Automated localization of breast cancer in DCE-MRI

doi:10.1016/j.media.2014.12.001

Medical Image Analysis

Volume 20, Issue 1, February 2015, Pages 265-274

https://doi.org/10.1016/j.media.2014.12.001 Get rights and content

Highlights

•
A computer-aided detection system for breast cancer in DCE-MRI is proposed.
•
The system was trained to detect mass-like and non-mass-like malignant lesions.
•
Performance is evaluated on a dataset containing abnormal and normal cases.
•
The CAD system reaches 89% sensitivity at 4 false positives per normal case.

Abstract

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is increasingly being used for the detection and diagnosis of breast cancer. Compared to mammography, DCE-MRI provides higher sensitivity, however its specificity is variable. Moreover, DCE-MRI data analysis is time consuming and depends on reader expertise. The aim of this work is to propose a novel automated breast cancer localization system for DCE-MRI. Such a system can be used to support radiologists in DCE-MRI analysis by marking suspicious areas. The proposed method initially corrects for motion artifacts and segments the breast. Subsequently, blob and relative enhancement voxel features are used to locate lesion candidates. Finally, a malignancy score for each lesion candidate is obtained using region-based morphological and kinetic features computed on the segmented lesion candidate. We performed experiments to compare the use of different classifiers in the region classification stage and to study the effect of motion correction in the presented system. The performance of the algorithm was assessed using free-response operating characteristic (FROC) analysis. For this purpose, a dataset of 209 DCE-MRI studies was collected. It is composed of 95 DCE-MRI studies with 105 breast cancers (55 mass-like and 50 non-mass-like malignant lesions) and 114 DCE-MRI studies from women participating in a screening program which were diagnosed to be normal. At 4 false positives per normal case, 89% of the breast cancers (91% and 86% for mass-like and non-mass-like malignant lesions, respectively) were correctly detected.

Graphical Abstract

Introduction

Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) of the breast is employed for breast cancer screening in women with cumulative lifetime breast cancer risk of more than 20–25% (US and EU guideline) (Mann et al., 2008). Other indications include, but are not limited to, preoperative staging, evaluation of women treated with neoadjuvant chemotherapy and problem solving in case of inconclusive findings from other modalities. Compared to mammography, which is the image modality commonly used in regular screening, breast DCE-MRI presents higher sensitivity (Peters et al., 2008), especially in women with dense breasts (Pisano et al., 2008). However, specificity is more variable, both for screening (Mann et al., 2008) and characterization (Peters et al., 2008) purposes, since the examination of breast DCE-MRI depends on many factors such as reader expertise and use of adequate visualization techniques. This fact can lead to a substantial amount of false positive findings. Furthermore, breast MRI analysis requires interpretation of four-dimensional DCE data, as well as correlation to multi-parametric data from other MRI imaging sequences, and is therefore a time consuming task.

In order to help overcome these limitations, dedicated workstations are currently used in clinical practice to assist the radiologist in the detection and classification of breast lesions in DCE-MRI data. These systems mainly provide an automated kinetic assessment by color-coding the intensity changes per voxel during enhancement of the breast tissue. This automated assessment aids in the interpretation of patterns of contrast enhancement (persistent, plateau and washout enhancement) across a series of MRI volumes (Dorrius et al., 2011), but human interaction is still required to identify and characterize suspicious areas, which increases the risk of misinterpreting or overlooking breast lesions (Pages et al., 2012, Yamaguchi et al., 2013) and may cause inter- and intra-observer variability (Newell et al., 2010).

A computer-aided detection (CAD) system that marks the most suspicious locations of the breast can aid radiologists in the analysis of breast DCE-MRI. Such a system can reduce the interpretation time of analyzing breast DCE-MRI data and, as shown when CAD has been applied to other modalities (Li et al., 2008, White et al., 2009, Kligerman et al., 2013, Burhenne et al., 2000), otherwise missed lesions could be detected.

