Explaining 3D Computed Tomography Classifiers with Counterfactuals
This work addresses the problem of interpretability in 3D medical imaging for clinicians and researchers, providing a solution for an incremental improvement in the field.
The authors tackled the challenge of explaining 3D computed tomography classifiers and achieved a memory-efficient and effective method for generating interpretable counterfactuals. Their approach was demonstrated on two models for clinical phenotype prediction and lung segmentation.
Counterfactual explanations enhance the interpretability of deep learning models in medical imaging, yet adapting them to 3D CT scans poses challenges due to volumetric complexity and resource demands. We extend the Latent Shift counterfactual generation method from 2D applications to explain 3D computed tomography (CT) scans classifiers. We address the challenges associated with 3D classifiers, such as limited training samples and high memory demands, by implementing a slice-based autoencoder and gradient blocking except for specific chunks of slices. This method leverages a 2D encoder trained on CT slices, which are subsequently combined to maintain 3D context. We demonstrate this technique on two models for clinical phenotype prediction and lung segmentation. Our approach is both memory-efficient and effective for generating interpretable counterfactuals in high-resolution 3D medical imaging.