RSNA 2017

Abstract Archives of the RSNA, 2017


BR245-SD-WEA1

Contrast Texture-Derived MRI Radiomics Correlate with Breast Cancer Clinico-Pathological Prognostic Factors

Wednesday, Nov. 29 12:15PM - 12:45PM Room: BR Community, Learning Center Station #1



Participants
Samir Kulkarni, MD, Dallas, TX (Abstract Co-Author) Nothing to Disclose
Yin Xi, Dallas, TX (Abstract Co-Author) Nothing to Disclose
Ramapriya Ganti, MD, Dallas, TX (Abstract Co-Author) Nothing to Disclose
Matthew A. Lewis, PhD, Dallas, TX (Abstract Co-Author) Research collaborations, CMR Naviscan Corporation and QT Ultrasound Labs
Robert E. Lenkinski, PhD, Dallas, TX (Abstract Co-Author) Research Grant, Koninklijke Philips NV Research Consultant, Aspect Imaging
Basak E. Dogan, MD, Dallas, TX (Presenter) Nothing to Disclose

For information about this presentation, contact:

basak.dogan@utsouthwestern.edu

PURPOSE

To identify the potential role of MRI radiomics as a virtual prognostic biomarker for breast cancer Estrogen Receptor (ER), HER2 expression, tumor grade, molecular subtype and T stage.

METHOD AND MATERIALS

In an IRB approved, HIPAA compliant study, consecutive patients with primary invasive breast cancer who underwent dynamic contrast-enhanced (DCE) breast MRI between July 2013 and July 2016 in our institution were retrospectively reviewed. Patient age, T size, grade, ER and HER2 status, Ki-67(%) were recorded. DCE images were segmented and Haralick texture features were extracted using a custom developed pyOsiriX script. Bootstrap Lasso (Bolasso) feature selection method was used to select a small subset of optimal texture features. The final selected model consisted of features that were present in at least 80% of the bootstrap replications. Their classification performance was assessed with the area under receiver operating characteristic curve (AUC) with leave-one-out cross-validation.

RESULTS

Two hundred and twenty patients were analyzed. Median age was 49 (range 21-79). Tumor size was T1 in 50(23%), T2 in 96 (43.6%), T3 in 49 (22.2%) and T4 in 25(11.4%). Tumor grade was 1-2 in 112(51%) and 3 in 96(43.6%) patients. Sensitivity, specificity, PPV, NPV and accuracy of our radiomics model for differentiating T1/2 vs T3/4 was 65%, 88%, 74.5%, 82%, 80%, while HER2(+) vs HER2(-) ,69%, 64%, 39%, 86%, 65.3% (AUC=0.69 [95%CI 0.61, 0.77]), high nuclear grade (Grade 3) vs low grade (Grades 1-2) was 70.4%,72.7%, 92.7%, 33.3% , 70.8% (AUC=0.71 [95%CI 0.62, 0.80]); and ER (+) vs ER(-) status (AUC=0.57 [95%CI 0.48, 0.66]) and ki67<=14% vs ki67>14%(AUC= 0.66 [95%CI 0.56, 0.77]). Radiomics performance for distinguishing each molecular subtype from others was: HER2 enriched 0.7 [95%CI 0.6, 0.7], Triple-negative 0.63 [95%CI 0.5, 0.7]), Luminal A 0.61 [95%CI 0.5,0.7] and Luminal B 0.57 [95%CI 0.4, 0.6].

CONCLUSION

Quantitative radiomics using MRI contrast texture has moderate diagnostic performance in predicting poor breast cancer clinicopathological prognostic factors of advanced stage, high nuclear grade, high ki-i67 and HER2-enriched molecular subtype.

CLINICAL RELEVANCE/APPLICATION

Quantitative radiomics shows promise for future use as a clinical prognostic tool, which can potentially help with clinical management decisions.