Using computer‐extracted image phenotypes from tumors on breast magnetic resonance imaging to predict breast cancer pathologic stage

Abstract

Background To demonstrate that computer-extracted image phenotypes (CEIPs) of biopsy-proven breast cancer on MRI can accurately predict pathologic stage. Methods We used a dataset of de-identified breast MRIs organized by the National Cancer Institute in The Cancer Imaging Archive. We analyzed 91 biopsy-proven breast cancer cases with pathologic stage (stage I = 22; stage II = 58; stage III = 11) and surgically proven nodal status (negative nodes = 46, ≥ 1 positive node = 44, no nodes examined = 1). We characterized tumors by (a) radiologist measured size, and (b) CEIP. We built models combining two CEIPs to predict tumor pathologic stage and lymph node involvement, evaluated them in leave-one-out cross-validation with area under the ROC curve (AUC) as figure of merit. Results Tumor size was the most powerful predictor of pathologic stage but CEIPs capturing biologic behavior also emerged as predictive (e.g. stage I+II vs. III demonstrated AUC = 0.83). No size measure was successful in the prediction of positive lymph nodes but adding a CEIP describing tumor “homogeneity,” significantly improved this discrimination (AUC = 0.62, p=.003) over chance. Conclusions Our results indicate that MRI phenotypes show promise for predicting breast cancer pathologic stage and lymph node status.

Joint Prediction of Breast Cancer Histological Grade and Ki-67 Expression Level Based on DCE-MRI and DWI Radiomics

“…Our results are partly consistent with those of the previous study because the Ki-67 index has also been used as a proliferation marker to distinguish between the luminal A and luminal B molecular subtypes of ER-positive breast cancer [50]. The feature of homogeneity, which showed the most discriminative power in a similar study [23], also had relatively high prediction performance in either the Ki-67 or tumor grade prediction task in breast cancer. Additionally, Burnside et al provided effective greatest dimensions and surface areas as important features for prediction [23].…”

“…The feature of homogeneity, which showed the most discriminative power in a similar study [23], also had relatively high prediction performance in either the Ki-67 or tumor grade prediction task in breast cancer. Additionally, Burnside et al provided effective greatest dimensions and surface areas as important features for prediction [23]. Our findings show that morphological features such as maximum tumor diameter, tumor volume, tumor diameter derived from image slices, mean radius, and diameter from tumor volume have high individual performance (AUC larger than 0.70).…”

“…However, to predict tumor characteristics, only a single parametric MRI is usually used. For example, studies indicate that radiomic analysis of MRI may be helpful in predicting breast cancer subtypes [13][14][15][16], neoadjuvant chemotherapy responses [17][18][19], Ki-67 expression level [20][21][22], pathological stage [23], histological grade [24], tumor malignancy [25,26], and breast cancer risk [27] based on either DCE-MRI or DWI. These studies were performed without considering complementary information that could be provided by different parametric images.…”

Unsupervised Clustering of Quantitative Image Phenotypes Reveals Breast Cancer Subtypes with Distinct Prognoses and Molecular Pathways

“…The availability of large-scale curated image and gene expression datasets has spurred a significant interest in linking tumor phenotypes at the molecular and tissue (imaging) level(14,25–34). These studies used a similar study design to identify the correlation of individual imaging features with specific molecular features, such as gene expression, mutation, or predefined molecular subtypes.…”

Imaging-Genomics Study Of Head-Neck Squamous Cell Carcinoma: Associations Between Radiomic Phenotypes And Genomic Mechanisms Via Integration Of TCGA And TCIA

“…Precise and effective treatment of cancer requires the integration of disease information from multiple sources. Imaging-genomics research combines radiographic image analysis with genomic research to improve disease diagnosis and prognosis, discover novel biomarkers, and identify genomic mechanisms associated with phenotype formation [11][12][13][14][15] . Such imaging-genomics studies have been performed for multiple cancer types, including breast invasive carcinoma [11][12][13][14][15] , lung cancer [16][17] , glioblastoma multiforme 18 , and clear cell renal cell carcinoma 19 .…”