基于数字病理与术前增强CT的胰腺腺鳞癌跨尺度风险表型构建及术前无创预测

简美诚; 陈成伟; 邵成伟; 边云

doi:10.19732/j.cnki.2096-6210.2026.02.004

您当前的位置：

首页 >

文章列表页 >

基于数字病理与术前增强CT的胰腺腺鳞癌跨尺度风险表型构建及术前无创预测

专题论著 | 更新时间：2026-05-07

- 基于数字病理与术前增强CT的胰腺腺鳞癌跨尺度风险表型构建及术前无创预测
- Construction and preoperative noninvasive prediction of a cross-scale risk phenotype in pancreatic adenosquamous carcinoma using digital pathology and contrast-enhanced CT
- 肿瘤影像学 2026年35卷第2期页码：239-250
- 作者机构：
  
  海军军医大学第一附属医院放射诊断科，上海 200434
- 作者简介：
  
  [ "边　云，影像医学与核医学博士，病理学博士后。现为中国人民解放军海军军医大学第一附属医院（上海长海医院）放射诊断科副主任，党支部副书记，主任医师，教授，博士研究生导师。兼任中国影像医学技术研究会委员，中国老年医学学会放射学分会委员，上海市医师协会放射医师分会人工智能学组副组长，上海市医学会放射科专科分会青年学组委员兼秘书，上海市抗癌协会胰腺癌专业委员会委员，上海市抗癌协会胰腺神经内分泌肿瘤专业委员会委员。主要研究方向是腹部影像学。近5年以第一作者和通信作者在Nature Medicine、Radiology、Nature Communications等国际期刊发表论文90余篇，主持国家及上海市科研课题15项，发明专利20项，获批软件著作19项，主编专著5部。入选上海市东方英才计划、上海长海医院长剑领军人才培养计划。" ]
- 基金信息：
  
  国家自然科学基金(82572215)
- DOI：10.19732/j.cnki.2096-6210.2026.02.004
  中图分类号： R735.9;R445.2
- 收稿：2026-03-27，
  
  修回：2026-04-15，
  
  录用：2026-04-16，
  
  纸质出版：2026-04-28
- 稿件说明：
移动端阅览
简美诚, 陈成伟, 邵成伟, 等. 基于数字病理与术前增强CT的胰腺腺鳞癌跨尺度风险表型构建及术前无创预测［J］. 肿瘤影像学, 2026, 35(2): 239-250.

JIAN M C, CHEN C W, SHAO C WCitation:, et al. Construction and preoperative noninvasive prediction of a cross-scale risk phenotype in pancreatic adenosquamous carcinoma using digital pathology and contrast-enhanced CT［J］. Oncoradiology, 2026, 35(2): 239-250.
简美诚, 陈成伟, 邵成伟, 等. 基于数字病理与术前增强CT的胰腺腺鳞癌跨尺度风险表型构建及术前无创预测［J］. 肿瘤影像学, 2026, 35(2): 239-250. DOI： 10.19732/j.cnki.2096-6210.2026.02.004.

JIAN M C, CHEN C W, SHAO C WCitation:, et al. Construction and preoperative noninvasive prediction of a cross-scale risk phenotype in pancreatic adenosquamous carcinoma using digital pathology and contrast-enhanced CT［J］. Oncoradiology, 2026, 35(2): 239-250. DOI： 10.19732/j.cnki.2096-6210.2026.02.004.

摘要

目的

构建胰腺腺鳞癌（pancreatic adenosquamous carcinoma，PASC）跨尺度风险评估体系。在术后病理学层面，基于全切片图像（whole slide image，WSI）构建鳞状表型病理学指数（squamous phenotype pathology index，SPPI）并评估其患者预后分层价值；在术前影像层面，探索增强计算机体层成像（computed tomography，CT）能否无创预测SPPI高低风险。

