口腔医学研究

口腔医学研究 ›› 2023, Vol. 39 ›› Issue (4): 308-315.DOI: 10.13701/j.cnki.kqyxyj.2023.04.005

• 口腔肿瘤学研究 • 上一篇    下一篇

cyclin D1表达模式深度学习识别和预测模型对HPV阴性口腔鳞状细胞癌和口咽鳞状细胞癌患者预后的预测作用

杨可1,2, 孙雅楠1,2, 胡雅英1,2, 吕宜楠1,2, 郑小风1,2, 李易玮1,2, 张佳莉1,2*   

  1. 1.武汉大学口腔医院口腔病理科 湖北 武汉 430079;
    2.口腔基础医学省部共建国家重点实验室培育基地和口腔生物医学教育部重点实验室武汉大学口腔医学院 湖北 武汉 430079
  • 收稿日期:2022-10-08 出版日期:2023-04-28 发布日期:2023-04-19
  • 通讯作者: *张佳莉,E-mail: jiali_zhang@whu.edu.cn
  • 作者简介:杨可(1995~ ),男,浙江绍兴人,硕士在读,主要从事口腔肿瘤临床的研究。
  • 基金资助:
    国家自然科学基金(编号:81972552)

Prediction of Prognosis of HPV-negative Oral and Oropharyngeal Squamous Cell Carcinoma by Deep Learning Identification and Prediction Model of Cyclin D1 Expression Pattern

YANG Ke1,2, SUN Yanan1,2, HU Yaying1,2, LV Yinan1,2, ZHEN Xiaofeng1,2, LI Yiwei1,2, ZHANG Jiali1,2*   

  1. 1. Department of Oral Histopathology, Hospital of Stomatology, Wuhan University, Wuhan 430079, China;
    2. The State Key Laboratory Breeding Base of Basic Science of Stomatology (Hubei_MOST) & Key Laboratory of Oral Biomedicine Ministry of Education, School of Stomatology, Wuhan University, Wuhan 430079, China
  • Received:2022-10-08 Online:2023-04-28 Published:2023-04-19

PDF (PC)

139

摘要： 目的: 通过评估cyclin D1的表达状态与HPV阴性口腔鳞状细胞癌(oral squamous cell carcinoma,OSCC)及口咽鳞状细胞癌(oropharyngeal squamous cell carcinoma,OPSCC)患者预后的关系,建立基于cyclin D1表达模式的图像识别评分和生存预测模型。方法: 回顾性分析610例HPV阴性OSCC和OPSCC患者的临床病理资料。通过比较cyclin D1在不同评价方式以及联合p16/p53表达等多种因素评价体系下患者总体生存(OS)及无进展生存(PFS)的差异,检测cyclin D1对OSCC、OPSCC患者预后的预测作用。利用YOLOv5图像识别算法建立cyclinD1表达模式评分模型, 并在此基础上用DeepHit和DeepSurv算法分别建立预后模型。结果: cyclin D1在OSCC和OPSCC癌巢中存在三级表达模式。该表达模式与OSCC(P<0.0001)、OPSCC(P<0.05)患者预后显著相关,且优于cyclin D1表达水平与患者预后的相关性。cyclin D1表达模式在OSCC(P<0.0001)及OPSCC(P<0.05)中均为独立的预后风险因素。基于cyclin D1表达模式的评分模型在测试集中的准确率达(78.48±4.31)%。在OS预测中,DeepHit算法建立的模型在测试集中的C-index为0.709±0.019,DeepSurv算法建立模型测试集中C-index为0.715±0.029。结论: 基于深度学习的cyclin D1表达模式评分模型联合生存预测模型对HPV阴性OSCC和OPSCC总体生存预后具有较好的预测作用。

关键词: cyclin D1, 口腔鳞状细胞癌, 口咽鳞状细胞癌, 预后, 深度学习

Abstract: Objective: To evaluate the relationship between the expression of cyclin D1 and the prognosis of patients with HPV-negative oral squamous cell carcinoma (OSCC) and oropharyngeal squamous cell carcinoma (OPSCC), and to establish the image recognition scoring and survival prediction models based on cyclin D1 expression pattern. Methods: The clinicopathological data of 610 patients with HPV negative OSCC and OPSCC were analyzed retrospectively. The differences of overall survival (OS) rate and progression-free survival (PFS) rate of patients under different evaluation methods of cyclin D1 combining with p16/p53 expression and other factors were compared. The image recognition model to scoring cyclin D1 expression pattern was established by YOLOv5 algorithm. On this basis, the survival prediction model was established by DeepHit and DeepSurv algorithms, respectively. Results: There were three expression patterns of cyclin D1 in OSCC and OPSCC cancer nests. Superior to the expression level scoring method, the expression pattern scoring of cyclin D1 was significantly correlated with the prognosis of patients with OSCC (P<0.0001) and OPSCC (P<0.05). And it was an independent prognostic risk factor in both OSCC (P<0.0001) and OPSCC (P<0.05). Based on cyclin D1 expression pattern score model, the average accuracy of the test sets was (78.48±4.31)%. In OS prediction models established by DeepHit algorithm, the C-index of test set was 0.709±0.019, and in the models established by DeepSurv algorithm, the C-index of test set reached 0.715±0.029. Conclusion: Based on image recognition model of cyclin D1 expression pattern, the survival prediction model has a relatively good prediction effect on OS prognosis of HPV-negative OSCC and OPSCC.

Key words: cyclin D1, oral squamous cell carcinoma, oropharyngeal squamous cell carcinoma, prognosis, deep learning