Signature Activation: A Sparse Signal View for Holistic Saliency
This work addresses explainability for medical imaging, specifically for clinicians using CNNs in healthcare, but appears incremental as it builds on existing saliency methods with a focus on medical images.
The paper tackles the need for model transparency in healthcare by introducing Signature Activation, a saliency method that provides holistic and class-agnostic explanations for CNN outputs, and demonstrates its potential in clinical settings by aiding lesion detection in coronary angiograms.
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.