Liver-tumor boundary detection: human observer vs computer edge detection.

Area of Science:

• Medical Imaging
• Computational Pathology
• Radiology

Background:

• Accurate tumor volume calculation is crucial for treatment planning and monitoring.
• Defining the precise liver-tumor interface in computed tomography (CT) images presents a significant challenge.
• Existing methods for manual tumor delineation can be time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and evaluate a computer edge detection program for defining the liver-tumor interface in CT images.
• To assess the program's accuracy in calculating tumor volumes and cross-sectional areas.
• To compare the program's performance against manual measurements and known phantom volumes.

Main Methods:

• Development of a novel edge-linking algorithm for liver-tumor interface detection.
• Testing the program using CT images from a lucite liver/tumor phantom, simulated pseudotumors, and 12 patient livers.
• Comparison of computer-calculated tumor sizes with measured volumes and manual delineations.

Main Results:

• The edge-linking algorithm demonstrated reasonable success in calculating areas of pseudotumors with sufficient contrast and edge gradients.
• While systematic overestimation occurred in phantom slices, total tumor volume errors were within acceptable ranges for most sizes.
• Variability in manual measurements for ill-defined tumors (14.0%) was significantly higher than for well-defined tumors (7.1%).

Conclusions:

• The developed computer edge detection program shows promise for accurately defining the liver-tumor interface in CT images.
• This automated approach can potentially reduce the variability associated with manual tumor measurements.
• Further refinement may enhance accuracy, particularly for tumors with ill-defined borders, improving tumor volume computation.