@precisa-saude/fhir-ocr-utils 0.31.3 → 0.31.5

package/dist/cli.js

@@ -20,13 +20,8 @@ import {
   getAllSearchPatterns,
   UNIT_TO_UCUM
 } from "@precisa-saude/fhir";
-var CONFIDENCE_VALUE_ADJACENT = 1;
-var CONFIDENCE_NAME_ONLY = 0.7;
-var CONFIDENCE_AMBIGUOUS = 0.4;
-var MAX_OCCURRENCES_PER_NAME = 5;
-function normalize(text) {
-  return text.normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase().replace(/(?<=\p{L})-(?=[\p{L}\p{N}])/gu, " ").replace(/[^\S\n]+/g, " ");
-}
+
+// src/anchor-lexicon.ts
 var UNAMBIGUOUS_SHORT_NAMES = /* @__PURE__ */ new Set([
   "hdl",
   "ldl",
@@ -140,6 +135,85 @@ var QUALITATIVE_VALUE_TERMS = /* @__PURE__ */ new Set([
   "undetectable",
   "yellow"
 ]);
+
+// src/body-region.ts
+var WHOLE_BODY_COMPOSITION_CODES = /* @__PURE__ */ new Set([
+  "BMC",
+  "BodyFatPct",
+  "FatFreeMass",
+  "FatMass",
+  "LeanMass",
+  "TotalMass"
+]);
+var LIMB_PAIR_CODES = /* @__PURE__ */ new Set(["ArmsFatMass", "ArmsLeanMass", "LegsFatMass", "LegsLeanMass"]);
+var BODY_REGION = String.raw`(?:arms?|legs?|trunk|head|android|gynoid|bracos?|pernas?|tronco|cabeca|androide|ginoide)`;
+var BODY_SIDE = String.raw`(?:right|left|direit[oa]|esquerd[oa])`;
+var ENDS_WITH_BODY_REGION = new RegExp(
+  String.raw`(?<![\p{L}\p{N}])${BODY_REGION}(?: (?:total|difference|diferenca|${BODY_SIDE}))* ?$`,
+  "u"
+);
+var ENDS_WITH_BODY_SIDE = new RegExp(String.raw`(?<![\p{L}\p{N}])${BODY_SIDE} ?$`, "u");
+function qualifiedByBodyRegion(code, before) {
+  if (WHOLE_BODY_COMPOSITION_CODES.has(code)) {
+    return ENDS_WITH_BODY_REGION.test(before);
+  }
+  if (LIMB_PAIR_CODES.has(code)) {
+    return ENDS_WITH_BODY_SIDE.test(before);
+  }
+  return false;
+}
+
+// src/anchor.ts
+var CONFIDENCE_VALUE_ADJACENT = 1;
+var CONFIDENCE_NAME_ONLY = 0.7;
+var CONFIDENCE_AMBIGUOUS = 0.4;
+var MAX_OCCURRENCES_PER_NAME = 5;
+function normalizeBase(text) {
+  return text.normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase().replace(/(?<=\p{L})-(?=[\p{L}\p{N}])/gu, " ").replace(/[^\S\n]+/g, " ");
+}
+function normalize(text) {
+  return foldCommas(normalizeBase(text));
+}
+var cachedNamesByHead = null;
+function isIndexedName(normalized) {
+  return normalized.length >= 3 || UNAMBIGUOUS_SHORT_NAMES.has(normalized);
+}
+function getNamesByHead() {
+  if (!cachedNamesByHead) {
+    const map = /* @__PURE__ */ new Map();
+    for (const pattern of getPatterns()) {
+      for (const name of pattern.names) {
+        const normalized = normalizeBase(name).trim();
+        if (!normalized || !isIndexedName(normalized)) {
+          continue;
+        }
+        const head = normalized.split(" ")[0];
+        map.set(head, [...map.get(head) ?? [], normalized]);
+      }
+    }
+    cachedNamesByHead = map;
+  }
+  return cachedNamesByHead;
+}
+function startsWithCatalogName(text) {
+  const head = /^[\p{L}\p{N}]+/u.exec(text)?.[0];
+  if (!head) {
+    return false;
+  }
+  return (getNamesByHead().get(head) ?? []).some((name) => {
+    if (!text.startsWith(name)) {
+      return false;
+    }
+    const after = text.slice(name.length).replace(/^s/, "");
+    return !/^[\p{L}\p{N}]/u.test(after);
+  });
+}
+function foldCommas(text) {
+  return text.replace(
+    /(?<=\p{L}),([^\S\n]*)(?=\p{L})/gu,
+    (comma, space, offset) => startsWithCatalogName(text.slice(offset + 1 + space.length)) ? comma : " "
+  );
+}
 var GENETIC_CONTEXT_PATTERNS = [
   /\b[nx][mrpc]_\d{6,}/,
   // RefSeq: NM_000384.2, NP_, NR_, XM_
@@ -233,10 +307,7 @@ function getNamePatterns() {
     for (const pattern of getPatterns()) {
       for (const name of pattern.names) {
         const normalized = normalize(name).trim();
