  const matches: AnchorMatch[] = [];\n  const normalizedPatterns = getNormalizedPatterns();\n\n  for (const [normalizedName, entries] of normalizedPatterns) {\n    if (normalizedName.length < 3 && !UNAMBIGUOUS_SHORT_NAMES.has(normalizedName)) {\n      continue;\n    }\n\n    let position = -1;\n    if (normalizedName.length <= 4) {\n      const regex = new RegExp(`\\\\b${normalizedName}\\\\b`);\n      const match = regex.exec(normalizedText);\n      if (match) {\n        position = match.index;\n      }\n    } else {\n      position = normalizedText.indexOf(normalizedName);\n    }\n\n    if (position !== -1) {\n      for (const entry of entries) {\n        if (!matchedCodes.has(entry.code)) {\n          matchedCodes.add(entry.code);\n          matches.push({\n            code: entry.code,\n            confidence: 1.0,\n            loinc: entry.loinc,\n            matchedName: entry.original,\n            position,\n          });\n        }\n      }\n    }\n  }\n\n  const scanTimeMs = Date.now() - startTime;\n  const matchedCodesArray = Array.from(matchedCodes);\n\n  return {\n    filteredReference: generateFilteredLLMReference(matchedCodesArray),\n    matches,\n    stats: {\n      matchedCount: matches.length,\n      scanTimeMs,\n      totalPatterns: getPatterns().length,\n    },\n  };\n}\n\n/**\n * Get the list of matched biomarker codes from an anchor result.\n */\nexport function getMatchedCodes(result: AnchorResult): string[] {\n  return result.matches.map((m) => m.code);\n}\n"]}

package/dist/index.d.cts

@@ -0,0 +1,35 @@
+/**
+ * OCR Anchor — Biomarker text anchoring
+ *
+ * Scans OCR text for biomarker names BEFORE sending to LLM.
+ * This prevents hallucination by constraining what biomarkers
+ * the LLM is allowed to extract.
+ */
+interface AnchorMatch {
+    code: string;
+    confidence: number;
+    loinc?: string;
+    matchedName: string;
+    position: number;
+}
+interface AnchorResult {
+    filteredReference: string;
+    matches: AnchorMatch[];
+    stats: {
+        totalPatterns: number;
+        matchedCount: number;
+        scanTimeMs: number;
+    };
+}
+/**
+ * Find all biomarker names present in OCR text.
+ * Uses exact string matching on normalized text.
+ * Returns unique matches (same biomarker won't be matched twice).
+ */
+declare function findBiomarkersInText(ocrText: string): AnchorResult;
+/**
+ * Get the list of matched biomarker codes from an anchor result.
+ */
+declare function getMatchedCodes(result: AnchorResult): string[];
+
+export { type AnchorMatch, type AnchorResult, findBiomarkersInText, getMatchedCodes };

package/dist/index.d.ts

@@ -0,0 +1,35 @@
+/**
+ * OCR Anchor — Biomarker text anchoring
+ *
+ * Scans OCR text for biomarker names BEFORE sending to LLM.
+ * This prevents hallucination by constraining what biomarkers
+ * the LLM is allowed to extract.
+ */
+interface AnchorMatch {
+    code: string;
+    confidence: number;
+    loinc?: string;
+    matchedName: string;
+    position: number;
+}
+interface AnchorResult {
+    filteredReference: string;
+    matches: AnchorMatch[];
+    stats: {
+        totalPatterns: number;
+        matchedCount: number;
+        scanTimeMs: number;
+    };
+}
+/**
+ * Find all biomarker names present in OCR text.
+ * Uses exact string matching on normalized text.
+ * Returns unique matches (same biomarker won't be matched twice).
+ */
+declare function findBiomarkersInText(ocrText: string): AnchorResult;
+/**
+ * Get the list of matched biomarker codes from an anchor result.
+ */
+declare function getMatchedCodes(result: AnchorResult): string[];
+
+export { type AnchorMatch, type AnchorResult, findBiomarkersInText, getMatchedCodes };

