启用Shapley值的渐进型伪袋增大用于全幻灯片图像分类
IEEE transactions on medical imaging
|September 2, 2024
概括
这项研究引入了一种新的计算病理学方法,使用Shapley值进行更准确的全幻灯片图像分类. 该方法改善了多实例学习中的实例重要性得分,增强了诊断洞察力.
科学领域:
- 计算病理学计算病理学
- 数字病理学数字病理学
- 机器学习在医学中的应用
背景情况:
- 由于千兆像素分辨率和有限的注释,全幻灯片图像 (WSI) 的分类具有挑战性.
- 多实例学习 (MIL) 是WSI的弱监督方法,但从袋级标签中提炼实例级信息是困难的.
- 传统的MIL方法常常会出现注意力分布偏差和实例识别不准确的问题.
研究的目的:
- 开发一种改进的方法,例如MIL中的重要性评分 (IIS),用于WSI分类.
- 提高MIL模型在计算病理学中的准确性和可解释性.
- 解决传统基于注意力的IIS估计的局限性.
主要方法:
- 提出了一种新的方法,灵感来自于合作游戏理论,利用Shapley值来评估实例贡献,以改进IIS估计.
- 使用注意力机制加速Shapley值计算,同时保持增强的实例识别和优先级.
- 引入了一个基于估计的IIS进行渐进的伪袋分配的框架,以促进平衡的注意力分配.
主要成果:
- 在CAMELYON-16,BRACS,TCGA-LUNG和TCGA-BRCA数据集上表现出高于最先进的方法的性能.
- 在WSI分类中实现了增强的可解释性和类明智的洞察力.
- 验证了Shapley值和伪袋分配在计算病理学中的MIL的有效性.
结论:
- 提出的沙普利基于价值的MIL方法显著提高了WSI分类的准确性和可解释性.
- 该方法有效地完善实例级信息,克服传统注意力机制的局限性.
- 这项工作为推进数字病理学中弱监督学习提供了一个有希望的方向.
相关概念视频
02:16Light Acquisition
In order to produce glucose, plants need to capture sufficient light energy. Many modern plants have evolved leaves specialized for light acquisition. Leaves can be only millimeters in width or tens of meters wide, depending on the environment. Due to competition for sunlight, evolution has driven the evolution of increasingly larger leaves and taller plants, to avoid shading by their neighbors with contaminant elaboration of root architecture and mechanisms to transport water and nutrients.
01:05Histogram
The histogram is a graphical representation in the x-y form of data distribution in a data set. The horizontal x-axis is labeled with what the data represents (for instance, distance from your home to school). The vertical y-axis is labeled either frequency or relative frequency (or percent frequency or probability).
A histogram graph consists of contiguous (adjoining) boxes. The heights of the bars correspond to frequency values. The graph will have the same shape with respective labels. The...
A histogram graph consists of contiguous (adjoining) boxes. The heights of the bars correspond to frequency values. The graph will have the same shape with respective labels. The...