补充缺页判断

photo_review/auto_photo_review.py

@@ -21,7 +21,8 @@ from services.paddle_services import IE_KEY
 from ucloud import ufile
 from util import image_util, common_util, html_util, model_util
 from util.data_util import handle_date, handle_decimal, parse_department, handle_name, handle_insurance_type, \
-    handle_original_data, handle_hospital, handle_department, handle_id, handle_age, parse_money, parse_hospital
+    handle_original_data, handle_hospital, handle_department, handle_id, handle_age, parse_money, parse_hospital, \
+    parse_page_num, handle_tiny_int
 
 
 # 尝试从二维码中获取高清图片
@@ -124,6 +125,7 @@ def information_extraction(phrec, pk_phhd, identity):
             info_extract = model_util.ie_settlement(
                 better_img_path, common_util.ocr_result_to_layout(model_util.ocr(better_img_path))
             )
+        ocr_text = None  # 此处肯定不是出院记录，后续用不到
     else:
         target_image = model_util.det_book(img_path)  # 识别文档区域并裁剪
         dewarped_image = model_util.dewarp(target_image)  # 去扭曲
@@ -158,7 +160,7 @@ def information_extraction(phrec, pk_phhd, identity):
                                creator=HOSTNAME, update_time=now, updater=HOSTNAME))
         session.commit()
         session.close()
-    return rec_type, info_extract
+    return rec_type, info_extract, ocr_text
 
 
 # 从keys中获取准确率最高的value
@@ -395,6 +397,9 @@ def cost_task(pk_phhd, cost_list_ie_result):
     cost_data["admission_date"] = handle_date(cost_data["admission_date_str"])
     cost_data["discharge_date"] = handle_date(cost_data["discharge_date_str"])
     cost_data["medical_expenses"] = handle_decimal(cost_data["medical_expenses_str"])
+    page_nums, page_count = parse_page_num(cost_list_ie_result[IE_KEY['page']])
+    cost_data['page_nums'] = handle_original_data(','.join(page_nums))
+    cost_data['page_count'] = handle_tiny_int(page_count)
     save_or_update_ie(ZxIeCost, pk_phhd, cost_data)
     return cost_data
 
@@ -416,19 +421,24 @@ def photo_review(pk_phhd, name):
     session.close()
     # 同一批图的标识
     identity = int(time.time())
+    discharge_text = ''
     for phrec in phrecs:
-        rec_type, ie_result = information_extraction(phrec, pk_phhd, identity)
+        rec_type, ie_result, ocr_text = information_extraction(phrec, pk_phhd, identity)
         if rec_type == '基本医保结算单':
             rec_result = settlement_result
         elif rec_type == '出院记录':
             rec_result = discharge_result
+            discharge_text += ocr_text
         elif rec_type == '费用清单':
             rec_result = cost_result
         else:
             rec_result = None
         if rec_result is not None:
             for key, value in ie_result.items():
-                rec_result[key] += value
+                if key == '页码':
+                    rec_result[key].append(value)  # 页码要区分来源，所以多包一层
+                else:
+                    rec_result[key] += value
 
     settlement_data = settlement_task(pk_phhd, settlement_result)
     discharge_data = discharge_task(pk_phhd, discharge_result)
@@ -447,8 +457,26 @@ def photo_review(pk_phhd, name):
     review_result['has_discharge'] = bool(discharge_result)
     review_result['has_cost'] = bool(cost_result)
     # 三项资料缺页判断
-    # TODO:缺页需要对页码进行抽取，暂未训练相关模型
+    page_description = []
-    review_result['full_page'] = True
+    # todo:关键词需根据实际情况调整
+    discharge_key = ['入院诊断', '入院日期', '出院日期', '出院诊断', '入院情况', '诊疗经过', '出院情况', '出院医嘱']
+    if not all(key in discharge_text for key in discharge_key):
+        page_description.append('《出院记录》缺页')
+
+    cost_missing_page = {}
+    if cost_data['page_nums']:
+        page_nums = cost_data['page_nums'].split(',')
+        required_set = set(range(1, cost_data['page_count'] + 1))
+        page_set = set(page_nums)
+        cost_missing_page = required_set - page_set
+    if cost_missing_page:
+        page_description.append(f"《住院费用清单》，缺第{','.join(cost_missing_page)}页")
+
+    if page_description:
+        review_result['full_page'] = False
+        review_result['page_description'] = ';'.join(page_description)
+    else:
+        review_result['full_page'] = True
 
     if (review_result['has_settlement'] and review_result['has_discharge'] and review_result['has_cost']
             and review_result['full_page']):

services/paddle_services/__init__.py

@@ -16,5 +16,6 @@ IE_KEY = {
     'admission_id': '住院号',
     'settlement_id': '医保结算单号码',
     'age': '年龄',
-    'upper_case_medical_expenses': '大写总额'
+    'upper_case_medical_expenses': '大写总额',
+    'page': '页码',
 }

services/paddle_services/ie_cost.py

@@ -10,7 +10,7 @@ from utils import process_request
 
 app = Flask(__name__)
 COST_LIST_SCHEMA = tuple(IE_KEY[key] for key in [
-    'name', 'admission_date', 'discharge_date', 'medical_expenses'
+    'name', 'admission_date', 'discharge_date', 'medical_expenses', 'page'
 ])
 COST = Taskflow('information_extraction', schema=COST_LIST_SCHEMA, model='uie-x-base',
                 task_path='model/cost_list_model', layout_analysis=False, precision='fp16')

util/data_util.py

@@ -178,3 +178,22 @@ def parse_hospital(string):
     split_hospitals = string_without_company.replace('医院', '医院 ')
     result += split_hospitals.strip().split(' ')
     return result
+
+
+def parse_page_num(page_list):
+    if not page_list:
+        return None, None
+    pages = []
+    total = []
+    for page in page_list:
+        join = ''.join(page)
+        numbers = re.findall(r'\d+', join)
+        pages.append(min(numbers))
+        total.append(max(numbers))
+    return pages, max(total)
+
+
+def handle_tiny_int(num):
+    if not num:
+        return None
+    return num if num <= 127 else 127