Only a few authors have presented algorithms to automatically detect lesions in breast DCE-MRI (Vignati et al., 2011, Renz et al., 2012, Chang et al., 2014). The method developed by Vignati et al. (2011) used subtracted mean intensity projection images over time, which were normalized using the contrast uptake of the mammary vessel. Renz et al. (2012) evaluated a fully automatic CAD system that segmented lesions with a hierarchical 3D Gaussian pyramid approach. More recently, Chang et al. (2014) combined kinetic and morphological features to identify focal tumor breast lesions. However, in these previous studies, results were reported for relatively small datasets that do not cover the entire spectrum of malignant breast lesions since non-mass-like enhancing lesions, which are lesions with abnormal enhancement larger than focus without space-occupying effect (Morris et al., 2014), and thus without blob-like shape, are not included. Therefore, automatic lesion detection in DCE-MRI is still an open problem. The problem is also clinically relevant since recent studies showed that lesions are regularly overlooked or misinterpreted in breast cancer screening programs with MRI (Pages et al., 2012, Yamaguchi et al., 2013).

In this work, we propose a novel CAD system to automatically detect breast lesions in DCE-MRI. The system is a multi-stage approach that uses blob features in combination with kinetic and morphological information of the lesion in motion corrected data. The performance of the algorithm was assessed using free-response operating characteristic (FROC) analysis, using a dataset composed of 95 studies with manually annotated malignant mass-like and non-mass-like lesions and 114 studies from patients participating in a screening program for high and intermediate risk women that were evaluated as normal.

Section snippets

Study dataset

The dataset used in this study contained 209 T1-weighted coronal DCE-MRI studies from different patients (age: 22–81 years, mean age: 48.5 years). Of the complete dataset, 114 DCE-MRI studies were obtained from patients participating in a high-risk screening program. These MRIs were scored as either BI-RADS 1 ( $n = 87$ ) or BI-RADS 2 ( $n = 27$ ). For each included MRI, at least 2 years of follow up were available with no sign of breast cancer and no previous history of breast cancer or breast surgery was

Results

Results of the first experiment can be observed in Fig. 6, which shows the performance of the proposed CAD system using different classifiers in the final region classification stage. Table 2 shows p-values for the comparisons between the classifiers. For the detection of malignant lesions, LDA, kNN, gentleboost, SVM and random forest obtained mean sensitivity values in the range between 0.1 and 4 FPs/case of 0.48, 0.60, 0.66, 0.64 and 0.72, respectively. The random forests classifier

Discussion

In this work, we developed a CAD system for breast cancer in DCE-MRI that provides automated localization of malignant lesions on motion corrected data. First, relative enhancement and blob features are used to detect suspicious areas. Then, the final malignancy likelihood is obtained using region-based morphological and kinetic information computed on segmented lesion candidates. The use of different classifiers in the final region classification stage and the effect of applying motion

Conclusion

In this paper we have presented a fully automatic algorithm for the detection of breast cancer in DCE-MRI. We introduced blob features in an initial voxel candidate detection stage. Morphological and kinetic features are employed by a second step to reduce false positives. Evaluation on a dataset of 209 DCE-MRI cases, which contained 114 DCE-MRI studies representative for normal patients participating in a screening program, showed that our approach yields high sensitivity for the detection of

Acknowledgments

The research leading to these results has received funding from the European Unions Seventh Framework Programme FP7 under Grant Agreement No. 306088. Albert Gubern-Mérida held a FPU Grant AP2009-2835.

References (45)

S. Behrens et al.
Computer assistance for MR based diagnosis of breast cancer: present and future challenges
Comput. Med. Imaging Graph
(2007)
Y.C. Chang et al.
Classification of breast mass lesions using model-based analysis of the characteristic kinetic curve derived from fuzzy c-means clustering
J. Magn. Reson. Imaging
(2012)
W. Chen et al.
A fuzzy c-means (FCM)-based approach for computerized segmentation of breast lesions in dynamic contrast-enhanced MR images
Acad. Radiol.
(2006)
A.B. Ashraf et al.
A multichannel markov random field framework for tumor segmentation with an application to classification of gene expression-based breast cancer recurrence risk
IEEE Trans. Med. Imaging
(2013)
L. Breiman
Random forests
Machine Learning
(2001)
L.J.W. Burhenne et al.
Potential contribution of computer-aided detection to the sensitivity of screening mammography
Radiology
(2000)
Chang, C.C., Lin, C.J., 2001. LIBSVM: a library for support vector machines....
Y.C. Chang et al.
Computerized breast lesions detection using kinetic and morphologic analysis for dynamic contrast-enhanced MRI
J. Magn. Reson. Imaging
(2014)
W. Chen et al.
Automatic identification and classification of characteristic kinetic curves of breast lesions on DCE-MRI
Med. Phys.
(2006)
E.W.L. Chuwa et al.
Early detection of breast cancer through population-based mammographic screening in asian women: a comparison study between screen-detected and symptomatic breast cancers
Breast J.
(2009)

M.D. Dorrius et al.