方法

回顾并连续收集2014年6月—2024年6月海军军医大学第一附属医院经手术切除且术后病理学检查证实为PASC的患者158例，按时间分为训练集（2014年6月—2021年6月，102例）和验证集（2021年7月—2024年6月，56例）。数字病理学阶段：在训练集中随机选取100张WSI，由病理科医师结合免疫组化逐像素标注鳞癌和腺癌区域（约10 000个256×256补丁），训练DeepLab-v3+自动分割；基于分割结果提取5类患者级表型：成分占比（C）、离散/碎片化（D）、边界复杂度（B）、空间界面关系（S）和肿瘤负荷（V），标准化后纳入Ridge-Cox模型构建连续风险指数SPPI，并以训练集中位数作为分层阈值。术前影像阶段：从158例中筛选术前完成增强CT者129例（训练85例、验证44例），以nnMamba分割门脉期图像并配准至各期相，从肿瘤核心及瘤周感兴趣区（region of interest，ROI）提取213个候选特征，经最小绝对收缩和选择算子（least absolute shrinkage and selection operator，LASSO）算法筛选后联合肿瘤最大径构建预测模型，以冻结的SPPI分层标签为终点。

结果

SPPI每升高1个标准差，训练集死亡风险增加60.9%（HR 1.609，95% CI 1.294~2.002），验证集增加90.3%（HR 1.903，95% CI 1.381~2.622）；对应C-index分别为0.632和0.709。阈值分层后，两队列高风险组中位总生存期（overall survival，OS）均不足10个月（9.53和8.68个月），低风险组分别为18.77和34.21个月。影像方面，LASSO最终保留5个特征，联合模型曲线下面积（area under curve，AUC）在训练集和验证集分别为0.831和0.865。模型预测的高低风险分组同样表现出显著生存差异（均

0.01），且经多因素Cox校正后仍为独立预后因素。

结论

SPPI整合了PASC鳞癌成分的多维空间病理学信息，预后分层效能优于传统单一比例指标。增强CT联合模型能够在术前较可靠地判断SPPI高低风险，初步验证了“影像—病理—预后”两阶段风险评估思路。

Abstract

Objective

To develop a cross-scale risk assessment framework for pancreatic adenosquamous carcinoma (PASC). At the postoperative pathology level

a squamous phenotype pathology index (SPPI) was constructed from whole slide images (WSIs) and its prognostic stratification value was evaluated. At the preoperative imaging level

the ability of contrast-enhanced computed tomography (CT) to noninvasively predict SPPI-defined high versus low risk was investigated.

Methods

A total of 158 patients with surgically resected

pathologically confirmed PASC were retrospectively enrolled from The First Affiliated Hospita

l of Naval Medical University between June 2014 and June 2024. The cohort was split chronologically into a training set (June 2014 to June 2021

=102) and a validation set (July 2021 to June 2024

=56). In the digital pathology stage

100 representative WSIs were randomly sampled from the training set; pathologists annotated squamous and adenocarcinoma regions pixel by pixel with immunohistochemical guidance

yielding approximately 10 000 patches (256×256 pixels) that were used to train a DeepLab-v3+ segmentation model. Five patient-level histological phenotype categories—component proportion (C)

dispersion/fragmentation (D)

boundary complexity (B)

spatial interface relationship (S)

and tumor burden (V)—were extracted from the segmentation output

standardized

and fed into a Ridge-penalized Cox model to produce the continuous risk index SPPI; the training-set median served as the stratification threshold. In the preoperative imaging stage

129 of the 158 patients who had undergone contrast-enhanced CT before surgery (training 85

validation 44) formed an imaging subcohort. Portal-phase images were segmented with nnMamba and registered to the remaining phases; 213 candidate radiomic features were extracted from intratumoral and peritumoral regions of interest

subjected to least absolute shrinkage and selection operator (LASSO) selection

and combined with maximum tumor diameter in a logistic model targeting the frozen SPPI risk labels.

Results

Each one-standard-deviation increase in SPPI was associated with a 60.9% rise in the hazard of death in the training set (HR 1.609

95% CI 1.294–2.002) and a 90.3% rise in the validation set (HR 1.903

95% CI 1.381–2.622)

with corresponding C-indexes of 0.632 and 0.709. After threshold-based stratification

median overall survival (OS) in the high-risk group was below 10 months in both cohorts (9.53 and 8.68 months)

whereas the low-risk group reached 18.77 and 34.21 months

respectively. On the imagin

g side

LASSO retained five features

and the combined model achieved area under curves (AUCs) of 0.831 and 0.865 in the training and validation sets. The model-predicted risk groups likewise showed significant survival separation (both

0.01) and remained independently prognostic after multivariable Cox adjustment.

