Ignore PDF prefix truncation AI issues

2026-05-27 16:06:10 +08:00
parent e0574d16a3
commit 54ea338bd6
4 changed files with 101 additions and 185 deletions
--- a/患者首页处理/数据可视化网页端/app/main.py
+++ b/患者首页处理/数据可视化网页端/app/main.py
@@ -4,14 +4,12 @@ import json
 import os
 import base64
 import hashlib
-import tempfile
 import secrets
 import re
 import threading
 import time
 import urllib.error
 import urllib.request
-import zipfile
 from datetime import date, datetime, timedelta
 from decimal import Decimal
 from pathlib import Path
@@ -23,7 +21,7 @@ from zoneinfo import ZoneInfo
 import psycopg2
 import psycopg2.extras
 from dotenv import load_dotenv
-from fastapi import BackgroundTasks, FastAPI, HTTPException, Request, Response
+from fastapi import FastAPI, HTTPException, Request, Response
 from fastapi.responses import FileResponse, JSONResponse
 from fastapi.staticfiles import StaticFiles
 from pydantic import BaseModel, Field
@@ -965,12 +963,6 @@ def session_from_request(request: Request) -> dict[str, Any] | None:
    return SESSIONS.get(token)


-def require_admin_user(request: Request) -> None:
-    user = getattr(request.state, "user", {}) or {}
-    if user.get("username") != admin_username():
-        raise HTTPException(status_code=403, detail="只有 admin 可以使用导出功能")
-
-
 def page_permission_for_path(path: str, method: str) -> str | tuple[str, ...] | None:
    if path in {"/api/status", "/api/schema"}:
        return None
@@ -1960,22 +1952,6 @@ def get_pdf_path(source_file: str) -> Path | None:
    return path if path.exists() and path.is_file() else None


-def safe_download_name(value: Any, fallback: str = "未命名") -> str:
-    text = re.sub(r'[\\/:*?"<>|\r\n]+', "_", str(value or "").strip())
-    text = re.sub(r"\s+", "_", text).strip("._ ")
-    return text[:80] or fallback
-
-
-def record_pdf_download_name(record: dict[str, Any]) -> str:
-    source_file = Path(str(record.get("source_file") or "")).name
-    stem = Path(source_file).stem or f"record_{record.get('id', '')}"
-    suffix = Path(source_file).suffix or ".pdf"
-    patient = safe_download_name(record.get("patient_name"), "")
-    inpatient_no = safe_download_name(record.get("inpatient_no") or record.get("medical_record_no"), "")
-    parts = [part for part in [inpatient_no, patient, stem] if part]
-    return "_".join(parts) + suffix
-
-
 def parse_json_content(content: str) -> Any:
    text = content.strip()
    if text.startswith("```"):
@@ -2369,8 +2345,9 @@ def build_ai_prompt(record: dict[str, Any], pdf_context: dict[str, Any], privacy
 6. 如果手术表格中能看到“手术及操作编码”列但对应单元格为空，写“编码栏可见但为空白”，不要写“编码区域未在首页显示”；单元格本来为空且无需补录时不要因此判为 problem。
 7. 手术操作名称可能因为换行被结构化解析截断；如果 PDF定位文本或局部截图中显示完整多行名称，请把完整名称放入 suggested_updates。
 8. 门急诊诊断编码只能用于 outpatient_diagnosis_code；主要诊断请修正 discharge_diagnoses 中“诊断类别=主要诊断”的行，不要把门急诊诊断编码复制到 discharge_diagnoses[].疾病编码，除非出院诊断表格对应“疾病编码”单元格本身清楚显示该编码。
-9. remaining_issues 只写当前文档复核人应该特别注意的内容；不要写如何修改，不要重复 suggested_updates，不要写“无需补录/无需处理/首页原貌”这类已判定无问题的说明。
-10. 不要编造 PDF 中看不见的内容，不要输出置信度。
+9. PDF局部截图可能因遮挡、截屏边界或缩放只显示编码/文字前半段；如果 PDF 显示值是结构化字段值的前缀，且没有相反证据，应判为 match/ok，不要写“需确认完整编码”，也不要建议把结构化字段截短。
+10. remaining_issues 只写当前文档复核人应该特别注意的内容；不要写如何修改，不要重复 suggested_updates，不要写“无需补录/无需处理/首页原貌”这类已判定无问题的说明。
+11. 不要编造 PDF 中看不见的内容，不要输出置信度。