package/dist/index.js

@@ -0,0 +1,126 @@
+// src/anchor.ts
+import {
+  generateFilteredLLMReference,
+  getAllSearchPatterns
+} from "@precisa-saude/fhir";
+function normalize(text) {
+  return text.normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase().replace(/\s+/g, " ");
+}
+var UNAMBIGUOUS_SHORT_NAMES = /* @__PURE__ */ new Set([
+  "hdl",
+  "ldl",
+  "lh",
+  "tsh",
+  "crp",
+  "pcr",
+  "ggt",
+  "alt",
+  "ast",
+  "bun",
+  "wbc",
+  "rbc",
+  "mcv",
+  "mch",
+  "rdw",
+  "mpv",
+  "psa",
+  "fsh",
+  "hba1c",
+  "egfr",
+  "acr",
+  "esr",
+  "vhs",
+  "bmc",
+  "bmd",
+  "vat",
+  "dxa",
+  "dmo",
+  "cmo",
+  "ffm",
+  "lbm",
+  "mlg",
+  "tav"
+]);
+var cachedPatterns = null;
+var cachedNormalized = null;
+function getPatterns() {
+  if (!cachedPatterns) {
+    cachedPatterns = getAllSearchPatterns();
+  }
+  return cachedPatterns;
+}
+function getNormalizedPatterns() {
+  if (!cachedNormalized) {
+    const patterns = getPatterns();
+    const map = /* @__PURE__ */ new Map();
+    for (const pattern of patterns) {
+      for (const name of pattern.names) {
+        const normalized = normalize(name);
+        const existing = map.get(normalized) || [];
+        existing.push({
+          code: pattern.code,
+          ...pattern.loinc && { loinc: pattern.loinc },
+          original: name
+        });
+        map.set(normalized, existing);
+      }
+    }
+    cachedNormalized = map;
+  }
+  return cachedNormalized;
+}
+function findBiomarkersInText(ocrText) {
+  const startTime = Date.now();
+  const normalizedText = normalize(ocrText);
+  const matchedCodes = /* @__PURE__ */ new Set();
+  const matches = [];
+  const normalizedPatterns = getNormalizedPatterns();
+  for (const [normalizedName, entries] of normalizedPatterns) {
+    if (normalizedName.length < 3 && !UNAMBIGUOUS_SHORT_NAMES.has(normalizedName)) {
+      continue;
+    }
+    let position = -1;
+    if (normalizedName.length <= 4) {
+      const regex = new RegExp(`\\b${normalizedName}\\b`);
+      const match = regex.exec(normalizedText);
+      if (match) {
+        position = match.index;
+      }
+    } else {
+      position = normalizedText.indexOf(normalizedName);
+    }
+    if (position !== -1) {
+      for (const entry of entries) {
+        if (!matchedCodes.has(entry.code)) {
+          matchedCodes.add(entry.code);
+          matches.push({
+            code: entry.code,
+            confidence: 1,
+            loinc: entry.loinc,
+            matchedName: entry.original,
+            position
+          });
+        }
+      }
+    }
+  }
+  const scanTimeMs = Date.now() - startTime;
+  const matchedCodesArray = Array.from(matchedCodes);
+  return {
+    filteredReference: generateFilteredLLMReference(matchedCodesArray),
+    matches,
+    stats: {
+      matchedCount: matches.length,
+      scanTimeMs,
+      totalPatterns: getPatterns().length
+    }
+  };
+}
+function getMatchedCodes(result) {
+  return result.matches.map((m) => m.code);
+}
+export {
+  findBiomarkersInText,
+  getMatchedCodes
+};
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/anchor.ts"],"sourcesContent":["/**\n * OCR Anchor — Biomarker text anchoring\n *\n * Scans OCR text for biomarker names BEFORE sending to LLM.\n * This prevents hallucination by constraining what biomarkers\n * the LLM is allowed to extract.\n */\n\nimport {\n  type BiomarkerSearchPattern,\n  generateFilteredLLMReference,\n  getAllSearchPatterns,\n} from '@precisa-saude/fhir';\n\nexport interface AnchorMatch {\n  code: string;\n  confidence: number;\n  loinc?: string;\n  matchedName: string;\n  position: number;\n}\n\nexport interface AnchorResult {\n  filteredReference: string;\n  matches: AnchorMatch[];\n  stats: {\n    totalPatterns: number;\n    matchedCount: number;\n    scanTimeMs: number;\n  };\n}\n\n/**\n * Normalize text for comparison:\n * - Removes diacritics (ã→a, ç→c, é→e)\n * - Converts to lowercase\n * - Normalizes whitespace\n */\nfunction normalize(text: string): string {\n  return text\n    .normalize('NFD')\n    .replace(/[\\u0300-\\u036f]/g, '')\n    .toLowerCase()\n    .replace(/\\s+/g, ' ');\n}\n\nconst UNAMBIGUOUS_SHORT_NAMES = new Set([\n  'hdl', 'ldl', 'lh', 'tsh', 'crp', 'pcr', 'ggt', 'alt', 'ast', 'bun',\n  'wbc', 'rbc', 'mcv', 'mch', 'rdw', 'mpv', 'psa', 'fsh', 'hba1c', 'egfr',\n  'acr', 'esr', 'vhs', 'bmc', 'bmd', 'vat', 'dxa', 'dmo', 'cmo', 'ffm',\n  'lbm', 'mlg', 'tav',\n]);\n\ntype PatternEntry = { original: string; code: string; loinc?: string };\n\nlet cachedPatterns: BiomarkerSearchPattern[] | null = null;\nlet cachedNormalized: Map<string, PatternEntry[]> | null = null;\n\nfunction getPatterns(): BiomarkerSearchPattern[] {\n  if (!cachedPatterns) {\n    cachedPatterns = getAllSearchPatterns();\n  }\n  return cachedPatterns;\n}\n\nfunction getNormalizedPatterns(): Map<string, PatternEntry[]> {\n  if (!cachedNormalized) {\n    const patterns = getPatterns();\n    const map = new Map<string, PatternEntry[]>();\n    for (const pattern of patterns) {\n      for (const name of pattern.names) {\n        const normalized = normalize(name);\n        const existing = map.get(normalized) || [];\n        existing.push({\n          code: pattern.code,\n          ...