From the Department of Radiology (J.G.N., C.M.P., J.M.G., H.K.), Department of Thoracic and Cardiovascular Surgery (S.P., Y.T.K.), and Department of Pathology (Y.K.J., D.H.C.), Seoul National University Hospital and College of Medicine, 101 Daehak-ro, Jongno-gu, Seoul 03080, Republic of Korea; Artificial Intelligence Collaborative Network, Seoul National University Hospital, Seoul, Republic of Korea (J.G.N.); Institute of Radiation Medicine (C.M.P., J.M.G.) and Institute of Medical and Biological Engineering (C.M.P.), Seoul National University Medical Research Center, Seoul, Republic of Korea; and Cancer Research Institute, Seoul National University, Seoul, Republic of Korea (Y.K.J., J.M.G., Y.T.K.).
Radiology. 2022 Nov;305(2):441-451. doi: 10.1148/radiol.213262. Epub 2022 Jul 5.
Background A preoperative CT-based deep learning (DL) prediction model was proposed to estimate disease-free survival in patients with resected lung adenocarcinoma. However, the black-box nature of DL hinders interpretation of its results. Purpose To provide histopathologic evidence underpinning the DL survival prediction model and to demonstrate the feasibility of the model in identifying patients with histopathologic risk factors through unsupervised clustering and a series of regression analyses. Materials and Methods For this retrospective study, data from patients who underwent curative resection for lung adenocarcinoma without neoadjuvant therapy from January 2016 to September 2020 were collected from a tertiary care center. Seven histopathologic risk factors for the resected adenocarcinoma were documented: the aggressive adenocarcinoma subtype (cribriform, morular, solid, or micropapillary-predominant subtype); mediastinal nodal metastasis (pN2); presence of lymphatic, venous, and perineural invasion; visceral pleural invasion (VPI); and mutation status. Unsupervised clustering using 80 DL model-driven CT features was performed, and associations between the patient clusters and the histopathologic features were analyzed. Multivariable regression analyses were performed to investigate the added value of the DL model output to the semantic CT features (clinical T category and radiologic nodule type [ie, solid or subsolid]) for histopathologic associations. Results A total of 1667 patients (median age, 64 years [IQR, 57-71 years]; 975 women) were evaluated. Unsupervised patient clusters 3 and 4 were associated with all histopathologic risk factors ( < .01) except for mutation status ( = .30 for cluster 3). After multivariable adjustment, model output was associated with the aggressive adenocarcinoma subtype (odds ratio [OR], 1.03; 95% CI: 1.002, 1.05; = .03), venous invasion (OR, 1.03; 95% CI: 1.004, 1.06; = .02), and VPI (OR, 1.08; 95% CI: 1.06, 1.10; < .001), independently of the semantic CT features. Conclusion The deep learning model extracted CT imaging surrogates for the histopathologic profiles of lung adenocarcinoma. © RSNA, 2022 . See also the editorial by Yanagawa in this issue.
背景 术前基于深度学习（DL）的预测模型被提出，用于估计接受肺腺癌切除术患者的无病生存率。然而，DL 的黑盒性质阻碍了对其结果的解释。目的 提供 DL 生存预测模型的组织病理学依据，并通过无监督聚类和一系列回归分析证明该模型识别具有组织病理学危险因素患者的可行性。材料与方法 这项回顾性研究纳入了 2016 年 1 月至 2020 年 9 月在一家三级护理中心接受无新辅助治疗的肺腺癌切除术的患者的数据。记录了七种与切除腺癌相关的组织病理学危险因素：侵袭性腺癌亚型（筛状、微团状、实体或微乳头状优势型）；纵隔淋巴结转移（pN2）；淋巴管、静脉和神经周围侵犯；脏层胸膜侵犯（VPI）；和 突变状态。使用 80 个 DL 模型驱动的 CT 特征进行无监督聚类，分析患者聚类与组织病理学特征之间的关联。进行多变量回归分析，以研究 DL 模型输出对语义 CT 特征（临床 T 分期和放射学结节类型[即实性或部分实性]）与组织病理学关联的附加价值。结果 共评估了 1667 例患者（中位年龄 64 岁[IQR，5771 岁]；975 例女性）。无监督患者聚类 3 和 4 与所有组织病理学危险因素相关（ <.01），但 突变状态除外（聚类 3 为.30）。多变量调整后，模型输出与侵袭性腺癌亚型（比值比[OR]，1.03；95%CI：1.0021.05； =.03）、静脉侵犯（OR，1.03；95%CI：1.0041.06； =.02）和 VPI（OR，1.08；95%CI：1.061.10； <.001）相关，独立于语义 CT 特征。结论 DL 模型提取了 CT 成像替代物，用于肺腺癌的组织病理学特征。©RSNA，2022。也可参见本期 Yanagawa 的述评。
