Towards Estimating the Uncertainty Associated with Three-Dimensional Geometry Reconstructed from Medical Image Data

Area of Science:

• Medical Imaging and Image Analysis
• Computational Modeling
• Biomedical Engineering

Background:

• Patient-specific computational modeling relies heavily on accurate medical imaging for treatment planning.
• Image artifacts and segmentation processes introduce variability and potential inaccuracies in anatomical data extraction.
• Quantifying uncertainty in medical image analysis is crucial for reliable patient-specific treatment plans.

Purpose of the Study:

• To develop and analyze image datasets to estimate the uncertainty associated with extracting anatomical data from medical images.
• To assess the impact of image artifacts and segmentation variability on the accuracy of anatomical measurements.
• To investigate methods for improving segmentation accuracy in medical image analysis.

Main Methods:

• Development of two image datasets: one from a "virtual voxelization" of a CAD model (idealized) and another from CT scanning of physical spherical phantoms.
• Standard image analysis procedures were applied to extract anatomical data.
• Investigation of established thresholding procedures to enhance segmentation accuracy.

Main Results:

• For the idealized sphere, diameter error was ≤2% with 5+ voxels across the diameter.
• Measurement error increased to approximately 4% for physical phantoms with similar voxelization.
• Thresholding procedures were adapted to explore improvements in segmentation accuracy.

Conclusions:

• Image acquisition artifacts and segmentation variability introduce measurable errors in anatomical data extraction.
• The study quantifies the impact of these errors, showing higher uncertainty in physical scans compared to idealized models.
• Accurate estimation of uncertainty is essential for reliable patient-specific computational modeling and treatment planning.