 必须返回这个 JSON 结构：
 {{
@@ -2538,12 +2515,90 @@ def ai_needs_review_text(value: Any) -> bool:
    return bool(text) and not any(marker in text for marker in AI_NO_REVIEW_MARKERS)


+def ai_compact_compare_value(value: Any) -> str:
+    text = str(value or "").strip()
+    text = text.replace("．", ".").replace("Ｘ", "X").replace("ｘ", "x").replace("×", "x")
+    return re.sub(r"\s+", "", text).upper()
+
+
+def ai_pdf_value_is_structured_prefix(pdf_value: Any, structured_value: Any) -> bool:
+    pdf_text = ai_compact_compare_value(pdf_value)
+    structured_text = ai_compact_compare_value(structured_value)
+    if len(pdf_text) < 3 or len(structured_text) <= len(pdf_text):
+        return False
+    return structured_text.startswith(pdf_text)
+
+
+def ai_text_pdf_prefix_issue(text: Any) -> bool:
+    compact = re.sub(r"\s+", "", str(text or ""))
+    if not compact:
+        return False
+    patterns = [
+        r"PDF(?:显示|中显示|可见|值)?(?:为|是|[:：])?([A-Za-z0-9][A-Za-z0-9.+\-*/xX]*)[，,；;。、]?(?:结构化字段|结构化值|字段|系统字段)(?:为|是|[:：])?([A-Za-z0-9][A-Za-z0-9.+\-*/xX]*)",
+        r"PDF[^，,；;。]*?([A-Za-z0-9][A-Za-z0-9.+\-*/xX]{2,})[^，,；;。]*?(?:结构化字段|结构化值|字段|系统字段)[^，,；;。]*?([A-Za-z0-9][A-Za-z0-9.+\-*/xX]{2,})",
+    ]
+    return any(
+        ai_pdf_value_is_structured_prefix(match.group(1), match.group(2))
+        for pattern in patterns
+        for match in re.finditer(pattern, compact, flags=re.IGNORECASE)
+    )
+
+
+def ai_pdf_prefix_truncation_issue(value: Any) -> bool:
+    if isinstance(value, dict):
+        pdf_value = value.get("pdf_value") or value.get("pdf") or value.get("PDF值") or value.get("图片值")
+        structured_value = (
+            value.get("structured_value")
+            or value.get("structured")
+            or value.get("field_value")
+            or value.get("结构化值")
+            or value.get("结构化字段")
+        )
+        if ai_pdf_value_is_structured_prefix(pdf_value, structured_value):
+            return True
+        return ai_text_pdf_prefix_issue(value)
+    return ai_text_pdf_prefix_issue(value)
+
+
+def ai_has_nonblank_suggested_updates(parsed: dict[str, Any]) -> bool:
+    suggested_updates = parsed.get("suggested_updates")
+    if not isinstance(suggested_updates, list):
+        return False
+    return any(isinstance(item, dict) and not blank_ai_value(suggested_update_value(item)) for item in suggested_updates)
+
+
+def ai_problem_evidence(parsed: dict[str, Any]) -> list[dict[str, Any]]:
+    evidence = parsed.get("evidence")
+    if not isinstance(evidence, list):
+        return []
+    problem_results = {"mismatch", "uncertain", "missing", "problem", "not_match", "not ok", "not_ok"}
+    return [
+        item
+        for item in evidence
+        if isinstance(item, dict) and str(item.get("result") or "").strip().lower() in problem_results
+    ]
+
+
+def ai_only_pdf_prefix_truncation(parsed: dict[str, Any]) -> bool:
+    if ai_has_nonblank_suggested_updates(parsed):
+        return False
+    issues = ai_remaining_issues(parsed)
+    if issues:
+        return all(ai_pdf_prefix_truncation_issue(issue) for issue in issues)
+    evidence = ai_problem_evidence(parsed)
+    if evidence:
+        return all(ai_pdf_prefix_truncation_issue(item) for item in evidence)
+    return ai_pdf_prefix_truncation_issue(parsed.get("summary")) or ai_pdf_prefix_truncation_issue(parsed.get("decision"))
+
+
 def normalize_ai_parsed(parsed: dict[str, Any]) -> dict[str, Any]:
    normalized = dict(parsed)
    remaining = normalized.get("remaining_issues")
    if isinstance(remaining, list):
-        normalized["remaining_issues"] = [item for item in remaining if ai_needs_review_text(item)]
-    elif remaining and ai_needs_review_text(remaining):
+        normalized["remaining_issues"] = [
+            item for item in remaining if ai_needs_review_text(item) and not ai_pdf_prefix_truncation_issue(item)
+        ]
+    elif remaining and ai_needs_review_text(remaining) and not ai_pdf_prefix_truncation_issue(remaining):
        normalized["remaining_issues"] = [remaining]
    else:
        normalized["remaining_issues"] = []
@@ -2567,7 +2622,7 @@ def ai_has_confirmed_problem(parsed: dict[str, Any]) -> bool:
            if not isinstance(item, dict):
                continue
            result = str(item.get("result") or "").strip().lower()
-            if result in {"mismatch", "uncertain", "missing", "problem"}:
+            if result in {"mismatch", "uncertain", "missing", "problem"} and not ai_pdf_prefix_truncation_issue(item):
                return True

