全文下载:
20.pdf
文章编号: 1672-6987(2025)02-0150-09 DOI: 10.16351/j.1672-6987.2025.02.020
臧敏a, 曹源a, 丰艳a*, 于彬a,b*(青岛科技大学 a. 信息科学技术学院; b. 数据科学学院, 山东 青岛 266061)
摘要: 乳腺癌是全球女性患病率最高的癌症,组织病理学诊断被认为是诊断乳腺癌的最佳方法。然而,组织病理学图像的复杂程度和病理学家的知识储备会影响诊断结果的准确性。为了解决以上问题同时协助临床医生诊断疾病,本工作提出了一种新颖的乳腺癌组织病理图像预测模型,称为双信息流移动倒置瓶颈卷积神经网络(DMBC-Net)。首先,利用深度可分离卷积、坐标注意力和跳跃连接构成DMBC模块。该模块为DMBC-Net的基础模块,可以在扩大感受野的同时获得具有更多有效信息的特征映射。其次,在ImageNet和BACH数据集的基础上利用二次迁移学习训练和优化模型参数。最后,该模型在公开的BreaKHis数据集进行了验证。在40、100、200、400倍放大倍数的影响下,分类准确率分别达到97.86%、96.10%、98.00%、97.70%。将DMBC-Net与其他先进模型比较,实验结果表明,DMBC-Net具有优秀的预测能力。
关键词: 乳腺癌; 组织病理图像; 移动倒置瓶颈; 迁移学习; 坐标注意力
中图分类号: TP 391.4 文献标志码: A
引用格式: 臧敏, 曹源, 丰艳, 等. 基于移动倒置瓶颈和迁移学习的乳腺癌组织病理图像二值分类[J]. 青岛科技大学学报(自然科学版), 2025, 46(2): 150-158.
ZANG Min, CAO Yuan, FENG Yan, et al. Binary classification of breast cancer histopathological images based on mobile inverted bottleneck and transfer learning[J]. Journal of Qingdao University of Science and Technology(Natural Science Edition), 2025, 46(2): 150-158.
Binary Classification of Breast Cancer Histopathological Images Based on Mobile Inverted Bottleneck and Transfer Learning
ZANG Mina, CAO Yuana, FENG Yana, YU Bina,b(a. College of Information Science and Technology;b. College of Data Science, Qingdao University of Science and Technology, Qingdao 266061, China)
Abstract: Breast cancer is the most common cancer in women worldwide, and histopathological diagnosis is considered the best way to diagnose breast cancer. However, the complexity of histopathological images and the knowledge reserve of pathologists can affect the accuracy of diagnostic results. In order to solve the above problems and assist clinicians in diagnosing diseases, this paper proposed a novel model for predicting histopathological images of breast cancer, called dual-mobile inverted bottleneck convolutional neural network (DMBC-Net). First, a DMBC module is formed using depthwise separable convolution, coordinate attention, and skip connections. This module is the basic module of DMBC-Net, which can obtain feature mapping with more effective information while expanding the receptive field. Then, secondary transfer learning is used to train and optimize model parameters based on the ImageNet and BACH datasets. Finally, DMBC-Net is verified on the public BreaKHis dataset. Under the influence of 40,100, 200 and 400 magnifications, the classification accuracy reached 97.86%, 96.10%, 98.00% and 97.70% respectively. Comparing DMBC-Net with other advanced models, the experimental results show that DMBC-Net has excellent prediction ability.
Key words: breast cancer; histopathological images; mobile inverted bottleneck; transfer learning; coordinate attention
收稿日期: 2024-10-17
基金项目: 国家自然科学基金项目(62172248);山东省自然科学基金项目(ZR2021MF098).
作者简介: 臧敏(1998—),女,硕士研究生. * 通信联系人.