Computer-aided detection in breast MRI: a systematic review and meta-analysis

Eur. Radiol.

(2011)

B. Efron

Bootstrap methods: another look at the jackknife

Ann. Stat.

(1979)

J. Friedman et al.

Special invited paper. Additive logistic regression: a statistical view of boosting

Ann. Stat.

(2000)

C. Gallego Ortiz et al.

Automatic atlas-based segmentation of the breast in MRI for 3D breast volume computation

Med. Phys.

(2012)

K.G. Gilhuijs et al.

Computerized analysis of breast lesions in three dimensions using dynamic magnetic-resonance imaging

Med. Phys.

(1998)

K.G.A. Gilhuijs et al.

Breast MR imaging in women at increased lifetime risk of breast cancer: clinical system for computerized assessment of breast lesions initial results

Radiology

(2002)

Gubern-Mérida, A., Wang, L., Kallenberg, M., Martí, R., Hahn, H., Karssemeijer, N., 2013. Breast segmentation in MRI:...

A. Gubern-Mérida et al.

Breast segmentation and density estimation in breast MRI: a fully automatic framework

IEEE J. Biomed. Health Inform.

(2014)

S. Klein et al.

elastix: a toolbox for intensity-based medical image registration

IEEE Trans. Med. Imaging

(2010)

S. Kligerman et al.

The effect of computer-aided detection on radiologist performance in the detection of lung cancers previously missed on a chest radiograph

J. Thorac. Imaging

(2013)

C.K. Kuhl

The coming of age of nonmammographic screening for breast cancer

JAMA

(2008)

G.D. Leonard et al.

Ductal carcinoma in situ, complexities and challenges

J. Natl. Cancer Inst.

(2004)

Cited by (122)