    if resolution in {"false_positive", "ok", "no_issue", "误报", "无问题"}:
@@ -2596,6 +2651,8 @@ def ai_remaining_issues(parsed: dict[str, Any]) -> list[Any]:


 def ai_has_unresolved_problem(parsed: dict[str, Any]) -> bool:
+    if ai_only_pdf_prefix_truncation(parsed):
+        return False
    if ai_remaining_issues(parsed):
        return True

@@ -2630,22 +2687,20 @@ def ai_classification(parsed: dict[str, Any]) -> str:
    resolution = str(parsed.get("issue_resolution") or "").strip().lower()
    ok_values = {"ok", "pass", "passed", "no_issue", "no issue", "无问题", "通过", "已通过"}
    problem_values = {"problem", "not_ok", "not ok", "confirm", "不ok", "不通过", "需确认", "待确认", "需复核"}
+    if ai_only_pdf_prefix_truncation(parsed):
+        return AI_OK_STATUS
    if classification in ok_values or decision in ok_values or resolution in {"false_positive", "ok", "no_issue", "误报", "无问题", "通过"}:
        return AI_OK_STATUS
    if classification in problem_values or decision in problem_values or resolution in {"confirmed_problem", "uncertain", "problem", "待确认", "已证实"}:
+        if not ai_remaining_issues(parsed) and ai_has_nonblank_suggested_updates(parsed):
+            return AI_OK_STATUS
        return AI_PROBLEM_STATUS
    if ai_remaining_issues(parsed) or ai_has_unresolved_problem(parsed):
        return AI_PROBLEM_STATUS
    suggested_updates = parsed.get("suggested_updates")
    if isinstance(suggested_updates, list) and suggested_updates:
        return AI_OK_STATUS
-    evidence = parsed.get("evidence")
-    if isinstance(evidence, list):
-        for item in evidence:
-            if not isinstance(item, dict):
-                continue
-            result = str(item.get("result") or "").strip().lower()
-            if result in {"mismatch", "uncertain", "missing", "problem"}:
+    if any(not ai_pdf_prefix_truncation_issue(item) for item in ai_problem_evidence(parsed)):
        return AI_PROBLEM_STATUS
    return AI_OK_STATUS

