The system integrates a spatial prior probability map created by spatially normalizing manually segmented training data with regional mapping enhanced by total variation regularization. It employs a level set framework that incorporates propagated shape constraints to refine organ boundaries across adjacent slices. Edges detected through image gradient computations are combined with the probability map to delineate candidate kidney regions, which are then segmented into left and right compartments. An iterative energy minimization process further sharpens these boundaries, and a morphological closing step smooths contours before converting pixel counts into a total organ volume using image metadata.
Description
This technical solution is differentiated by its fully automated, robust approach to accurately measuring kidney volumes in challenging cases such as autosomal dominant polycystic kidney disease. It overcomes the limitations of manual segmentation by reducing errors and observer variability. The integration of spatial priors, noise-reducing regional mapping, and consistent boundary refinement minimizes interference from adjacent organs and irregular shapes caused by cysts. These combined features ensure reliable, high-resolution segmentation from MR images, offering a significant advancement in monitoring disease progression through precise volumetric assessments.
Applications
- Automated kidney segmentation
- Kidney volume measurement software
- ADPKD progression monitoring
Advantages
- Automated kidney segmentation reduces time, errors, and inter-observer variability compared to manual methods.
- Enhances segmentation accuracy by integrating spatial priors, noise reduction with total variation regularization, and boundary refinement with propagated shape constraints.
- Precisely distinguishes and segments left and right kidneys, supporting accurate total kidney volume (TKV) measurements.
- Facilitates reliable longitudinal monitoring of ADPKD progression and could be adapted for segmenting other organs.
IP Status
https://patents.google.com/patent/US10706548B2