-        if (!normalized) {
-          continue;
-        }
-        if (normalized.length < 3 && !UNAMBIGUOUS_SHORT_NAMES.has(normalized)) {
+        if (!normalized || !isIndexedName(normalized)) {
           continue;
         }
         let slot = map.get(normalized);
@@ -295,6 +366,56 @@ function collectCandidates(normalizedText) {
   }
   return candidates;
 }
+function joinLines(text) {
+  const parts = [];
+  const offsets = [];
+  let cursor = 0;
+  for (const lineBreak of text.matchAll(/[^\S\n]*\n[^\S\n]*/g)) {
+    for (let i = cursor; i < lineBreak.index; i += 1) {
+      offsets.push(i);
+    }
+    parts.push(text.slice(cursor, lineBreak.index), " ");
+    offsets.push(lineBreak.index + lineBreak[0].indexOf("\n"));
+    cursor = lineBreak.index + lineBreak[0].length;
+  }
+  for (let i = cursor; i < text.length; i += 1) {
+    offsets.push(i);
+  }
+  parts.push(text.slice(cursor));
+  return { joined: parts.join(""), offsets };
+}
+function collectWrappedCandidates(normalizedText) {
+  if (!normalizedText.includes("\n")) {
+    return [];
+  }
+  const { joined, offsets } = joinLines(normalizedText);
+  const namePatterns = getNamePatterns();
+  const candidates = [];
+  for (const [name, slot] of namePatterns) {
+    if (!name.includes(" ") || !joined.includes(name)) {
+      continue;
+    }
+    const regex = slot.regex ??= buildNamePattern(name);
+    regex.lastIndex = 0;
+    let occurrences = 0;
+    let match = regex.exec(joined);
+    while (match !== null && occurrences < MAX_OCCURRENCES_PER_NAME) {
+      const start = offsets[match.index];
+      const end = offsets[match.index + match[0].length - 1] + 1;
+      const lineBreak = normalizedText.indexOf("\n", start);
+      const nextBreak = normalizedText.indexOf("\n", lineBreak + 1);
+      const wrapped = lineBreak !== -1 && lineBreak < end && (nextBreak === -1 || nextBreak >= end);
+      if (wrapped && !namePatterns.has(normalizedText.slice(start, lineBreak).trim()) && !hasValueEvidence(
+        normalizedText.slice(getLineBounds(normalizedText, start).start, lineBreak)
+      )) {
+        candidates.push({ end, entries: slot.entries, start });
+        occurrences += 1;
+      }
+      match = regex.exec(joined);
+    }
+  }
+  return candidates;
+}
 function resolveOverlaps(candidates) {
   const sorted = [...candidates].sort(
     (a, b) => b.end - b.start - (a.end - a.start) || a.start - b.start
@@ -315,33 +436,37 @@ function findBiomarkersInText(ocrText) {
   const startTime = Date.now();
   const normalizedText = normalize(ocrText);
   const bestByCode = /* @__PURE__ */ new Map();
-  const geneticLines = /* @__PURE__ */ new Map();
-  const valueLines = /* @__PURE__ */ new Map();
-  const girthLines = /* @__PURE__ */ new Map();
-  for (const candidate of resolveOverlaps(collectCandidates(normalizedText))) {
-    const { end: lineEnd, start: lineStart } = getLineBounds(normalizedText, candidate.start);
-    let genetic = geneticLines.get(lineStart);
-    if (genetic === void 0) {
-      genetic = hasGeneticContext(normalizedText.slice(lineStart, lineEnd));
-      geneticLines.set(lineStart, genetic);
+  const contexts = /* @__PURE__ */ new Map();
+  const candidates = [
+    ...collectCandidates(normalizedText),
+    ...collectWrappedCandidates(normalizedText)
+  ];
+  for (const candidate of resolveOverlaps(candidates)) {
+    const lineStart = getLineBounds(normalizedText, candidate.start).start;
+    const lineEnd = getLineBounds(normalizedText, candidate.end - 1).end;
+    const key = `${lineStart}:${lineEnd}`;
+    let context = contexts.get(key);
+    if (!context) {