@@ -2774,12 +2829,16 @@ def ai_suggested_updates(result: dict[str, Any], before: dict[str, Any]) -> tupl
            compare_old = rows[index].get(key)
            if ai_outpatient_code_leak(path, value, before, compare_old):
                continue
+            if ai_pdf_value_is_structured_prefix(value, compare_old):
+                continue
            if comparable(compare_old) == comparable(value):
                continue
            rows[index][key] = value
            new_value = rows
            label = f"{FIELD_META[field]['label']}[{index + 1}].{key}"

+        if index is None and ai_pdf_value_is_structured_prefix(value, compare_old):
+            continue
        if comparable(compare_old) == comparable(value):
            continue
        working[field] = new_value
@@ -3397,77 +3456,11 @@ def list_records(q: str = "", status_filter: str = "review_all", limit: int = 30
    return {"records": rows, "limit": limit, "offset": offset, "has_more": has_more}


-@app.get("/api/records/export")
-def export_records(request: Request, background_tasks: BackgroundTasks, q: str = "", status_filter: str = "review_all"):
-    require_admin_user(request)
-    where_sql, params = record_filter_sql(q, status_filter)
-    max_export = 5000
-    query = sql.SQL(
-        """
-        SELECT
-          id, source_file, inpatient_no, medical_record_no, patient_name, review_status,
-          NULLIF(review_logs->-1->>'changed_at', '')::timestamp AS last_activity_at
-        FROM {table}
-        {where}
-        {order_by}
-        LIMIT %s
-        """
-    ).format(table=table_identifier(), where=where_sql, order_by=RECORD_LIST_ORDER_SQL)
-    with connect() as conn, conn.cursor() as cur:
-        cur.execute(query, [*params, max_export])
-        rows = [row_to_json(dict(row)) for row in cur.fetchall()]
-
-    export_items: list[tuple[dict[str, Any], Path]] = []
-    for row in rows:
-        pdf_path = get_pdf_path(row.get("source_file") or "")
-        if pdf_path:
-            export_items.append((row, pdf_path))
-    if not export_items:
-        raise HTTPException(status_code=404, detail="当前筛选下没有可导出的 PDF")
-
-    handle = tempfile.NamedTemporaryFile(prefix="frontpage_export_", suffix=".zip", delete=False)
-    zip_path = Path(handle.name)
-    handle.close()
-    used_names: set[str] = set()
-    with zipfile.ZipFile(zip_path, "w", compression=zipfile.ZIP_DEFLATED, allowZip64=True) as archive:
-        for index, (record, pdf_path) in enumerate(export_items, start=1):
-            name = record_pdf_download_name(record)
-            arcname = f"{index:04d}_{name}"
-            while arcname in used_names:
-                arcname = f"{index:04d}_{record.get('id')}_{name}"
-            used_names.add(arcname)
-            archive.write(pdf_path, arcname)
-    background_tasks.add_task(zip_path.unlink, missing_ok=True)
-    stamp = datetime.now(APP_TIMEZONE).strftime("%Y%m%d_%H%M%S")
-    return FileResponse(
-        zip_path,
-        media_type="application/zip",
-        filename=f"患者首页影像导出_{status_filter}_{stamp}.zip",
-        background=background_tasks,
-        headers={"Cache-Control": "no-store", "X-Content-Type-Options": "nosniff"},
-    )
-
-
@app.get("/api/records/{record_id}")
 def get_record(record_id: int):
    return {"record": fetch_record(record_id)}


