Journal of the Egyptian National Cancer Institute · 2026
MSA U-Net: a multi-scale attention-enhanced network for automatic segmentation of uterine lymph nodes in MRI images.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Hong Jingde, Chen Chunxia, Qiu Juping, Lin Zeng, Xie Yuhang, Lin Yongping
Аннотация
OBJECTIVE: Lymph nodes (LNs) in medical images often have fuzzy boundaries, vary in size and shape, and have intensities similar to those of neighboring tissues, making accurate segmentation challenging. To address this issue, we propose a multi-scale attention-enhanced network (MSA U-Net) that integrates channel-wise and spatial attention mechanisms for automatic segmentation of metastatic pelvic LNs associated with uterine malignancies from sagittal magnetic resonance imaging (MRI).
METHODS: This network integrates a U-Net backbone, fuses features from different encoder levels at skip connections, and feeds the fused features into the decoder at each layer. Different receptive fields are used at the bottleneck nodes to capture multiscale contextual information. A detection head is added to the bottom layer of the network and the detection results are used to assist in the final segmentation of the image. A lightweight design of the convolutional block attention module is also implemented to optimize feature representation.
RESULTS: The experimental results demonstrate that the proposed network achieves better segmentation performance compared to the baseline model. The proposed network achieves a mean intersection over union of 0.76, an average pixel accuracy of 0.97, a precision of 0.78, a recall of 0.97, a Dice coefficient of 0.82, and a Hausdorff distance of 2.21.
CONCLUSIONS: The proposed MSA U-Net segmentation network effectively segments LNs in uterine MRI images, outperforming existing segmentation methods. This study provides a reliable and automated method to help clinicians detect LN, thus improving clinical decision-making.
METHODS: This network integrates a U-Net backbone, fuses features from different encoder levels at skip connections, and feeds the fused features into the decoder at each layer. Different receptive fields are used at the bottleneck nodes to capture multiscale contextual information. A detection head is added to the bottom layer of the network and the detection results are used to assist in the final segmentation of the image. A lightweight design of the convolutional block attention module is also implemented to optimize feature representation.
RESULTS: The experimental results demonstrate that the proposed network achieves better segmentation performance compared to the baseline model. The proposed network achieves a mean intersection over union of 0.76, an average pixel accuracy of 0.97, a precision of 0.78, a recall of 0.97, a Dice coefficient of 0.82, and a Hausdorff distance of 2.21.
CONCLUSIONS: The proposed MSA U-Net segmentation network effectively segments LNs in uterine MRI images, outperforming existing segmentation methods. This study provides a reliable and automated method to help clinicians detect LN, thus improving clinical decision-making.