Controlling for Biasing Signals in Images for Prognostic Models: Survival Predictions for Lung Cancer with Deep Learning
This work addresses bias in AI-driven healthcare predictions for lung cancer patients, representing an incremental step toward causal deep learning in medical imaging.
The study tackled the problem of biased prognostic models in medical imaging by using deep learning to control for confounding signals in CT scans of lung cancer, achieving unbiased survival predictions through a dual-task network that enforces independence between outcome and collider activations.
Deep learning has shown remarkable results for image analysis and is expected to aid individual treatment decisions in health care. To achieve this, deep learning methods need to be promoted from the level of mere associations to being able to answer causal questions. We present a scenario with real-world medical images (CT-scans of lung cancers) and simulated outcome data. Through the sampling scheme, the images contain two distinct factors of variation that represent a collider and a prognostic factor. We show that when this collider can be quantified, unbiased individual prognosis predictions are attainable with deep learning. This is achieved by (1) setting a dual task for the network to predict both the outcome and the collider and (2) enforcing independence of the activation distributions of the last layer with ordinary least squares. Our method provides an example of combining deep learning and structural causal models for unbiased individual prognosis predictions.