Appearance Constrained Semi-Automatic Segmentation from DCE-MRI is Reproducible and Feasible for Breast Cancer Radiomics: A Feasibility Study

Harini Veeraraghavan,Meredith Sadinski,Elizabeth J Sutton,Natsuko Onishi,Elizabeth Morris,Brittany Z Dashevsky,Joseph O Deasy

doi:10.1038/s41598-018-22980-9

Harini Veeraraghavan, Meredith Sadinski + Show 5 more

Open Access

https://doi.org/10.1038/s41598-018-22980-9

Copy DOI

Abstract

We present a segmentation approach that combines GrowCut (GC) with cancer-specific multi-parametric Gaussian Mixture Model (GCGMM) to produce accurate and reproducible segmentations. We evaluated GCGMM using a retrospectively collected 75 invasive ductal carcinoma with ERPR+ HER2− (n = 15), triple negative (TN) (n = 9), and ER-HER2+ (n = 57) cancers with variable presentation (mass and non-mass enhancement) and background parenchymal enhancement (mild and marked). Expert delineated manual contours were used to assess the segmentation performance using Dice coefficient (DSC), mean surface distance (mSD), Hausdorff distance, and volume ratio (VR). GCGMM segmentations were significantly more accurate than GrowCut (GC) and fuzzy c-means clustering (FCM). GCGMM’s segmentations and the texture features computed from those segmentations were the most reproducible compared with manual delineations and other analyzed segmentation methods. Finally, random forest (RF) classifier trained with leave-one-out cross-validation using features extracted from GCGMM segmentation resulted in the best accuracy for ER-HER2+ vs. ERPR+/TN (GCGMM 0.95, expert 0.95, GC 0.90, FCM 0.92) and for ERPR + HER2− vs. TN (GCGMM 0.92, expert 0.91, GC 0.77, FCM 0.83).

Highlights

We present a segmentation approach that combines GrowCut (GC) with cancer-specific multiparametric Gaussian Mixture Model (GCGMM) to produce accurate and reproducible segmentations
We evaluated the reproducibility of manual delineations produced by multiple users using six consecutive cases with two from estrogen positive (ER)-HER2+, two from ERPR + HER2− and two from triple negative cancers to benchmark segmentation performance
GCGMM segmentations were more accurate compared with GC and fuzzy c-means clustering (FCM) methods for both mild and marked background parenchymal enhancements (Table 1), and for cancers that presented as masses

Summary

Introduction

We present a segmentation approach that combines GrowCut (GC) with cancer-specific multiparametric Gaussian Mixture Model (GCGMM) to produce accurate and reproducible segmentations. Accurate and reasonably fast segmentation is critical for radiomics analysis[6] which consists of extracting image features from large datasets with the purpose of identifying non-invasive image-based surrogates for diagnosis (differentiating disease aggressiveness) and for predicting treatment response. Semi-automatic segmentations including GrowCut (GC)[17] have been reported to produce more reproducible texture features compared with features computed from manually delineated lung tumors[18], thereby, underscoring the importance and utility of computer-generated segmentations for high-throughput radiomics. Repetitive interactions resulting either from the algorithm itself which present as queries or from users can become time consuming for high-throughput radiomics analysis This in turn limits the applicability of such methods for high-throughput analysis in comparison to fully automatic methods such as unsupervised fuzzy clustering[26]

Objectives

Methods

Results

Discussion

Conclusion