We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
MRAUnet++: A Novel Multi-Scale Residual Attention Network for Enhanced Rectal Cancer Segmentation.
- Authors
Zhengpeng Li; Jun Hu; Zhuang Liang; Jiansheng Wu
- Abstract
Deep learning (DL) models play a crucial role in medical image analysis, with their performance reliant on the scale and diversity of available training data. However, medical imaging faces challenges like data acquisition difficulty, sample class imbalance, and high annotation costs. To tackle these issues and enable automatic and accurate segmentation of rectal tumor regions in CT images, aiding physicians in diagnosis and treatment, we present the innovative Multi-scale Residual Attention-based Unet++ Network (MRAUnet++). This model replaces UNet++'s convolutional blocks with multi-scale residual blocks, extracting features across different scales for enriched diversity. Residual connections optimize the network structure, enhancing its ability to express complex features. The integrated attention mechanism suppresses irrelevant features, boosting segmentation performance by allowing selective focus on crucial rectal tumor features. Empirical evaluations on a rectal tumor dataset demonstrate MRAUnet++'s outstanding performance, achieving a Dice coefficient of 83.32% and an IoU value of 70.06%, surpassing other comparative models. These results highlight the model's enhanced accuracy in rectal tumor segmentation, providing a reliable diagnostic tool for medical practitioners.
- Subjects
RECTAL cancer; RECTUM tumors; COMPUTER-assisted image analysis (Medicine); ENDORECTAL ultrasonography; IMAGE segmentation; IMAGE analysis; COMPUTED tomography
- Publication
Engineering Letters, 2024, Vol 32, Issue 4, p880
- ISSN
1816-093X
- Publication type
Article