+      const line = normalizedText.slice(lineStart, lineEnd);
+      context = {
+        genetic: hasGeneticContext(line),
+        girth: hasGirthContext(line),
+        hasValue: hasValueEvidence(line)
+      };
+      contexts.set(key, context);
     }
+    const { genetic, girth, hasValue } = context;
     if (genetic) {
       continue;
     }
-    let hasValue = valueLines.get(lineStart);
-    if (hasValue === void 0) {
-      hasValue = hasValueEvidence(normalizedText.slice(lineStart, lineEnd));
-      valueLines.set(lineStart, hasValue);
-    }
-    let girth = girthLines.get(lineStart);
-    if (girth === void 0) {
-      girth = hasGirthContext(normalizedText.slice(lineStart, lineEnd));
-      girthLines.set(lineStart, girth);
-    }
+    const before = normalizedText.slice(lineStart, candidate.start);
     for (const entry of candidate.entries) {
       if (entry.ambiguous && !hasValue) {
         continue;
       }
+      if (qualifiedByBodyRegion(entry.code, before)) {
+        continue;
+      }
       if (girth && SKINFOLD_SITE_CODES.has(entry.code)) {
         continue;
       }
@@ -884,7 +1009,7 @@ async function main() {
     strict: false
   });
   if (values.version) {
-    process.stdout.write(`${"0.31.3"}
+    process.stdout.write(`${"0.31.5"}
 `);
     return;
   }

package/dist/index.cjs

@@ -4,13 +4,8 @@
 
 
 var _fhir = require('@precisa-saude/fhir');
-var CONFIDENCE_VALUE_ADJACENT = 1;
-var CONFIDENCE_NAME_ONLY = 0.7;
-var CONFIDENCE_AMBIGUOUS = 0.4;
-var MAX_OCCURRENCES_PER_NAME = 5;
-function normalize(text) {
-  return text.normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase().replace(/(?<=\p{L})-(?=[\p{L}\p{N}])/gu, " ").replace(/[^\S\n]+/g, " ");
-}
+
+// src/anchor-lexicon.ts
 var UNAMBIGUOUS_SHORT_NAMES = /* @__PURE__ */ new Set([
   "hdl",
   "ldl",
@@ -124,6 +119,85 @@ var QUALITATIVE_VALUE_TERMS = /* @__PURE__ */ new Set([
   "undetectable",
   "yellow"
 ]);
+
+// src/body-region.ts
+var WHOLE_BODY_COMPOSITION_CODES = /* @__PURE__ */ new Set([
+  "BMC",
+  "BodyFatPct",
+  "FatFreeMass",
+  "FatMass",
+  "LeanMass",
+  "TotalMass"
+]);
+var LIMB_PAIR_CODES = /* @__PURE__ */ new Set(["ArmsFatMass", "ArmsLeanMass", "LegsFatMass", "LegsLeanMass"]);
+var BODY_REGION = String.raw`(?:arms?|legs?|trunk|head|android|gynoid|bracos?|pernas?|tronco|cabeca|androide|ginoide)`;
+var BODY_SIDE = String.raw`(?:right|left|direit[oa]|esquerd[oa])`;
+var ENDS_WITH_BODY_REGION = new RegExp(
+  String.raw`(?<![\p{L}\p{N}])${BODY_REGION}(?: (?:total|difference|diferenca|${BODY_SIDE}))* ?$`,
+  "u"
+);
+var ENDS_WITH_BODY_SIDE = new RegExp(String.raw`(?<![\p{L}\p{N}])${BODY_SIDE} ?$`, "u");
+function qualifiedByBodyRegion(code, before) {
+  if (WHOLE_BODY_COMPOSITION_CODES.has(code)) {
+    return ENDS_WITH_BODY_REGION.test(before);
+  }
+  if (LIMB_PAIR_CODES.has(code)) {
+    return ENDS_WITH_BODY_SIDE.test(before);
+  }
+  return false;
+}
+
+// src/anchor.ts
+var CONFIDENCE_VALUE_ADJACENT = 1;
+var CONFIDENCE_NAME_ONLY = 0.7;
+var CONFIDENCE_AMBIGUOUS = 0.4;
+var MAX_OCCURRENCES_PER_NAME = 5;
+function normalizeBase(text) {
+  return text.normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase().replace(/(?<=\p{L})-(?=[\p{L}\p{N}])/gu, " ").replace(/[^\S\n]+/g, " ");
+}
+function normalize(text) {
+  return foldCommas(normalizeBase(text));
+}
+var cachedNamesByHead = null;
+function isIndexedName(normalized) {
+  return normalized.length >= 3 || UNAMBIGUOUS_SHORT_NAMES.has(normalized);
+}
+function getNamesByHead() {
+  if (!cachedNamesByHead) {
+    const map = /* @__PURE__ */ new Map();
+    for (const pattern of getPatterns()) {
+      for (const name of pattern.names) {