-@app.get("/api/records/{record_id}/export")
-def export_record(record_id: int, request: Request):
-    require_admin_user(request)
-    record = fetch_record(record_id)
-    pdf_path = get_pdf_path(record.get("source_file") or "")
-    if not pdf_path:
-        raise HTTPException(status_code=404, detail="PDF 文件不存在")
-    return FileResponse(
-        pdf_path,
-        media_type="application/pdf",
-        filename=record_pdf_download_name(record),
-        headers={"Cache-Control": "no-store", "X-Content-Type-Options": "nosniff"},
-    )
-
-
@app.get("/api/records/{record_id}/ai-question-image/{log_index}/{module_index}")
 def ai_question_image(record_id: int, log_index: int, module_index: int):
    if log_index < 0 or module_index < 0:
--- a/患者首页处理/数据可视化网页端/app/static/app.js
+++ b/患者首页处理/数据可视化网页端/app/static/app.js
@@ -190,10 +190,6 @@ function userPermissions() {
  return state.currentUser?.permissions || {};
 }

-function isAdminUser() {
-  return state.currentUser?.source === "env" || state.currentUser?.username === "admin";
-}
-
 function canOpenPage(page) {
  const key = page === "auditHistory" ? "audit_history" : page;
  return userPermissions()[key] !== false;
@@ -402,9 +398,6 @@ function renderRecords() {
    const active = record.id === state.selectedId ? " is-active" : "";
    const manual = record.manual_corrected ? " · 人工已改" : "";
    const activity = record.last_activity_at ? `<span class="record-time">最近 ${escapeHtml(formatTime(record.last_activity_at))}</span>` : "";
-    const exportButton = isAdminUser() && record.has_pdf
-      ? `<button class="record-download" type="button" data-export-record="${record.id}" title="下载影像/PDF">下载</button>`
-      : "";
    return `
      <div class="record-item${active}" role="button" tabindex="0" data-id="${record.id}">
        <div class="record-main">
@@ -414,17 +407,13 @@ function renderRecords() {
        <div class="record-sub">
          ${escapeHtml(record.inpatient_no || record.medical_record_no || "")} · ${escapeHtml(record.major_department || record.discharge_dept || "")}${manual}<br>
          ${escapeHtml(record.primary_diagnosis || "")}
-          <div class="record-footer">
-            <span>${activity}</span>
-            ${exportButton}
-          </div>
+          ${activity}
        </div>
      </div>
    `;
  }).join("") + loadingNotice + endNotice;
  list.querySelectorAll(".record-item").forEach((item) => {
-    item.addEventListener("click", (event) => {
-      if (event.target.closest("[data-export-record]")) return;
+    item.addEventListener("click", () => {
      selectRecord(Number(item.dataset.id));
    });
    item.addEventListener("keydown", (event) => {
@@ -433,18 +422,11 @@ function renderRecords() {
      selectRecord(Number(item.dataset.id));
    });
  });
-  list.querySelectorAll("[data-export-record]").forEach((button) => {
-    button.addEventListener("click", (event) => {
-      event.stopPropagation();
-      exportRecordPdf(Number(button.dataset.exportRecord));
-    });
-  });
  list.scrollTop = previousScrollTop;
 }

