主要组件分析和微调视觉转换整合模型可解释性用于乳腺癌预测
Huong Hoang Luong1,2, Phuc Phan Hong3, Dat Vo Minh3
1College of Information and Communication Technology, Can Tho University, Can Tho 900000, Vietnam.
Visual computing for industry, biomedicine, and art
|March 10, 2025
概括
这项研究使用视觉转换器 (ViT) 模型增强了乳腺癌诊断. 改进的模型在从组织病理学图像分类瘤方面达到97.13%的准确性,为新的基准设定了新的基准.
科学领域:
- 在瘤学瘤学.
- 计算机科学 计算机科学
- 医疗成像医学成像
背景情况:
- 乳腺癌是全球主要的健康问题,需要准确的诊断工具.
- 组织病理学对于乳腺癌的诊断和分类至关重要,推动了医学和计算领域的研究.
- 开发用于组织病理学图像分析的有效计算方法对于改善患者的治疗结果至关重要.
研究的目的:
- 开发和评估一种先进的机器学习模型,用于从组织病理图像准确地分类乳腺癌.
- 通过新的注意力机制和维度减小技术来提高视觉变压器 (ViT) 模型的性能.
- 确保模型可解释性和可解释性,使用已建立的AI可解释性方法.
主要方法:
- 视觉变压器 (ViT) 模型被用于使用BreakHis数据集将乳腺瘤分类为良性或恶性.
- 在微调过程中引入了多头本地大内核自我注意,以提高模型性能.
- 主要组件分析 (PCA) 应用于减小维度,以减轻过拟合和减少计算负载.
- 使用VGGIN进行集体学习,以进一步提高分类准确性.
- 为了模型的可解释性,使用了夏普利添加式解释 (SHAP),局部可解释的模型不可知解释 (LIME) 和梯度加权类激活映射 (Grad-CAM).
主要成果:
- 微调的ViT模型具有多头本地大内核自我注意,在100×放大时实现了95.94%的精度,比标准ViT有3.34%的改进.
- PCA应用导致了额外的3.34%的准确性改进,证明了其在减少过度装配和计算复杂性的有效性.
- 与VGGIN一起学习进一步提高了分类准确度,达到97.13%.
- 与现有的最先进的方法相比,拟议的方法显示了精度改进,从0.98%到17.13%不等.
- 可解释性方法提供了对特征重要性和模型关注的见解,有助于理解分类决策.
结论:
- 该研究成功开发了乳腺癌组织病理学图像分类的高性能模型,取得了最先进的结果.
- 新型注意力机制,PCA和集体学习的整合显著提高了分类的准确性和效率.
- 使用可解释性技术提高了AI模型在乳腺癌诊断中的可信度和临床适用性.
- 这项研究为自动化乳腺癌组织病理图像分析建立了新的基准,为改进的诊断工具铺平了道路.
相关概念视频
02:43Mouse Models of Cancer Study
Mice have long served as models for studying human biology and pathology because of their phylogenetic and physiological similarity with humans. They are also easy to maintain and breed in the laboratory, and hence, many inbred strains are now available for research. Studies on mice have contributed immeasurably to our understanding of cancer biology.
The development of transgenic, knockout, and knock-in mice has led to an exponential increase in their use as model organisms in research,...
The development of transgenic, knockout, and knock-in mice has led to an exponential increase in their use as model organisms in research,...
02:43Mouse Models of Cancer Study
Mice have long served as models for studying human biology and pathology because of their phylogenetic and physiological similarity with humans. They are also easy to maintain and breed in the laboratory, and hence, many inbred strains are now available for research. Studies on mice have contributed immeasurably to our understanding of cancer biology.
The development of transgenic, knockout, and knock-in mice has led to an exponential increase in their use as model organisms in research,...
The development of transgenic, knockout, and knock-in mice has led to an exponential increase in their use as model organisms in research,...