Objective To use the features extracted from magnetic resonance(MRI)images and machine learning methods to help distinguish the molecular subtypes of breast cancer.Methods The clinical data of 178 patients with breast cancer diagnosed in our hospital from October 2021 to October 2023 were retrospectively analyzed.The shape,MRI features and histogram based features of each patient's tumor were extracted from the MRI images before and after three times of enhancement using internal software.Simultaneously collect clinical and pathological data.Identify important imaging features based on machine learning models and establish models for predicting IDC subtypes.Using the left one method cross validation(LOOCV)to avoid overfitting of the model,Kruskal Wallis test was used to determine statistical significance.Results The LOOCV process generated a model with different features,with 11 out of the top 20 features showing significant differences between IDC subtypes(P＜0.05).Combining the first nine pathological and imaging features,the prediction model has an accuracy of 83.4％in identifying IDC subtypes.The accuracy of the combined pathological and imaging models for each subtype was 89.2％(ERPR1),63.6％(ERPR-/HER21),and 82.5％(TN),respectively.When only the first nine imaging features are combined,the overall accuracy of the prediction model in identifying IDC subtypes on LOOCV is 71.2％.The accuracy of the combined pathological and imaging models for each subtype was 69.9％(ERPR1),62.9％(ERPR-/HER21),and 81.0％(TN),respectively.Conclusion We have developed a machine learning based prediction model that uses features extracted from MRI to distinguish IDC subtypes with significant predictive ability.
关键词
磁共振/乳腺癌分子亚型/留一法交叉验证
Key words
Magnetic Resonance/Molecular Subtypes of Breast Cancer/Cross Validation with Retention Method
维普
万方数据