Total variation based DCE‐MRI decomposition by separating lesion from background for time‐intensity curve estimation. Issue 6 (22nd May 2017)

• Record Type: Journal Article
• Authors: Liu, Hui; Zheng, Yuanjie; Liang, Dong; Tang, Pinpin; Ren, Fuquan; Zhang, Lina; Zhao, Zuowei
• Is Part Of: Medical physics. Volume 44:Issue 6(2017)
• Page Start: 2321
• Page End: 2331
• Publication Date: 2017-05-22
• Subjects: Bregman iteration -- DCE‐MRI -- decomposition algorithm -- time intensity curve -- tracer‐kinetic model; Medical physics -- Periodicals; Medical physics; Geneeskunde; Natuurkunde; Toepassingen; Biophysics; Periodicals; Periodicals; Electronic journals; 610.153
• DOI: 10.1002/mp.12242
• Languages: English
• ISSNs: 0094-2405

Abstract

Abstract : Purpose: This study aims to obtain the accurate time intensity curve (TIC) of a dynamic contrast‐enhanced magnetic resonance image (DCE‐MRI) by eliminating the normal tissue enhancement and obtaining pure lesion information. The TIC of DCE‐MRI is sometimes distorted because of the influence of normal tissue. In this paper, a new tracer‐kinetic modeling based on total variation (DC‐TV) is proposed to address this problem by decomposing the DCE‐MRI into the normal tissue image and the lesion image. As TIC generation is not standardized and a credible program is expected, an accurate TIC generation is presented in this paper. Materials and methods: We propose a new tracer‐kinetic model DC‐TV to decompose the lesion region in breast DCE‐MRIs. The original image is decomposed into a normal tissue image and a lesion image to obtain the pure lesion enhancement information. The acquired lesion images are smooth and correspond to the diffusion of the contrast agent in the lesion. The normal tissue image sequences are stable and correspond to the enhanced normal tissue. To speed up the computational process of our convergent algorithm, the split Bregman iteration algorithm is applied. To compare the algorithm results, images generated by decomposed methods without normal tissue constraint based on total variation are compared with those generated by our method. The performance of the proposed method is evaluated by the correlation of normal tissue images with the lesion classification accuracy of lesion images. Results: Ninety‐eight lesions, including 40 benign and 58 malignant, are evaluated. The dataset includes various typical pathologies of the breast such as invasive ductal carcinoma, ductal carcinoma in situ, tubular carcinoma, phyllodes tumor, hyperplasia, and fibroadenoma, among others. The area under the ROC for the pure lesion enhancement images acquired by DC‐TV is greater than that acquired by the original DCE‐MRIs. Conclusions: The pure enhancement information from the original breast DCE‐MRI lesions can be successfully obtained using our DC‐TV. The TICs based on the acquired pure enhancement information closely conform to three‐time‐point model, which is a classic diagnosis rule. The experiment shows that DC‐TV provide a credible TIC generation program.