Vural Topuz Özge, Bağbudar Sidar, Aksu Ayşegül, Söylemez Akkurt Tuçe, Akkaş Burcu Esen
Department of Nuclear Medicine, Başakşehir Cam and Sakura City Hospital, University of Health Sciences, Istanbul, Turkey.
Department of Pathology, Başakşehir Cam and Sakura City Hospital, University of Health Sciences, Istanbul, Turkey.
Nuklearmedizin. 2025 Jun;64(3):194-204. doi: 10.1055/a-2512-8212. Epub 2025 Jan 28.
To determine the value of radiomics data extraction from baseline 18F FDG PET/CT in the prediction of tumor-infiltrating lymphocytes (TILs) among patients with primary breast cancer (BC).We retrospectively evaluated 74 patients who underwent baseline 18F FDG PET/CT scans for BC evaluation between October 2020 and April 2022. Radiomics data extraction resulted in a total of 131 radiomic features from primary tumors. TILs status was defined based on histological analyses of surgical specimens and patients were categorized as having low TILs or moderate & high TILs. The relationships between TILs groups and tumor features, patient characteristics and molecular subtypes were examined. Features with a correlation coefficient of less than 0.6 were analyzed by logistic regression to create a predictive model. The diagnostic performance of the model was calculated via receiver operating characteristics (ROC) analysis.Menopausal status, histological grade, nuclear grade, and four radiomics features demonstrated significant differences between the two TILs groups. Multivariable logistic regression revealed that nuclear grade and three radiomics features (Morphological COMShift, GLCM Correlation, and GLSZM Small Zone Emphasis) were independently associated with TIL grouping. The diagnostic performance analysis of the model showed an AUC of 0.864 (95% CI: 0.776-0.953; p < 0.001). The sensitivity, specificity, PPV, NPV and accuracy values of the model were 69.6%, 82.4%, 64%, 85.7% and 78.4%, respectivelyThe pathological TIL scores of BC patients can be predicted by using radiomics feature extraction from baseline 18F FDG PET/CT scans.
为了确定从基线18F FDG PET/CT中提取的影像组学数据在预测原发性乳腺癌（BC）患者肿瘤浸润淋巴细胞（TILs）方面的价值。我们回顾性评估了2020年10月至2022年4月期间接受基线18F FDG PET/CT扫描以评估BC的74例患者。影像组学数据提取共产生了来自原发性肿瘤的131个影像组学特征。TILs状态基于手术标本的组织学分析来定义，患者被分类为低TILs或中高TILs。研究了TILs组与肿瘤特征、患者特征和分子亚型之间的关系。对相关系数小于0.6的特征进行逻辑回归分析以创建预测模型。通过受试者操作特征（ROC）分析计算模型的诊断性能。绝经状态、组织学分级、核分级和四个影像组学特征在两个TILs组之间表现出显著差异。多变量逻辑回归显示核分级和三个影像组学特征（形态学COMShift、灰度共生矩阵相关性和灰度游程长度矩阵小区域强调）与TIL分组独立相关。模型的诊断性能分析显示AUC为0.864（95%CI：0.776 - 0.953；p < 0.001）。模型的敏感性、特异性、阳性预测值、阴性预测值和准确性分别为69.6%、82.4%、64%、85.7%和78.4%。通过从基线18F FDG PET/CT扫描中提取影像组学特征可以预测BC患者的病理TIL评分。
