Preoperative predictive value of machine learning model based on MRI omics characteristics for molecular classification of invasive breast cancer
GULIJIRE·Tailaiti
HARELEHASHI·Anmantai
WANG Yingying
HUO Zhong
Abstract：Objective To evaluate the efficacy of a machine learning model constructed based on MRI radiomics features in predicting the molecular subtyping of invasive breast cancer before surgery.Methods A retrospective analysis was conducted on 155 cases of invasive breast cancer patients who underwent radical mastectomy at our hospital from January 2023 to June 2024.Patients were divided into three groups according to postoperative immunohistochemical molecular subtyping results:Luminal type(including Luminal A and Luminal B),HER2-overexpressing type,and triple-negative type.Clinical pathological and MRI sign data of the three groups were compared;the lesions in the second phase of MRI images of all patients were manually segmented to extract ROI radiomics feature variables,and LASSO regression was used for feature variable dimensionality reduction and screening.All sample data were divided into training and testing sets in a 7:3 ratio;random forest,logistic regression,and support vector machine,three machine learning models,were used to model the screened radiomics feature variables,and ROC curves were plotted to evaluate the efficacy of different machine learning models in preoperative assessment of invasive breast cancer molecular subtyping.Results There was no statistical significance in general clinical pathological data and imaging indicators among the three groups(P>0.05);the confusion matrix showed that the diagnostic sensitivity,specificity,and accuracy of the random forest model were 96.55%,93.33%,and 93.62%,respectively,significantly higher than those of the logistic regression and support vector machine models(P<0.05);ROC curve analysis showed that the random forest model had the highest predictive AUC of 0.903(95%CI:0.815-0.958),significantly higher than the logistic regression model's 0.680(95%CI:0.565-0.782)and the support vector machine model's 0.693(95%CI:0.579-0.793),with statistically significant differences(P<0.05).Conclusion The random forest model constructed based on MRI radiomics features and machine learning algorithms has high efficacy in predicting the molecular subtyping of invasive breast cancer before surgery.
Keywords：machine learning modelradiomics featuresinvasive breast cancermolecular subtyping
Publication Date：2025-06-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 699-705 )