| 摘 要: 针对混合型肝癌病理切片中三级淋巴结构人工识别存在耗时费力、准确性不高等问题,基于SegFormer(Segmentation Transformer)模型提出了一种轻量化的识别模型。首先,使用非负矩阵法对病理切片进行标准化处理。其次,增加特征图层级,使模型能够获得更精细的感受视野。最后,引入SimAM(Similarity-Aware ActivationModule)注意力机制,辅助模型定位三级淋巴结构的关键信息。使用374张病理切片进行模型训练和验证,实验结果显示,该模型的平均交并比、平均召回率和准确率分别达到了92.50%、96.08%和96.39%,在三级淋巴结构识别方面表现出色,具有显著优势。 |
| 关键词: SegFormer模型;三级淋巴结构;混合型肝癌;图像识别;深度学习 |
|
中图分类号: TP391.41;R735.7
文献标识码: A
|
|
| An Automatic Recognition Algorithm for Tertiary Lymphoid Structures in Combined Hepatocellular Carcinoma Based on an Improved SegFormer Model |
|
YANG Xiaodong1, SHI Xinyi1, DONG Wei1, LIU Baolin1
|
(1.School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China;
2.School of Computer Science and Engineering, Sichuan University of Science & Engineering, Zigong 643002, China;
3.Department of Pathology, Eastern Hepatobiliary Surgery Hospital, Naval Medical University (Second Military Medical University), Shanghai 200438, China)
y3305365587@163.com; 422801214@qq.com; well_dw@126.com; blliuk@163.com
|
| Abstract: To address the issues of time-consuming, labo-r intensive, and low accuracy in manual identification of Tertiary Lymphoid Structures (TLS) in pathological slides of combined Hepatocellular Carcinoma (cHCC), this study proposes a lightweight recognition model based on the SegFormer (Segmentation Transformer) architecture. Firstly, Non-negative Matrix Factorization (NMF) is employed to standardize pathological slides. Secondly, additional feature map layers are incorporated to enable the model to obtain a finer receptive field. Finally, the SimAM (Similarity-Aware Activation Module) attention mechanism is introduced to assist the model in locating critical TLS regions. The model is trained and validated on 374 pathological slides, and the experiment results show that the model achieves a mean Intersection-ove-r Union (mIoU) of 92.50% , an average recall of 96.08% , and an accuracy of 96.39% respectively. The proposed modelperforms exceptionally well in recognizing TLS, exhibiting significant advantages. |
| Keywords: SegFormer model; Tertiary Lymphoid Structures (TLS); combined Hepatocellular Carcinoma (cHCC);image recognition; deep learning |