+        const normalized = normalizeBase(name).trim();
+        if (!normalized || !isIndexedName(normalized)) {
+          continue;
+        }
+        const head = normalized.split(" ")[0];
+        map.set(head, [..._nullishCoalesce(map.get(head), () => ( [])), normalized]);
+      }
+    }
+    cachedNamesByHead = map;
+  }
+  return cachedNamesByHead;
+}
+function startsWithCatalogName(text) {
+  const head = _optionalChain([/^[\p{L}\p{N}]+/u, 'access', _ => _.exec, 'call', _2 => _2(text), 'optionalAccess', _3 => _3[0]]);
+  if (!head) {
+    return false;
+  }
+  return (_nullishCoalesce(getNamesByHead().get(head), () => ( []))).some((name) => {
+    if (!text.startsWith(name)) {
+      return false;
+    }
+    const after = text.slice(name.length).replace(/^s/, "");
+    return !/^[\p{L}\p{N}]/u.test(after);
+  });
+}
+function foldCommas(text) {
+  return text.replace(
+    /(?<=\p{L}),([^\S\n]*)(?=\p{L})/gu,
+    (comma, space, offset) => startsWithCatalogName(text.slice(offset + 1 + space.length)) ? comma : " "
+  );
+}
 var GENETIC_CONTEXT_PATTERNS = [
   /\b[nx][mrpc]_\d{6,}/,
   // RefSeq: NM_000384.2, NP_, NR_, XM_
@@ -217,10 +291,7 @@ function getNamePatterns() {
     for (const pattern of getPatterns()) {
       for (const name of pattern.names) {
         const normalized = normalize(name).trim();
-        if (!normalized) {
-          continue;
-        }
-        if (normalized.length < 3 && !UNAMBIGUOUS_SHORT_NAMES.has(normalized)) {
+        if (!normalized || !isIndexedName(normalized)) {
           continue;
         }
         let slot = map.get(normalized);
@@ -279,6 +350,56 @@ function collectCandidates(normalizedText) {
   }
   return candidates;
 }
+function joinLines(text) {
+  const parts = [];
+  const offsets = [];
+  let cursor = 0;
+  for (const lineBreak of text.matchAll(/[^\S\n]*\n[^\S\n]*/g)) {
+    for (let i = cursor; i < lineBreak.index; i += 1) {
+      offsets.push(i);
+    }
+    parts.push(text.slice(cursor, lineBreak.index), " ");
+    offsets.push(lineBreak.index + lineBreak[0].indexOf("\n"));
+    cursor = lineBreak.index + lineBreak[0].length;
+  }
+  for (let i = cursor; i < text.length; i += 1) {
+    offsets.push(i);
+  }
+  parts.push(text.slice(cursor));
+  return { joined: parts.join(""), offsets };
+}
+function collectWrappedCandidates(normalizedText) {
+  if (!normalizedText.includes("\n")) {
+    return [];
+  }
+  const { joined, offsets } = joinLines(normalizedText);
+  const namePatterns = getNamePatterns();
+  const candidates = [];
+  for (const [name, slot] of namePatterns) {
+    if (!name.includes(" ") || !joined.includes(name)) {
+      continue;
+    }
+    const regex = slot.regex ??= buildNamePattern(name);
+    regex.lastIndex = 0;
+    let occurrences = 0;
+    let match = regex.exec(joined);
+    while (match !== null && occurrences < MAX_OCCURRENCES_PER_NAME) {
+      const start = offsets[match.index];
+      const end = offsets[match.index + match[0].length - 1] + 1;
+      const lineBreak = normalizedText.indexOf("\n", start);
+      const nextBreak = normalizedText.indexOf("\n", lineBreak + 1);
+      const wrapped = lineBreak !== -1 && lineBreak < end && (nextBreak === -1 || nextBreak >= end);
+      if (wrapped && !namePatterns.has(normalizedText.slice(start, lineBreak).trim()) && !hasValueEvidence(
+        normalizedText.slice(getLineBounds(normalizedText, start).start, lineBreak)
+      )) {
+        candidates.push({ end, entries: slot.entries, start });
+        occurrences += 1;
+      }
+      match = regex.exec(joined);
+    }
+  }
+  return candidates;
+}
 function resolveOverlaps(candidates) {
