The Value of MSCT Based Radiomics in Differential Diagnosis of Borrmann Ⅳ Gastric Cancer and Primary Gastric Lymphoma
10.13471/j.cnki.j.sun.yat-sen.univ（med.sci）.2022.0520
- VernacularTitle:MSCT影像组学对Borrmann Ⅳ型胃癌与原发性胃淋巴瘤的鉴别诊断价值
- Author:
Qin-xian CHEN
1
;
Yu LIU
2
;
Lie-bin HUANG
1
;
Bao FENG
2
;
Hui-min XUE
1
;
Chang-lin LI
2
;
Yong QUAN
3
;
Wan-sheng LONG
1
Author Information
1. Department of Radiology，Jiangmen Central Hospital，Jiangmen 529030，China
2. School of Electronic Information and Automation， Guilin University of Aerospace Technology， Guilin 541004， China
3. Department of Radiology， Zhongshan People's Hospital， Zhongshan 528403， China
- Publication Type:Journal Article
- Keywords:
Borrmann type Ⅳ gastric cancer;
primary gastric lymphoma;
radiomics;
differential diagnosis
- From:
Journal of Sun Yat-sen University(Medical Sciences)
2022;43(5):852-860
- CountryChina
- Language:Chinese
-
Abstract:
ObjectiveTo explore the predictive value of CT based radiomics model in differentiating Borrmann type Ⅳ gastric cancer （GC） from primary gastric lymphoma （PGL）. MethodsA total of 186 cases （Borrmann type Ⅳ GC： 132； PGL： 86） pathologically diagnosed by surgical resection and/or endoscopic biopsy were enrolled from June 2008 to April 2018 retrospectively. Radiomics features were extracted from CT arterial phase and venous phase images by computed algorithm， and selected by least absolute shrinkage and selection operator （Lasso） logistic regression， and then the CT-based radiomics models were established. CT subjective signs were reviewed to build CT subjective signs model， while CT subjective signs and radiomics signature were assembled to build combined model. The receiver operating characteristic （ROC） curve was used to evaluate the performance of CT subjective sign model， radiomics model and the combined model. ResultsTwo signs（the bright line sign of serosa and the irregular nodular protrusion on the serosa side）were selected into the CT subjective sign model. Among the radiomics features， 9 venous phase features， 8 arterial phase features and 14 arteriovenous combination features related to tumor classification were selected， and the corresponding radiomics models were constructed respectively. When the cut-off value of CT subjective sign model was 0.188， the area under curve （AUC） was 0.846， the sensitivity was 61.9%， the specificity was 81.7%， and the accuracy was 76.5%. The cut-off values of arterial phase， venous phase and arteriovenous phase radiomics model were -0.315， -0.669 and -0.858， respectively， and the AUCs were 0.864， 0.955 and 0.890， the sensitivity were 71.4%， 95.2% and 81.0%， the specificity were 85.0%， 88.3% and 80.0%， the accuracy were 81.5%， 90.1% and 80.3%， respectively. The cut-off values of arterial phase， venous phase and arteriovenous phase in the combined model were 0.257， 0.556 and 0.497， respectively， and the AUCs were 0.883， 0.956 and 0.918， the sensitivity was 71.4%， 90.5% and 71.4%， the specificity was 85.0%， 93.3% and 90.0% and the accuracy were 81.5%， 92.6% and 85.2%， respectively. The diagnostic performance of the models from high to low were the combined model， radiomics model and CT subjective finding model （ P< 0.001）， and CT venous phase images were more effective in the differential diagnosis of the two tumors. ConclusionsThe radiomics model based on the arterial and venous phases CT images could differentiate Borrmann type Ⅳ gastric carcinoma from primary gastric lymphoma effectively.
