Fig. 1: Schematic of the deep learning workflow for prognostication of emphysema progression.

This workflow diagram illustrates the process of local emphysema progression prediction using a local foundational model of density evolution. Starting with high-resolution CT scans, regions of interest (ROI) are identified and fed into the FM. The encoder part of the autoencoder compresses the ROI into a lower-dimensional representation z, which is then reconstructed back to the original data space by the decoder. The feature vector z is subsequently processed through a multilayer perceptron (MLP) to estimate the progression risk score. The risk score is depicted as the local emphysema progression. The final output includes the localized visualization of emphysema progression on the CT scan, highlighted for clinical assessment.