This repository contains models and code for predicting Signal-to-Noise Ratio (SNR) from CT phantom images using a causal representation learning approach. The models are based on Variational Autoencoders (VAEs) and support ablation testing to evaluate the contribution of various acquisition metadata: voltage, time, and contrast agent.
- Build a causal-aware SNR prediction model from CT phantom image data.
- Incorporate acquisition parameters (voltage, time, contrast agent) as structured metadata.
- Evaluate the influence of each variable through ablation and intervention.
- Simulate
do()operations to interpret causal effects on SNR.