Signature Activation: A Sparse Signal View for Holistic Saliency
Signature Activation: A Sparse Signal View for Holistic Saliency
复制标题
DOI：
10.48550/arxiv.2309.11443
复制
发表时间：
2023-09
期刊：
影响因子：
--
通讯作者：
José Roberto Tello-Ayala;A. Fahed;Weiwei Pan;E. Pomerantsev;P. Ellinor;A. Philippakis;F. Doshi-Velez
中科院分区：
文献类型：
--
作者：
José Roberto Tello-Ayala;A. Fahed;Weiwei Pan;E. Pomerantsev;P. Ellinor;A. Philippakis;F. Doshi-Velez
在医疗保健中采用机器学习需要模型的透明度和可解释性。在这项工作中，我们介绍了签名激活，这是一种显着性方法，可以为卷积神经网络（CNN）输出生成整体和类不可知的解释。我们的方法利用了某些类型的医学图像（例如血管造影片）具有清晰的前景和背景对象的事实。我们给出了理论解释来证明我们的方法。我们显示了我们的方法在临床环境中的潜在用途，通过评估其有效性，以帮助检测冠状动脉造影病变。
The adoption of machine learning in healthcare calls for model transparency and explainability. In this work, we introduce Signature Activation, a saliency method that generates holistic and class-agnostic explanations for Convolutional Neural Network (CNN) outputs. Our method exploits the fact that certain kinds of medical images, such as angiograms, have clear foreground and background objects. We give theoretical explanation to justify our methods. We show the potential use of our method in clinical settings through evaluating its efficacy for aiding the detection of lesions in coronary angiograms.