Related Experiment Video
Updated: Sep 17, 2025
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Development of a model to predict Ki-67 expression status in non-Hodgkin's lymphoma based on PET radiomics
Ke Tian1, Pengfei Tong2, Keliu Wu3
1Department of Radiology, Air Force Medical Center, Air Force Medical University, Beijing, China.
Introduction:
This study aims to evaluate the effectiveness of conventional metabolic parameters and radiomic features from 18F-deoxyglucose(FDG) PET in predicting Ki-67 expression status in patients with non-Hodgkin's lymphoma.
Methods:
We analyzed clinical, immunohistochemical(IHC), and 18F-FDG PET/CT data from 197 patients diagnosed with non-Hodgkin's lymphoma at our institution between May 2018 and July 2023. Patients were randomly assigned to a training set (60%) and a validation set (40%) to develop PET image-based radiomics, clinical, and combined models. The models' predictive abilities were evaluated using receiver operating characteristic (ROC) curves and a nomogram was created to estimate high Ki-67 expression probabilities.
Results:
Among the patients, 70 exhibited low Ki-67 expression while 127 had high Ki-67 expression (113 males, 84 females, aged 5-85 years). The high Ki-67 group showed a higher proportion of fever(75.9% vs. 24.1%, P < 0.05) and tumor SUV max value/mediastinal SUV max value (T/MB) (P < 0.01). Five radiomic features formed the radiomics score (AUC: training 0.827; validation 0.883). The combined model showed the highest AUC(training 0.921; validation 0.916), indicating strong predictive capability.
Conclusion:
The radiomics model derived from 18F-FDG PET demonstrates superior predictive performance for Ki-67 expression status compared to T/MB. The combined model further improves prediction accuracy, highlighting its potential clinical applicability.
