Figure 1:

Octrees enable compressed image representation while maintaining image and segmentation quality. (A) Octree compression with an intensity threshold τ can decrease the memory (Mem) footprint while avoiding substantial image degradation. In this example, τ = 200 HU leads to 15% memory use with minimal image (root mean square error [RMSE], 0.98 HU) and segmentation (left ventricular [LV] Dice coefficient, 0.979; LV border Dice coefficient [LV bDice], 0.884) errors. (B) To achieve the desired image memory footprint, τ thresholds were identified through a binary search, with increasing τ values leading to more significant memory savings (a lower memory footprint).