18F-FDG PET radiomics score for treatment response and prognosis prediction in patients with primary gastrointestinal diffuse large B-cell lymphoma
10.3760/cma.j.cn321828-20240910-00319
- VernacularTitle:18F-FDG PET影像组学评分预测原发性胃肠道DLBCL患者的治疗反应和预后
- Author:
Jincheng ZHAO
1
;
Jian RONG
;
Yue TENG
;
Man CHEN
;
Jianxin CHEN
;
Jingyan XU
Author Information
1. 中国药科大学南京鼓楼医院血液内科，南京 210008
- Publication Type:Journal Article
- Keywords:
Lymphoma, large B-cell, diffuse;
Gastrointestinal tract;
Radiomics;
Positron-emission tomography;
Fluorodeoxyglucose F18
- From:
Chinese Journal of Nuclear Medicine and Molecular Imaging
2025;45(12):726-731
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the value of a cross-combination machine learning approach in constructing a PET radiomics score (RadScore) for predicting early treatment response and prognosis in patients with primary gastrointestinal diffuse large B-cell lymphoma (PGI-DLBCL).Methods:This retrospective cohort study was conducted on 108 patients (59 males and 49 females, age (55.6±12.1) years) diagnosed with PGI-DLBCL between November 2016 and December 2021 at Nanjing Drum Tower Hospital, Affiliated Hospital of Medical School, Nanjing University ( n=85) and West China Hospital, Sichuan University ( n=23). Patients were divided into a training set ( n=86) and a validation set ( n=22) with the ratio of 8∶2 using stratified random sampling method. Seven machine learning models were employed to generate 49 feature selection-classification candidates, and the optimal candidate was selected to construct the RadScore, with five-fold cross-validation applied to determine the best-performing model. Logistic regression analysis was performed to identify risk factors for early treatment response, and a radiomics nomogram was developed by integrating RadScore with clinical predictors. Survival results between different groups of RadScore was compared by log-rank test. Results:Nineteen predictive features were selected from 111 radiomic features to construct the RadScore. In the training set, lactate dehydrogenase (LDH) (odds ratio ( OR)=3.53, 95% CI: 1.21-10.31, P=0.021), intestinal involvement ( OR=3.04, 95% CI: 1.04-8.88, P=0.042), total lesion glycolysis (TLG; OR=6.73, 95% CI: 2.23-20.29, P<0.001) and RadScore ( OR=15.11, 95% CI: 3.95-57.80, P<0.001) were identified as independent risk factors for predicting early treatment response. The combined model integrating RadScore, LDH, intestinal involvement, and TLG demonstrated good discriminatory ability for early treatment response (AUC=0.860 in the training set; AUC=0.902 in the validation set). Significant differences were observed in progression-free survival (PFS) and overall survival (OS) between different RadScore groups ( χ2 values: 13.92 and 8.56, both P<0.01). Conclusions:The machine learning-based RadScore may effectively predict survival outcomes in patients with PGI-DLBCL. The combined model integrating RadScore, clinical factors, and metabolic indicators can predict early treatment response in PGI-DLBCL patients.