Kinetic Heterogeneity Improves the Specificity of Dynamic Enhanced MRI in Differentiating Benign and Malignant Breast Tumours
2024, Academic Radiology
To investigate whether kinetic heterogeneity in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) improves the specificity of breast cancer (BC) diagnosis.
The DCE-MRI data of patients with benign breast tumours and BC from June 2020 to July 2022 were retrospectively evaluated. MATLAB and SPM were used to determine six major kinetic parameters: peak, enhancement volume, heterogeneity, as well as persistent, plateau, and washout proportions. Continuous variables were compared using the Student's t-test or Mann–Whitney U tests, and categorical variables were compared using the chi-square or Fisher's exact tests. Receiver operating characteristic curves were plotted. The intraclass correlation coefficient (ICC) was used to evaluate agreement between the two observers. Multivariate logistic regression analysis was conducted to calculate the odds ratios (ORs) with 95% confidence intervals (CIs) for the association between benign and malignant breast tumours.
In total, 147 patients (mean age, 47 years old) were included in the study, 76 of whom had BC. Data analysis by the two observers showed good consistency in the peak, enhancement volume, persistent proportion, plateau proportion, washout proportion, and heterogeneity, with ICCs of 0.865, 0.988, 0.906, 0.940, 0.740, and 0.867, respectively (p < 0.001). In the DCE kinetic analysis, differences in all the six kinetic parameters were statistically significant (p < 0.05). The area under the curve for heterogeneity was 0.92 (95% CI:0.88,0.97), and the sensitivity and specificity were 0.895 and 0.845, respectively. Multivariate logistic regression analysis showed that heterogeneity was an independent predictor of BC compared to benign breast tumours (OR = 2.020; 95% CI:1.316, 3.100; p = 0.001).
The kinetic heterogeneity of DCE-MRI can effectively distinguish between benign and malignant breast tumours and improve the specificity of BC diagnosis.
BreastDM: A DCE-MRI dataset for breast tumor image segmentation and classification
2023, Computers in Biology and Medicine
Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has shown high sensitivity to diagnose breast cancer. However, few computer-aided algorithms focus on employing DCE-MR images for breast cancer diagnosis due to the lack of publicly available DCE-MRI datasets. To address this issue, our work releases a new DCE-MRI dataset called BreastDM for breast tumor segmentation and classification. In particular, a dataset of 232 patients selected with DCE-MR images for benign and malignant cases is established. Each case consists of three types of sequences: pre-contrast, post-contrast, and subtraction sequences. To show the difficulty of breast DCE-MRI tumor image segmentation and classification tasks, benchmarks are achieved by state-of-the-art image segmentation and classification algorithms, including conventional hand-crafted based methods and recently-emerged deep learning-based methods. More importantly, a local–global cross attention fusion network (LG-CAFN) is proposed to further improve the performance of breast tumor images classification. Specifically, LG-CAFN achieved the highest accuracy (88.20%, 83.93%) and AUC value (0.9154,0.8826) in both groups of experiments. Extensive experiments are conducted to present strong baselines based on various typical image segmentation and classification algorithms. Experiment results also demonstrate the superiority of the proposed LG-CAFN to other breast tumor images classification methods. The related dataset and evaluation codes are publicly available at smallboy-code/Breast-cancer-dataset.
Joint-phase attention network for breast cancer segmentation in DCE-MRI
2023, Expert Systems with Applications
Breast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays an important role in the screening and treatment evaluation of high-risk breast cancer. The segmentation of cancerous regions is an essential step for the comprehensive analysis of breast MRI. Nevertheless, automatic and robust segmentation is still very challenging because of the large differences of cancers in size, morphology, and intensity appearance. To alleviate these issues, we propose a simple yet effective two-stage approach, which simultaneously exploits pre- and post-contrast enhanced information to segment the breast cancer. In particular, we first offer a breast segmentation network to predict the breast region of interest (ROI) and therefore excluding confounding information from thorax region in the whole MRI scans. Moreover, inspired by the radiologists’ examination routine which takes full advantage of the MRI sequences to make the diagnosis, we suggest a joint-phase attention network in order to mine both pre- and post-contrast representations for the segmentation of cancerous regions. The accuracy and generalizability of the proposed network is validated on our collected DCE-MRI dataset containing 550 subjects (with 748 biopsy-proven breast cancers) from 3 different centers (one as internal dataset and two as external datasets). The primary evaluation metrics are Dice similarity coefficient (Dice), Jaccard index (Jaccard), and average symmetric surface distance (ASSD). Our network consistently achieves satisfactory segmentation results, by generating an average Dice of 88.77%/82.77%/83.03%, Jaccard of 81.27%/71.89%/73.23%, and ASSD of 2.21/3.63/2.69, on one internal and two external datasets, respectively. Our method offers an effective cancer segmentation approach for the breast DCE-MRI examination. The code is publicly available at https://github.com/ryandok/JPA.
Three-dimensional affinity learning based multi-branch ensemble network for breast tumor segmentation in MRI
2022, Pattern Recognition
Accurate and automatic breast tumor segmentation based on dynamic contrast-enhancement magnetic resonance imaging (DCE-MRI) plays an important role in breast cancer analysis. However, the background parenchymal enhancement and large variations in tumor size, shape or appearance make the task very challenging, and also the segmentation performance is still not satisfactory, especially for non-mass enhancement (NME) and small size tumors ( $\leq$ 2 cm). To address these challenges, we propose a novel 3D affinity learning based multi-branch ensemble network for accurate breast tumor segmentation. Specifically, two different types of subnetworks are built to form a multi-branch network. The two subnetworks are equipped with effective operation components, i.e., residual connection and channel-wise attention or making use of dense connectivity patterns, which can process the input images in parallel. Second, we propose an end-to-end trainable 3D affinity learning based refinement module by calculating the similarities between features of voxels, which is useful in discovering more pixels belonging to breast tumors. Third, two local affinity matrices are constructed by 3D affinity learning, which are used to refine the segmentation outputs of two subnetworks, respectively. Finally, a novel ensemble module is proposed to combine the information derived from the subnetworks, which can hierarchically merge the local and global affinity matrices derived from subnetworks. A large-scale breast DCE-MR images dataset with 420 subjects are built for evaluation, and comprehensive experiments have been conducted to demonstrate that our proposed method achieves superior performance over state-of-the-art medical image segmentation methods.
Stacking approach for accurate Invasive Ductal Carcinoma classification
2022, Computers and Electrical Engineering
The accurate diagnosis of Breast cancer (BC) requires adequately exploiting Artificial intelligence (AI)-based methods in the diagnosing process. To tackle the issue of accurate BC diagnosis, we have proposed a deep learning-based stacking method (StackBC). In particular, we have incorporated deep learning models including Convolutional Neural Network (CNN), Long Short Term Memory (LSTM), and Gated Recurrent Unit (GRU) for the classification of Invasive Ductal Carcinoma (IDC). Additionally, Transfer Learning (TL) and Data Augmentation (DA) approaches have been incorporated to balance the dataset and adequately train the model. To further improve the predictive outputs of the model, we used the stacking technique. Among the three individual base classifiers, the performance of the GRU model was better. Hence, we selected the GRU as a meta classifier to distinguish between Non-IDC and IDC breast images. The experimental results confirmed that the StackBC method outperformed state-of-the-art methods.
Diffusion-Weighted Imaging (DWI) for Breast Lesion Characterization: The Olea Medical Perspective and the Utilization of Olea Sphere Software
2022, Diffusion MRI of the Breast
Magnetic resonance imaging (MRI) plays an increasingly significant role in diagnosis and follow-up of breast lesions. A myriad of sequences are now available to provide various biomarkers for lesion characterization and grading to supplement mammographic findings. More recently, along with dynamic-contrast enhanced (DCE) and T2-weighted imaging, diffusion-weighted imaging (DWI) has emerged as a contrast-agent-free technique allowing to discriminate benign from malignant lesions. However, the lack of consensus regarding acquisition protocols requires significant efforts from medical and scientific communities toward standardization. The absence of guidelines leaves the vendor uncertain as to what characteristics in their software product would be the most applicable for the market. In the meanwhile, until such consensus is reached, vendors can only offer a large spectrum of metrics to allow multiparametric analysis. Olea Sphere combined with breastscape offers a large panel of advanced postprocessing for diffusion, permeability, relaxometry, and texture to facilitate breast lesion characterization and, hopefully, contribute to the standardization of clinical practice for the utmost benefit to patients. In this chapter, the authors describe the various DWI models implemented within Olea Sphere, with specific focus on the clinical interest of such advanced metrics, based on the proprietary Bayesian postprocessing method.

