Using a Cross-Task Grid of Linear Probes to Interpret CNN Model Predictions On Retinal Images
This work provides insights into model interpretability for medical imaging, but it is incremental as it applies existing linear probe methods to a new dataset.
The researchers tackled the problem of interpreting CNN predictions on retinal images by analyzing a dataset using linear probes across 93 tasks, finding that middle-layer representations are more generalizable and that some target tasks are better predicted from correlated source tasks than from the same task.
We analyze a dataset of retinal images using linear probes: linear regression models trained on some "target" task, using embeddings from a deep convolutional (CNN) model trained on some "source" task as input. We use this method across all possible pairings of 93 tasks in the UK Biobank dataset of retinal images, leading to ~164k different models. We analyze the performance of these linear probes by source and target task and by layer depth. We observe that representations from the middle layers of the network are more generalizable. We find that some target tasks are easily predicted irrespective of the source task, and that some other target tasks are more accurately predicted from correlated source tasks than from embeddings trained on the same task.