Discrimination values of CT radiomics combined with machine learning algorithms for tumor deposits,lymph node metastasis and their simultaneous occurrence in colorectal cancer
10.19745/j.1003-8868.2024216
- VernacularTitle:CT影像组学联合机器学习算法在鉴别结直肠癌肿瘤沉积、淋巴结转移及二者同时发生中的价值
- Author:
Yue TENG
1
;
Yuan XU
;
Shu DING
Author Information
1. 盐城市第一人民医院影像科,江苏盐城 224001
- Keywords:
radiomics;
machine learning;
colorectal cancer;
tumor deposit;
lymph node metastasis
- From:
Chinese Medical Equipment Journal
2024;45(11):60-66
- CountryChina
- Language:Chinese
-
Abstract:
Objective To investigate the values of CT radiomics combined with machine learning algorithms for discriminating tumor deposits(TDs),lymph node metastasis(LNM)and their simultaneous occurrence in colorectal cancer(CRC).Methods Totally 261 CRC patients treated at some hospital from January 2017 to December 2023 had their clinical data analyzed retrospectively,who were classified into a TDs-positive group(TDs+group,64 cases),a LNM-positive group(LNM+group,99 cases)and a TDs-positive combined with LNM-positive group(TDs+combined with LNM+group,98 cases)according to A JCC 8th Edition:Colorectal Cancer.MaZda version 4.6 software was used to outline the region of interest,and then the texture feature parameters with the most discriminative value were selected by the screening method coming with this software.Texture feature parameters with statistically significant differences in one-way analysis(P＜0.05)were included to draw ROC curves,and the diagnostic efficacy of the included parameters was evaluated.The diagnostic efficacy of four machine learning algorithms such as Bayesian,decision tree,random forest and Logistic regression for the TDs+group,the LNM+group and the TDs+combined with LNM+group was assessed based on the included parameters and ten-fold cross-validation.Statistical analysis was performed using SPSS 26.0,MedCalc 19.1.3,Weka 3.8.6 and R language software.Results Texture feature parameters with statistically significant(P＜0.05)one-way ANOVA differences among the 3 groups included skewness,S(2,-2)and mean value,S(0,4)correlation,S(3,3)difference variance and S(0,4)inverse moments.The S(3,3)difference variance had the highest diagnostic efficacy in the TDs+group and the LNM+group,and the skewness did in the TDs+combined with LNM+group.Random forest algorithm behaved the best in diagnostic efficacy when compared with Bayesian,decision tree and Logistic regression algorithms,which had the accuracy rates being 0.897,0.830 and 0.861 and the AUCs being 0.951,0.957 and 0.958 respectively in the TDs+group,the LNM+group and the TDs+combined with LNM+group.Conclusion CT radiomics combined with random forest algorithm is of high value for discriminating TDs,LNM and their simultaneous occurrence in CRC patients,and thus can provide comprehensive diagnosis information for clinicians and help to formulate accurate treatment strategies for patients.[Chinese Medical Equipment Journal,2024,45(11):60-66]
