Shao Tingru, Ni Peirong, Wang Chun, Li Jiahui, Lv Xiaozhi
Department of Stomatology, ZhuJiang Hospital of Southern Medical University, 253 Gongye Avenue Middle, Haizhu District, Guangzhou, 510280, China.
Discov Oncol. 2025 Jun 1;16(1):976. doi: 10.1007/s12672-025-02144-8.
The aim of this study is to establish a deep learning model for predicting the pathological grade of oral squamous cell carcinoma(OSCC) based on whole slide images (WSIs). The histopathological images of 257 patients with OSCC were downloaded from TCGA database and randomly divided into training set, validation set and test set at a ratio of 7:2:1. The CPTAC-OSCC dataset, which contains 165 pathological slides of OSCC, was downloaded from the CPTAC database for external validation. The deep learning model constructed by CLAM algorithm achieved an AUC of 0.86 in the training set and an AUC of 0.71 in the external validation set. The results showed that the model had good prediction efficiency and good generalization ability. In addition, deep learning features were extracted from the model, and the prognostic risk model of OSCC was further constructed combined with transcriptome data to explore the differences in gene expression levels and immune cell infiltration levels between high and low risk groups. The deep learning model can accurately predict the pathological grade of patients with OSCC, which provides certain reference value for clinical diagnosis. The nomogram combining transcriptome data and patient clinical characteristics can be used as a prognostic classifier for clinical decision making and treatment.
本研究的目的是基于全切片图像（WSIs）建立一种用于预测口腔鳞状细胞癌（OSCC）病理分级的深度学习模型。从TCGA数据库下载了257例OSCC患者的组织病理学图像，并按照7:2:1的比例随机分为训练集、验证集和测试集。从CPTAC数据库下载了包含165张OSCC病理切片的CPTAC - OSCC数据集用于外部验证。由CLAM算法构建的深度学习模型在训练集中的AUC为0.86，在外部验证集中的AUC为0.71。结果表明该模型具有良好的预测效率和泛化能力。此外，从模型中提取深度学习特征，并结合转录组数据进一步构建OSCC的预后风险模型，以探索高风险组和低风险组之间基因表达水平和免疫细胞浸润水平的差异。该深度学习模型能够准确预测OSCC患者的病理分级，为临床诊断提供了一定的参考价值。结合转录组数据和患者临床特征的列线图可作为临床决策和治疗的预后分类器。
