Abstract:
|
Healthy tissue near non-small cell lung tumor (NSCLC) of inoperable low-risk patients is exposed to clinically relevant doses of radiation therapy (RT) causing radiographic radiation-induced lung damage (RILD). This study aims to build a statistical model of RILD over time that describes location-specific changes. CT-scans of one NSCLC patient were obtained at 3, 6, 12, 18, and 21 months post-treatment. CT-scans are 3D images with almost 27 million voxels but the integrity of the voxels across time cannot be assured, thus, we grouped them into larger subvolumes called patches. RILD is measured through three-threshold based ordinal categories (dense, hazy and none), then, each patch is represented by a vector in the form of compositional data. A zero-inflated, non-linear mixed effects model for longitudinal lung tissue changes following RT was built for different patches to capture the locational and RT dose dependence. The model was adjusted by the spatial correlation between neighboring patches. We will demonstrate the performance of the model regarding predictive accuracy. Extension of the model towards a population model with multiple similar patients will be discussed.
|