Correlative Analysis of FFDM and DCE-MRI for Improved Breast CADx

Verification of lesion correspondence in images from different modalities is an essential but nontrivial step in the multi-modality computer-aided diagnosis (CADx) of breast cancer. This study extends the correlative feature analysis (CFA) method to lesions that are imaged by both full-field digital mammography (FFDM) and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI), and incorporates the CFA method into a lesion characterization procedure. Computerized algorithms are firstly applied to segment lesions and extract features in images. For each pair-wise feature, a Bayesian artificial neural network (BANN) is employed to yield a correspondence score. A subset of correspondence scores is then automatically selected and merged by another BANN to yield an overall probability of correspondence (PC). Depending on the overall PC value, the probability of malignancy of a lesion is estimated by either a single-modality CADx or multi-modality CADx. Receiver operating characteristic (ROC) analysis is used to evaluate the performance of lesion correspondence identification and lesion classification. In the task of identifying corresponding lesions in images from different modalities, the PC values in the MRI vs. craniocaudal (CC) category yielded an area under the curve (AUC) of 0.78±0.03, those in the MRI vs. mediolateral oblique (MLO) category yielded an AUC of 0.86 ± 0.03, and those in the MRI vs. mediolateral (ML) category yielded an AUC of 0.84±0.02, The use of multiple feature-based correspondence scores statistically improves performance compared to the of a single correspondence score (overall α = 0.05). With CFA included in the task of distinguishing malignant lesions from benign lesions, the proposed scheme yielded an AUC of 0.84±0.03 for lesions imaged with MR, and an AUC of 0.82±0.03 for lesions imaged with FFDM. The results of this pilot study indicate that multi-modality CADx with CFA has the potential to provide improved diagnostic accuracy.

[1]  M. Giger,et al.  Computerized interpretation of breast MRI: investigation of enhancement-variance dynamics. , 2004, Medical physics.

[2]  Maryellen L Giger,et al.  Potential effect of different radiologist reporting methods on studies showing benefit of CAD. , 2008, Academic radiology.

[3]  M L Giger,et al.  Computerized analysis of breast lesions in three dimensions using dynamic magnetic-resonance imaging. , 1998, Medical physics.

[4]  Robert M. Haralick,et al.  Textural Features for Image Classification , 1973, IEEE Trans. Syst. Man Cybern..

[5]  M. Giger,et al.  Volumetric texture analysis of breast lesions on contrast‐enhanced magnetic resonance images , 2007, Magnetic resonance in medicine.

[6]  Maryellen L. Giger,et al.  Ideal observer approximation using Bayesian classification neural networks , 2001, IEEE Transactions on Medical Imaging.

[7]  C A Roe,et al.  Statistical Comparison of Two ROC-curve Estimates Obtained from Partially-paired Datasets , 1998, Medical decision making : an international journal of the Society for Medical Decision Making.

[8]  Christopher M. Bishop,et al.  Neural networks for pattern recognition , 1995 .

[9]  M. Giger,et al.  Automatic identification and classification of characteristic kinetic curves of breast lesions on DCE-MRI. , 2006, Medical physics.

[10]  Kenji Suzuki,et al.  A dual-stage method for lesion segmentation on digital mammograms. , 2007, Medical physics.

[11]  Hon J. Yu,et al.  Selection of diagnostic features on breast MRI to differentiate between malignant and benign lesions using computer-aided diagnosis: differences in lesions presenting as mass and non-mass-like enhancement , 2009, European Radiology.

[12]  Li Lan,et al.  Classification of breast lesions with multimodality computer-aided diagnosis: observer study results on an independent clinical data set. , 2006, Radiology.

[13]  H. Chan,et al.  Multi-modality CADx: ROC study of the effect on radiologists' accuracy in characterizing breast masses on mammograms and 3D ultrasound images. , 2009, Academic radiology.

[14]  A. Jemal,et al.  Cancer Statistics, 2009 , 2009, CA: a cancer journal for clinicians.

[15]  Yulei Jiang,et al.  An ROC comparison of four methods of combining information from multiple images of the same patient. , 2004, Medical physics.

[16]  M. Giger,et al.  Automated computerized classification of malignant and benign masses on digitized mammograms. , 1998, Academic radiology.

[17]  C. Metz,et al.  Maximum likelihood estimation of receiver operating characteristic (ROC) curves from continuously-distributed data. , 1998, Statistics in medicine.

[18]  M. Giger,et al.  Multimodality computerized diagnosis of breast lesions using mammography and sonography. , 2005, Academic radiology.

[19]  Maryellen L. Giger,et al.  A Fuzzy C-Means (FCM)-Based Approach for Computerized Segmentation of Breast Lesions in Dynamic Contrast-Enhanced MR Images1 , 2006 .

[20]  Maryellen L. Giger,et al.  Identifying Corresponding Lesions from CC and MLO Views Via Correlative Feature Analysis , 2008, Digital Mammography / IWDM.

[21]  Yading Yuan,et al.  Correlative feature analysis on FFDM. , 2008, Medical physics.

[22]  Laurent D. Cohen,et al.  A new Image Registration technique with free boundary constraints: application to mammography , 2003, Comput. Vis. Image Underst..

[23]  Maryellen L Giger,et al.  Prevalence scaling: applications to an intelligent workstation for the diagnosis of breast cancer. , 2008, Academic radiology.