Conclusion

By integrating multidimensional spatial information on the squamous component

SPPI outperformed the conventional single-proportion metric for prognostic stratification of PASC. The contrast-enhanced CT-based combined model provided reasonably reliable preoperative discrimination of SPPI risk categories

offering preliminary support for a two-stage "imaging-pathology-prognosis" risk assessment paradigm.

关键词

Keywords

references

SIEGEL R L , KRATZER T B , GIAQUINTO A N , et al . Cancer statistics, 2025 ［J］. CA A Cancer J Clin , 2025 , 75 ( 1 ): 10 - 45 .

MCGUIGAN A , KELLY P , TURKINGTON R C , et al . Pancreatic cancer: a review of clinical diagnosis, epidemiology, treatment and outcomes ［J］. World J Gastroenterol , 2018 , 24 ( 43 ): 4846 - 4861 .

BOYD C A , BENARROCH-GAMPEL J , SHEFFIELD K M , et al . 415 patients with adenosquamous carcinoma of the pancreas: a population-based analysis of prognosis and survival ［J］. J Surg Res , 2012 , 174 ( 1 ): 12 - 19 .

KATZ M H G , TAYLOR T H , AL-REFAIE W B , et al . Adenosquamous versus adenocarcinoma of the pancreas: a population-based outcomes analysis ［J］. J Gastrointest Surg , 2011 , 15 ( 1 ): 165 - 174 .

BRAUN R , KLINKHAMMER-SCHALKE M , ZEISSIG S R , et al . Clinical outcome and prognostic factors of pancreatic adenosquamous carcinoma compared to ductal adenocarcinoma: results from the German cancer registry group ［J］. Cancers , 2022 , 14 ( 16 ): 3946 .

NAGTEGAAL I D , ODZE R D , KLIMSTRA D , et al . The 2019 WHO classification of tumours of the digestive system ［J］. Histopathology , 2020 , 76 ( 2 ): 182 - 188 .

XIONG Q L , ZHANG Z W , XU Y F , et al . Pancreatic adenosquamous carcinoma: a rare pathological subtype of pancreatic cancer ［J］. J Clin Med , 2022 , 11 ( 24 ): 7401 .

MOSLIM M A , LEFTON M D , ROSS E A , et al . Clinical and histological basis of adenosquamous carcinoma of the pancreas: a 30-year experience ［J］. J Surg Res , 2021 , 259 : 350 - 356 .

TATSUGUCHI T , KITAHARA D , KOZONO S , et al . Increased proportion of the squamous cell carcinoma components is associated with aggressive behavior and a worse prognosis in resected pancreatic adenosquamous carcinoma ［J］. J Gastrointest Cancer , 2024 , 56 ( 1 ): 5 .

KARDON D E , THOMPSON L D R , PRZYGODZKI R M , et al . Adenosquamous carcinoma of the pancreas: a clinicopathologic series of 25 cases ［J］. Mod Pathol , 2001 , 14 ( 5 ): 443 - 451 .

BORAZANCI E . Adenosquamous carcinoma of the pancreas: Molecular characterization of 23 patients along with a literature review ［J］. World J Gastrointest Oncol , 2015 , 7 ( 9 ): 132 .

NIAZI M K K , PARWANI A V , GURCAN M N . Digital pathology and artificial intelligence ［J］. Lancet Oncol , 2019 , 20 ( 5 ): e253 - e261 .

SAKAMOTO T , FURUKAWA T , LAMI K , et al . A narrative review of digital pathology and artificial intelligence: focusing on lung cancer ［J］. Transl Lung Cancer Res , 2020 , 9 ( 5 ): 2255 - 2276 .

LI Y , CHEN C W , ZHANG Y S , et al . Development and validation of a pathomics model for accurate grading of pancreatic neuroendocrine tumors ［J］. NPJ Precis Oncol , 2025 , 9 ( 1 ): 235 .

