Name
Abstract | ||
|---|---|---|
Malignancy in women's breast is known to be the second most common form of cancer. Early detection can help diagnose the disease effectively, but it continues to grow manifolds due to reasons unknown. Therefore, to aid radiologists in the effective treatment of breast cancer, an end-to-end deep learning-based architecture for ROI-based breast mass segmentation is proposed. The architecture involving Residual connections and Group convolution in U-Net (RGU-Net), contains encoder and decoder blocks with different resolutions and feature sizes. The architecture captures multi-level features from the encoder-decoder architecture using the residual connections and group convolution. Moreover, to improve the field-of-view of the filters, atrous convolutions are added. Later, for better visualization, the predicted masks are labelled (structured learning) using a conditional random field (CRF) to analyse the mass boundaries explicitly. A publicly available INBreast dataset is used to validate the method, which is augmented to produce robust results. The experimental results produced from the proposed approach outperformed the conventional mass segmentation algorithms, demonstrating its effectiveness. |
Year | DOI | Venue |
|---|---|---|
2021 | 10.23919/EUSIPCO54536.2021.9616230 | 29TH EUROPEAN SIGNAL PROCESSING CONFERENCE (EUSIPCO 2021) |
Keywords | DocType | ISSN |
Breast Mass, Feature processing, Semantic Segmentation, Residual Mapping, Group Convolution, Image labeling | Conference | 2076-1465 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
2 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Ridhi Arora | 1 | 0 | 0.34 |
| Balasubramanian Raman | 2 | 0 | 0.34 |