   const sorted = [...candidates].sort(
     (a, b) => b.end - b.start - (a.end - a.start) || a.start - b.start
@@ -299,33 +420,37 @@ function findBiomarkersInText(ocrText) {
   const startTime = Date.now();
   const normalizedText = normalize(ocrText);
   const bestByCode = /* @__PURE__ */ new Map();
-  const geneticLines = /* @__PURE__ */ new Map();
-  const valueLines = /* @__PURE__ */ new Map();
-  const girthLines = /* @__PURE__ */ new Map();
-  for (const candidate of resolveOverlaps(collectCandidates(normalizedText))) {
-    const { end: lineEnd, start: lineStart } = getLineBounds(normalizedText, candidate.start);
-    let genetic = geneticLines.get(lineStart);
-    if (genetic === void 0) {
-      genetic = hasGeneticContext(normalizedText.slice(lineStart, lineEnd));
-      geneticLines.set(lineStart, genetic);
+  const contexts = /* @__PURE__ */ new Map();
+  const candidates = [
+    ...collectCandidates(normalizedText),
+    ...collectWrappedCandidates(normalizedText)
+  ];
+  for (const candidate of resolveOverlaps(candidates)) {
+    const lineStart = getLineBounds(normalizedText, candidate.start).start;
+    const lineEnd = getLineBounds(normalizedText, candidate.end - 1).end;
+    const key = `${lineStart}:${lineEnd}`;
+    let context = contexts.get(key);
+    if (!context) {
+      const line = normalizedText.slice(lineStart, lineEnd);
+      context = {
+        genetic: hasGeneticContext(line),
+        girth: hasGirthContext(line),
+        hasValue: hasValueEvidence(line)
+      };
+      contexts.set(key, context);
     }
+    const { genetic, girth, hasValue } = context;
     if (genetic) {
       continue;
     }
-    let hasValue = valueLines.get(lineStart);
-    if (hasValue === void 0) {
-      hasValue = hasValueEvidence(normalizedText.slice(lineStart, lineEnd));
-      valueLines.set(lineStart, hasValue);
-    }
-    let girth = girthLines.get(lineStart);
-    if (girth === void 0) {
-      girth = hasGirthContext(normalizedText.slice(lineStart, lineEnd));
-      girthLines.set(lineStart, girth);
-    }
+    const before = normalizedText.slice(lineStart, candidate.start);
     for (const entry of candidate.entries) {
       if (entry.ambiguous && !hasValue) {
         continue;
       }
+      if (qualifiedByBodyRegion(entry.code, before)) {
+        continue;
+      }
       if (girth && SKINFOLD_SITE_CODES.has(entry.code)) {
         continue;
       }
@@ -628,7 +753,7 @@ function buildPrompt(text, allowed) {
 }
 function stripFence(raw) {
   const fenced = /```(?:json)?\s*([\s\S]*?)```/.exec(raw);
-  return (_nullishCoalesce(_optionalChain([fenced, 'optionalAccess', _ => _[1]]), () => ( raw))).trim();
+  return (_nullishCoalesce(_optionalChain([fenced, 'optionalAccess', _4 => _4[1]]), () => ( raw))).trim();
 }
 async function extractWithModel(text, options) {
   const { apiKey, baseUrl, headers = {}, model, responseFormat, timeoutMs = 3e5 } = options;
@@ -668,7 +793,7 @@ async function extractWithModel(text, options) {
     throw new Error(`${String(response.status)} de ${baseUrl}: ${await response.text()}`);
   }
   const body = await response.json();
-  const raw = _nullishCoalesce(_optionalChain([body, 'access', _2 => _2.choices, 'optionalAccess', _3 => _3[0], 'optionalAccess', _4 => _4.message, 'optionalAccess', _5 => _5.content]), () => ( ""));
+  const raw = _nullishCoalesce(_optionalChain([body, 'access', _5 => _5.choices, 'optionalAccess', _6 => _6[0], 'optionalAccess', _7 => _7.message, 'optionalAccess', _8 => _8.content]), () => ( ""));
   const tookMs = Date.now() - startedAt;
   try {
     return { payload: JSON.parse(stripFence(raw)), raw, tookMs };