Artificial intelligence model based on radiomic features from preoperative CT for predicting Ki-67 expression in adrenocortical carcinoma
CHAO Yin-yao
DENG Yi-shu
CHEN Xian-da
MA Nan
ZHU Hong-zhang
YU Kai-lin
KE Zong-pan
XIAO Jun
GUO Sheng-jie
Abstract：Objective To establish an artificial intelligence(AI)model based on radiomic features extracted from preoperative non-contrast CT images to predict Ki-67 expression levels in adrenocortical carcinoma(ACC).Meth-ods A total of 93 patients diagnosed pathologically with primary ACC and who underwent adrenalectomy were retrospec-tively included from the First Affiliated Hospital of Sun Yat-sen University,Sun Yat-sen University Cancer Center,the First Affiliated Hospital of the University of Science and Technology of China,and the TCIA database between July 2010 and September 2023.Patients were divided into internal training and external validation cohorts.A Ki-67 index＞10％was defined as high Ki-67.Radiomic features were extracted and selected,and five AI algorithms were applied to build predictive models.Model performance was evaluated using the area under the receiver operating characteristic curve(AUC).Results There were no significant differences in clinical baseline characteristics such as age and tumor size a-mong patients.After feature selection,six radiomic features were retained for model development.In the internal training cohort,the gradient boosting algorithm achieved the highest AUC(0.94),with an accuracy of 0.87,sensitivity of 0.89,and specificity of 0.84.In the external validation cohort,the random forest algorithm showed the best performance,with an AUC of 0.86,accuracy of 0.87,and sensitivity of 0.88.Conclusion Radiomic features from preoperative non-con-trast CT can effectively predict Ki-67 expression in adrenocortical carcinoma,offering a potential non-invasive biomark-er for clinical decision-making.
Keywords：adrenocortical carcinomaradiomicsKi-67artificial intelligencemulticenter retrospective study
Publication Date：2025-06-15
Online Publishing Date：2025-08-18(First online date of this platform, not the publication date of the document)
Pages：6( 839-844 )