(pattern.loinc && { loinc: pattern.loinc }),\n          original: name,\n        });\n        map.set(normalized, existing);\n      }\n    }\n    cachedNormalized = map;\n  }\n  return cachedNormalized;\n}\n\n/**\n * Find all biomarker names present in OCR text.\n * Uses exact string matching on normalized text.\n * Returns unique matches (same biomarker won't be matched twice).\n */\nexport function findBiomarkersInText(ocrText: string): AnchorResult {\n  const startTime = Date.now();\n  const normalizedText = normalize(ocrText);\n  const matchedCodes = new Set<string>();\n  const matches: AnchorMatch[] = [];\n  const normalizedPatterns = getNormalizedPatterns();\n\n  for (const [normalizedName, entries] of normalizedPatterns) {\n    if (normalizedName.length < 3 && !UNAMBIGUOUS_SHORT_NAMES.has(normalizedName)) {\n      continue;\n    }\n\n    let position = -1;\n    if (normalizedName.length <= 4) {\n      const regex = new RegExp(`\\\\b${normalizedName}\\\\b`);\n      const match = regex.exec(normalizedText);\n      if (match) {\n        position = match.index;\n      }\n    } else {\n      position = normalizedText.indexOf(normalizedName);\n    }\n\n    if (position !== -1) {\n      for (const entry of entries) {\n        if (!matchedCodes.has(entry.code)) {\n          matchedCodes.add(entry.code);\n          matches.push({\n            code: entry.code,\n            confidence: 1.0,\n            loinc: entry.loinc,\n            matchedName: entry.original,\n            position,\n          });\n        }\n      }\n    }\n  }\n\n  const scanTimeMs = Date.now() - startTime;\n  const matchedCodesArray = Array.from(matchedCodes);\n\n  return {\n    filteredReference: generateFilteredLLMReference(matchedCodesArray),\n    matches,\n    stats: {\n      matchedCount: matches.length,\n      scanTimeMs,\n      totalPatterns: getPatterns().length,\n    },\n  };\n}\n\n/**\n * Get the list of matched biomarker codes from an anchor result.\n */\nexport function getMatchedCodes(result: AnchorResult): string[] {\n  return result.matches.map((m) => m.code);\n}\n"],"mappings":";AAQA;AAAA,EAEE;AAAA,EACA;AAAA,OACK;AA0BP,SAAS,UAAU,MAAsB;AACvC,SAAO,KACJ,UAAU,KAAK,EACf,QAAQ,oBAAoB,EAAE,EAC9B,YAAY,EACZ,QAAQ,QAAQ,GAAG;AACxB;AAEA,IAAM,0BAA0B,oBAAI,IAAI;AAAA,EACtC;AAAA,EAAO;AAAA,EAAO;AAAA,EAAM;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAC9D;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAS;AAAA,EACjE;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAAO;AAAA,EAC/D;AAAA,EAAO;AAAA,EAAO;AAChB,CAAC;AAID,IAAI,iBAAkD;AACtD,IAAI,mBAAuD;AAE3D,SAAS,cAAwC;AAC/C,MAAI,CAAC,gBAAgB;AACnB,qBAAiB,qBAAqB;AAAA,EACxC;AACA,SAAO;AACT;AAEA,SAAS,wBAAqD;AAC5D,MAAI,CAAC,kBAAkB;AACrB,UAAM,WAAW,YAAY;AAC7B,UAAM,MAAM,oBAAI,IAA4B;AAC5C,eAAW,WAAW,UAAU;AAC9B,iBAAW,QAAQ,QAAQ,OAAO;AAChC,cAAM,aAAa,UAAU,IAAI;AACjC,cAAM,WAAW,IAAI,IAAI,UAAU,KAAK,CAAC;AACzC,iBAAS,KAAK;AAAA,UACZ,MAAM,QAAQ;AAAA,UACd,GAAI,QAAQ,SAAS,EAAE,OAAO,QAAQ,MAAM;AAAA,UAC5C,UAAU;AAAA,QACZ,CAAC;AACD,YAAI,IAAI,YAAY,QAAQ;AAAA,MAC9B;AAAA,IACF;AACA,uBAAmB;AAAA,EACrB;AACA,SAAO;AACT;AAOO,SAAS,qBAAqB,SAA+B;AAClE,QAAM,YAAY,KAAK,IAAI;AAC3B,QAAM,iBAAiB,UAAU,OAAO;AACxC,QAAM,eAAe,oBAAI,IAAY;AACrC,QAAM,UAAyB,CAAC;AAChC,QAAM,qBAAqB,sBAAsB;AAEjD,aAAW,CAAC,gBAAgB,OAAO,KAAK,oBAAoB;AAC1D,QAAI,eAAe,SAAS,KAAK,CAAC,wBAAwB,IAAI,cAAc,GAAG;AAC7E;AAAA,IACF;AAEA,QAAI,WAAW;AACf,QAAI,eAAe,UAAU,GAAG;AAC9B,YAAM,QAAQ,IAAI,OAAO,MAAM,cAAc,KAAK;AAClD,YAAM,QAAQ,MAAM,KAAK,cAAc;AACvC,UAAI,OAAO;AACT,mBAAW,MAAM;AAAA,MACnB;AAAA,IACF,OAAO;AACL,iBAAW,eAAe,QAAQ,cAAc;AAAA,IAClD;AAEA,QAAI,aAAa,IAAI;AACnB,iBAAW,SAAS,SAAS;AAC3B,YAAI,CAAC,aAAa,IAAI,MAAM,IAAI,GAAG;AACjC,uBAAa,IAAI,MAAM,IAAI;AAC3B,kBAAQ,KAAK;AAAA,YACX,MAAM,MAAM;AAAA,YACZ,YAAY;AAAA,YACZ,OAAO,MAAM;AAAA,YACb,aAAa,MAAM;AAAA,YACnB;AAAA,UACF,CAAC;AAAA,QACH;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,QAAM,aAAa,KAAK,IAAI,IAAI;AAChC,QAAM,oBAAoB,MAAM,KAAK,YAAY;AAEjD,SAAO;AAAA,IACL,mBAAmB,6BAA6B,iBAAiB;AAAA,IACjE;AAAA,IACA,OAAO;AAAA,MACL,cAAc,QAAQ;AAAA,MACtB;AAAA,MACA,eAAe,YAAY,EAAE;AAAA,IAC/B;AAAA,EACF;AACF;AAKO,SAAS,gBAAgB,QAAgC;AAC9D,SAAO,OAAO,QAAQ,IAAI,CAAC,MAAM,EAAE,IAAI;AACzC;","names":[]}