PIETRYGA J A , MORGAN D E . Imaging preoperatively for pancreatic adenocarcinoma ［J］. J Gastrointest Oncol , 2015 , 6 ( 4 ): 343 - 357 .

REN S , ZHAO R , CUI W J , et al . Computed tomography-based radiomics signature for the preoperative differentiation of pancreatic adenosquamous carcinoma from pancreatic ductal adenocarcinoma ［J］. Front Oncol , 2020 , 10 : 1618 .

LI Q , LI X Z , LIU W B , et al . Non-enhanced magnetic resonance imaging-based radiomics model for the differentiation of pancreatic adenosquamous carcinoma from pancreatic ductal adenocarcinoma ［J］. Front Oncol , 2023 , 13 : 1108545 .

WESTON B R , BHUTANI M S . Optimizing diagnostic yield for EUS-guided sampling of solid pancreatic lesions: a technical review ［J］. Gastroenterol Hepatol , 2013 , 9 ( 6 ): 352 - 363 .

CRINÒ S F , LARGHI A , BERNARDONI L , et al . Touch imprint cytology on endoscopic ultrasound fine-needle biopsy provides comparable sample quality and diagnostic yield to standard endoscopic ultrasound fine-needle aspiration specimens in the evaluation of solid pancreatic lesions ［J］. Cytopathology , 2019 , 30 ( 2 ): 179 - 186 .

MOONS K G M , ALTMAN D G , REITSMA J B , et al . Transparent Reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): explanation and elaboration ［J］. Ann Intern Med , 2015 , 162 ( 1 ): W1 - W73 .

HESTER C A , AUGUSTINE M M , CHOTI M A , et al . Comparative outcomes of adenosquamous carcinoma of the pancreas: an analysis of the National Cancer Database ［J］. J Surg Oncol , 2018 , 118 ( 1 ): 21 - 30 .

FANG Y , SU Z , XIE J , et al . Genomic signatures of pancreatic adenosquamous carcinoma (PASC) ［J］. J Pathol , 2017 , 243 ( 2 ): 155 - 159 .

BRODY J R , COSTANTINO C L , POTOCZEK M , et al . Adenosquamous carcinoma of the pancreas harbors KRAS2, DPC4 and TP53 molecular alterations similar to pancreatic ductal adenocarcinoma ［J］. Mod Pathol , 2009 , 22 ( 5 ): 651 - 659 .

YANG D W , SUN X L , MONIRUZZAMAN R , et al . Loss of p53 and SMAD4 induces adenosquamous subtype pancreatic cancer in the absence of an oncogenic KRAS mutation ［J］. Cell Rep Med , 2024 , 5 ( 9 ): 101711 .

TANIGAWA M , NAITO Y , AKIBA J , et al . PD-L1 expression in pancreatic adenosquamous carcinoma: PD-L1 expression is limited to the squamous component ［J］. Pathol Res Pract , 2018 , 214 ( 12 ): 2069 - 2074 .

ZHANG Z W , XIONG Q L , XU Y F , et al . The PD-L1 expression and tumor-infiltrating immune cells predict an unfavorable prognosis in pancreatic ductal adenocarcinoma and adenosquamous carcinoma ［J］. J Clin Med , 2023 , 12 ( 4 ): 1398 .

YUAN X H , CHEN C W , SHI Z , et al . Deep learning CT model for stratified diagnosis of pancreatic cystic neoplasms: multicenter development, validation, and real-world clinical impact ［J］. npj Digit Med , 2025 , 8 : 609 .

浏览量

下载量

CNKI被引量

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于增强CT多区域影像组学特征联合临床危险因素构建结肠癌术前脉管侵犯预测模型

探讨CT影像组学模型对非小细胞肺癌免疫治疗效果的预测价值

探讨CT引导下射频消融治疗无法手术治疗的老年非小细胞肺癌患者的疗效及预后

基于CT影像组学联合预测模型对非小细胞肺癌远处转移的预测

胰腺神经内分泌肿瘤CT及MRI影像组学研究进展