Invasive ductal breast cancer molecular subtype prediction by MRI radiomic and clinical features based on machine learning

Abstract

BackgroundMost studies of molecular subtype prediction in breast cancer were mainly based on two-dimensional MRI images, the predictive value of three-dimensional volumetric features from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for predicting breast cancer molecular subtypes has not been thoroughly investigated. This study aimed to look into the role of features derived from DCE-MRI and how they could be combined with clinical data to predict invasive ductal breast cancer molecular subtypes.MethodsFrom January 2019 to December 2021, 190 Chinese women with invasive ductal breast cancer were studied (32 triple-negative, 59 HER2-enriched, and 99 luminal lesions) in this institutional review board-approved retrospective cohort study. The image processing software extracted 1130 quantitative radiomic features from the segmented lesion area, including shape-based, first-order statistical, texture, and wavelet features. Three binary classifications of the subtypes were performed: triple-negative vs. non-triple-negative, HER2-overexpressed vs. non-HER2-overexpressed, and luminal (A + B) vs. non-luminal. For the classification, five machine learning methods (random forest, logistic regression, support vector machine, naïve Bayes, and eXtreme Gradient Boosting) were employed. The classifiers were chosen using the least absolute shrinkage and selection operator method. The area evaluated classification performance under the receiver operating characteristic curve, sensitivity, specificity, accuracy, F1-Score, false positive rate, precision, and geometric mean.ResultsEXtreme Gradient Boosting model showed the best performance in luminal and non-luminal groups, with AUC, sensitivity, specificity, accuracy, F1-Score, false positive rate, precision, and geometric mean of 0.8282, 0.7524, 0.6542, 0.6964, 0.6086, 0.3458, 0.8524 and 0.7016, respectively. Meanwhile, the random forest model showed the best performance in HER2-overexpressed and non-HER2-overexpressed groups, with AUC, sensitivity, specificity, accuracy, F1-Score, false positive rate, precision, and geometric mean of 0.8054, 0.2941, 0.9744, 0.7679, 0.4348, 0.0256, 0.8333 and 0.5353, respectively. Furthermore, eXtreme Gradient Boosting model showed the best performance in the triple-negative and non-triple-negative groups, with AUC, sensitivity, specificity, accuracy, F1-Score, false positive rate, precision, and geometric mean of 0.9031, 0.9362, 0.4444, 0.8571, 0.9167, 0.5556, 0.8980 and 0.6450.ConclusionClinical data and three-dimension imaging features from DCE-MRI were identified as potential biomarkers for distinguishing between three molecular subtypes of invasive ductal carcinomas breast cancer. In the future, more extensive studies will be required to evaluate the findings.

Performance evaluation of ML models for preoperative prediction of HER2-low BC based on CE-CBBCT radiomic features: A prospective study

“…[ 24 , 35 ] To obtain the best prediction model, some scholars have compared the prediction performances of different ML algorithms. [ 24 , 35 , 46 ] Zhang [ 35 ] developed 5 ML models based on the radiomic features of dynamic contrast-enhanced MRI and non-mono-exponential model-based diffusion-weighted imaging to predict HER2+ and HER2− status, including LR, RF, AB, SVM, and LDA. Their findings revealed that both the RF and AB models exhibited superior predictive abilities compared to the other 3 models.…”

“…Their findings revealed that both the RF and AB models exhibited superior predictive abilities compared to the other 3 models. Similarly, Sheng et al [ 46 ] found that the RF model outperformed other ML models by extracting radiomic features from dynamic contrast-enhanced-MRI to predict HER2-overexpressed and non-HER2-overexpressed BC.…”

Radiomic model based on magnetic resonance imaging for predicting pathological complete response after neoadjuvant chemotherapy in breast cancer patients

“…These features were selected based on correlation. Sheng et al (33) combined dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) three-dimensional volume characteristics with clinical data to predict molecular subtypes of invasive ductal breast cancer using t tests and LASSO regression in R language for screening. The final features included shape-based features, texture features, wavelet features, first-order statistical features and the Laplacian of a Gaussian filter.…”

Freehand 1.5T MR-Guided Vacuum-Assisted Breast Biopsy (MR-VABB): Contribution of Radiomics to the Differentiation of Benign and Malignant Lesions

“…Moreover, it has been estimated that around 50–70% of MRI findings requiring biopsy turn out to be non-cancer cases [ 5 ]. In recent years, radiomics and artificial intelligence have been increasingly applied in breast MRI [ 6 ], not only to classify breast lesions or predict response to treatment but also to differentiate benign from malignant lesions [ 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 ]. Indeed, radiological practice includes a qualitative and subjective evaluation of the visible lesion describing different characteristics of the tumor aspect (e.g., the presence of spiculations and the presence of necrosis and microcalcifications).…”