View all citing articles on Scopus

View full text

Automated localization of breast cancer in DCE-MRI

Highlights

Abstract

Graphical Abstract

Introduction

Section snippets

Study dataset

Results

Discussion

Conclusion

Acknowledgments

Comput. Med. Imaging Graph

J. Magn. Reson. Imaging

Acad. Radiol.

A multichannel markov random field framework for tumor segmentation with an application to classification of gene expression-based breast cancer recurrence risk

IEEE Trans. Med. Imaging

Random forests

Machine Learning

Potential contribution of computer-aided detection to the sensitivity of screening mammography

Radiology

Computerized breast lesions detection using kinetic and morphologic analysis for dynamic contrast-enhanced MRI

J. Magn. Reson. Imaging

Automatic identification and classification of characteristic kinetic curves of breast lesions on DCE-MRI

Med. Phys.

Early detection of breast cancer through population-based mammographic screening in asian women: a comparison study between screen-detected and symptomatic breast cancers

Breast J.

Computer-aided detection in breast MRI: a systematic review and meta-analysis

Eur. Radiol.

Bootstrap methods: another look at the jackknife

Ann. Stat.

Special invited paper. Additive logistic regression: a statistical view of boosting

Ann. Stat.

Automatic atlas-based segmentation of the breast in MRI for 3D breast volume computation

Med. Phys.

Computerized analysis of breast lesions in three dimensions using dynamic magnetic-resonance imaging

Med. Phys.

Breast MR imaging in women at increased lifetime risk of breast cancer: clinical system for computerized assessment of breast lesions initial results

Radiology

Breast segmentation and density estimation in breast MRI: a fully automatic framework

IEEE J. Biomed. Health Inform.

elastix: a toolbox for intensity-based medical image registration

IEEE Trans. Med. Imaging

The effect of computer-aided detection on radiologist performance in the detection of lung cancers previously missed on a chest radiograph

J. Thorac. Imaging

The coming of age of nonmammographic screening for breast cancer

JAMA

Ductal carcinoma in situ, complexities and challenges

J. Natl. Cancer Inst.