package/package.json

@@ -0,0 +1,48 @@
+{
+  "name": "@precisa-saude/fhir-ocr-utils",
+  "version": "0.1.0",
+  "description": "OCR text anchoring utilities for biomarker extraction from lab result PDFs",
+  "type": "module",
+  "main": "./dist/index.cjs",
+  "module": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "import": {
+        "types": "./dist/index.d.ts",
+        "default": "./dist/index.js"
+      },
+      "require": {
+        "types": "./dist/index.d.cts",
+        "default": "./dist/index.cjs"
+      }
+    }
+  },
+  "files": ["dist"],
+  "scripts": {
+    "build": "tsup",
+    "lint": "tsc --noEmit",
+    "typecheck": "tsc --noEmit",
+    "test": "vitest run --passWithNoTests",
+    "test:coverage": "vitest run --coverage",
+    "clean": "rm -rf dist .turbo"
+  },
+  "keywords": ["fhir", "ocr", "biomarker", "lab-results", "text-anchoring"],
+  "license": "Apache-2.0",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/precisa-saude/fhir-brasil.git",
+    "directory": "packages/ocr-utils"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "dependencies": {
+    "@precisa-saude/fhir": "workspace:*"
+  },
+  "devDependencies": {
+    "tsup": "^8.3.5",
+    "typescript": "~5.7.3",
+    "vitest": "^2.1.8"
+  }
+}