 function renderFilterActions() {
  const button = $("approveAiNoIssueBtn");
-  const exportButton = $("batchExportBtn");
  if (!button) return;
  const filter = $("statusFilter")?.value || "review_all";
  const aiPassedCount = Number(state.status?.ai_passed || 0);
@@ -452,33 +434,6 @@ function renderFilterActions() {
  button.classList.toggle("is-hidden", !shouldShow);
  button.disabled = !shouldShow || aiPassedCount <= 0 || Boolean(state.bulkJob?.running);
  button.textContent = aiPassedCount > 0 ? `通过所有AI已通过（${aiPassedCount}）` : "通过所有AI已通过";
-  if (exportButton) {
-    exportButton.classList.toggle("is-hidden", !isAdminUser());
-    exportButton.disabled = !isAdminUser() || state.recordsLoading;
-  }
-}
-
-function downloadUrl(url) {
-  const link = document.createElement("a");
-  link.href = url;
-  link.rel = "noopener";
-  document.body.appendChild(link);
-  link.click();
-  link.remove();
-}
-
-function exportRecordPdf(recordId) {
-  if (!isAdminUser() || !recordId) return;
-  downloadUrl(`/api/records/${recordId}/export`);
-}
-
-function exportCurrentFilter() {
-  if (!isAdminUser()) return;
-  const params = new URLSearchParams({
-    q: $("searchInput").value.trim(),
-    status_filter: $("statusFilter").value || "review_all",
-  });
-  downloadUrl(`/api/records/export?${params.toString()}`);
 }

 function scrollSelectedRecordIntoView() {
@@ -711,7 +666,7 @@ function renderTargetSummary(targets, targetId = "targetSummary", formId = "deta
      <ul>${baseNotesHtml}</ul>
    </details>
    ${targets.hasAiNotes ? `
-      <details class="collapsible-panel target-panel is-ai-target">
+      <details class="collapsible-panel target-panel is-ai-target" open>
        <summary>
          <strong>AI建议复核点</strong>
          <span>${aiGroups.length ? `${aiGroups.length} 个模块` : "无待看点"}</span>
@@ -1894,7 +1849,6 @@ async function boot() {
  $("aiAllBtn").addEventListener("click", () => runAiReview("all"));
  $("aiCancelBtn").addEventListener("click", () => handleAiProgressAction().catch((error) => showMessage(error.message, "error")));
  $("approveAiNoIssueBtn").addEventListener("click", approveAiNoIssueRecords);
-  $("batchExportBtn").addEventListener("click", exportCurrentFilter);
  $("pdfZoomOutBtn").addEventListener("click", () => changePdfZoom(-PDF_ZOOM_STEP));
  $("pdfZoomInBtn").addEventListener("click", () => changePdfZoom(PDF_ZOOM_STEP));
  $("pdfZoomResetBtn").addEventListener("click", () => setPdfZoom(100));
--- a/患者首页处理/数据可视化网页端/app/static/index.html
+++ b/患者首页处理/数据可视化网页端/app/static/index.html
@@ -98,7 +98,6 @@
          <button id="aiCancelBtn" type="button">中止AI</button>
        </div>
        <button id="approveAiNoIssueBtn" class="bulk-action is-hidden" type="button">通过所有AI已通过</button>
-        <button id="batchExportBtn" class="bulk-action export-action is-hidden" type="button">批量导出当前筛选</button>
      </div>
      <div class="record-list" id="recordList" tabindex="0"></div>
    </aside>
--- a/患者首页处理/数据可视化网页端/app/static/styles.css
+++ b/患者首页处理/数据可视化网页端/app/static/styles.css
@@ -362,12 +362,6 @@ h2 {
  min-height: 34px;
 }

-.export-action {
-  border-color: #b6c3d0;
-  background: #f7fafc;
-  color: var(--accent-dark);
-}
-
 input,
 select,
 textarea {
@@ -478,32 +472,8 @@ textarea {

 .record-time {
  display: block;
-  color: #7b8794;
-}
-
-.record-footer {
-  min-height: 26px;
-  display: flex;
-  align-items: flex-end;
-  justify-content: space-between;
-  gap: 8px;
  margin-top: 2px;
-}
-
-.record-download {
-  flex: 0 0 auto;
-  min-height: 24px;
-  padding: 2px 8px;
-  border: 1px solid #b6c3d0;
-  border-radius: 5px;
-  background: #fff;
-  color: var(--accent);
-  font-size: 12px;
-}
-
-.record-download:hover {
-  border-color: var(--accent);
-  background: #eef6fb;
+  color: #7b8794;
 }

 .pdf-panel {