Mapping resinwood in CT scans of Scots pine with a hybrid Gaussian Mixture Model image segmentation pipeline

The presence of resinwood in Cronartium pini-infected Scots pine stems represents a major quality concern, diminishing timber value and challenging processing workflows. Although X‑ray computed tomography (CT) provides a non‑destructive means of visualising internal density, robust and automated methods to separate resinwood from heartwood and sapwood in such scans remain underdeveloped. To address this, we present a hybrid image analysis pipeline that combines a slice‑adaptive Gaussian Mixture Model (GMM), this statistically characterises tissue‑specific density patterns in each CT slice. This dataset was created through X-ray computed tomography (CT) imaging for developing segmentation method that capable identifying internal resinwood. Infected pine tree samples were inspected and harvested by trained inventory specialist, then scanned by MicroTec Mito industrial-scale CT scanner. The scanner produced data using a cone beam and two angled flat detectors on a helical scanning trajectory. The spatial resolution and resulting voxel dimensions were uniform at 0.5 × 0.5 × 0.5 mm³. The resolution of each image was 750 x 750 pixels on 2D. The resulting 3D images comprised 16-bit greyscale values representing density in kg/m³ at each location, and helical 1PI Katsevich was used for image reconstruction.