Lihui Wang Openreview

Leo Migdal

-Dec 6, 2025, 4:28 PM

To protect your privacy, all features that rely on external API calls from your browser are turned off by default. You need to opt-in for them to become active. All settings here will be stored as cookies with your web browser. For more information see our F.A.Q. Add open access links from to the list of external document links (if available). Privacy notice: By enabling the option above, your browser will contact the API of unpaywall.org to load hyperlinks to open access articles.

Although we do not have any reason to believe that your call will be tracked, we do not have any control over how the remote server uses your data. So please proceed with care and consider checking the Unpaywall privacy policy. For web page which are no longer available, try to retrieve content from the of the Internet Archive (if available). Privacy notice: By enabling the option above, your browser will contact the API of archive.org to check for archived content of web pages that are no longer available. Although we do not have any reason to believe that your call will be tracked, we do not have any control over how the remote server uses your data. So please proceed with care and consider checking the Internet Archive privacy policy.

To promote the generalization ability of breast tumor segmentation models, as well as to improve the segmentation performance for breast tumors with smaller size, low-contrast and irregular shape, we propose a progressive dual priori... The PDPNet first cropped tumor regions with a coarse-segmentation based localization module, then the breast tumor mask was progressively refined by using the weak semantic priori and cross-scale correlation prior knowledge. To validate the effectiveness of PDPNet, we compared it with several state-of-the-art methods on multi-center datasets. The results showed that, comparing against the suboptimal method, the DSC and HD95 of PDPNet were improved at least by 5.13% and 7.58% respectively on multi-center test sets. In addition, through ablations, we demonstrated that the proposed localization module can decrease the influence of normal tissues and therefore improve the generalization ability of the model. The weak semantic priors allow focusing on tumor regions to avoid missing small tumors and low-contrast tumors.

The cross-scale correlation priors are beneficial for promoting the shape-aware ability for irregular tumors. Thus integrating them in a unified framework improved the multi-center breast tumor segmentation performance. The source code and open data can be accessed at https://github.com/wangli100209/PDPNet. Breast tumor segmentation, DCE-MRI, deep learning, sementic prior, multicenter dataset. Breast cancer has become the most prevalent cancer worldwide [1, 2], its early diagnosis and precise treatment can undoubtedly increase the survival rate. To assist the diagnosis and to further facilitate the quantitative analysis, the detailed delineation of the tumor regions is usually required.

However, due to the complexity of shapes and sizes of the breast tumors, manual annotation of the regions of interest is very time-consuming and labor-intensive, in addition, the annotation accuracy highly relies on the... Therefore, investigating accurate and automatic breast tumor segmentation algorithms is of great significance. With the successful development of deep learning models in the filed of computer vision, numerous learning-based studies have been dedicated to automatically segmenting breast tumors from mammography images [3, 4, 5], ultrasound images [6,... For instance, Benjelloun et al. segmented breast tumors from DCE-MRIs with UNet and obtained a relative high mean intersection-over-union region (mIOU) value, demonstrating the feasibility of deep learning model for breast tumor segmentation [11]. To further reduce the effect of thoracic cavity region and promote the breast tumor segmentation performance, Zhang et al.

used a cascade FCNet to segment the whole-breast with the pre-contrast MRIs firstly and then delineated tumors from the breast region with all the DCE-MRIs, even though it achieved a better performance, its two-stage... Considering the sequential property of time-intensity curve (TIC) of DCE-MRIs and the merits of long and short-time memory (LSTM) network in exploring sequential data, Chen et al. combined LSTM and convolutional neural network (CNN) to segment breast tumors from 3D DCE-MRIs[13]. Moreover, the kinetic maps calculated from the DCE-MRIs can reflect pathological properties of tumors, which may provide complementary information for quantifying the tumor regions, accordingly, Qiao et al. used both semi-quantitative parameter maps and DCE-MRIs as the input of network to predict breast tumor regions, achieving a promising performance[14]. Although using the prior physical knowledge of DCE-MRIs can increase breast tumor segmentation accuracy, the segmentation performance for non-mass enhancement tumors with small size and irregular shapes is still not satisfactory due to the...

To address the above-mentioned issues, Wang et al. proposed a novel tumor-sensitive synthesis module to decrease the false-positive segmentation by introducing the differential loss between true and false breast tumors [15]. Considering that the multi-perspectives information are complementary and may improve the segmentation performance, Wang et al. fused the image features of adjacent slices and combined the multiscale features to improve the segmentation accuracy [16]. Meanwhile, given that multiscale feature representations are beneficial for capturing the details of the target and the transformer module is good at extracting global semantic information, Qin et al. introduced a multi-scale parallel convolution fusion module and a transformer module into the UNet architecture to delineate the tumor regions, which enables the model to enhance the edge and fine information [17].

To further increase the segmentation accuracy of tumors with irregular shapes, Peng et al. used the adaptive deformable convolution layer to capture the details of tumor edges[18], Hirsch et al. trained a deep network with non-overlapping image patches to recover the tumor positional information[19], and Zhu et al proposed a 3D affinity learning-based multi-branch ensemble network to refine the tumor and boundary voxels [20]. Even though modifying network structure and fully using kinetic parameters derived from DCE-MRI offer a promising paradigm for breast tumor segmentation, most of the existing models are two-stage learning-based, which requires to segment breast... In addition, most of them did not consider the multicenter segmentation problems and have poor generalizability across different image domains. Dr.

Lihui Wang is a Principal Investigator and a Ph.D. mentor at the Institute of Psychology and Behavioral Science, Shanghai Jiao Tong University, and Shanghai Mental Health Center. She is currently a Mercator Fellow of the German Research Foundation (DFG), a committee member of the Research Technologies in Neuroscience of Chinese Neuroscience Society, and an editorial board member of Journal of Psychological... She was awarded “Hanix Youth Scientist” by Chinese Psychology Society, and “Qiusuo Outstanding Youth” by Shanghai Center for Brain Science and Brain-inspired Technology. Published with Wowchemy — the free, open source website builder that empowers creators. A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity.© Copyright 2025 IEEE - All rights reserved.

Lihui Wang Openreview

People Also Search

To Protect Your Privacy, All Features That Rely On External

Although We Do Not Have Any Reason To Believe That

To Promote The Generalization Ability Of Breast Tumor Segmentation Models,

The Cross-scale Correlation Priors Are Beneficial For Promoting The Shape-aware

However, Due To The Complexity Of Shapes And Sizes Of