Measuring Efficiency of Semi-automated Brain Tumor Segmentation by Simulating User Interaction

Area of Science:

• Medical Imaging
• Radiology
• Computational Biology

Background:

• Traditional tumor extent quantification relies on crude 2D measurements, lacking volumetric accuracy.
• Algorithmic segmentation methods offer potential for improved accuracy but vary in required radiologist interaction levels.

Purpose of the Study:

• To evaluate an algorithm designed for brain tumor segmentation with varying degrees of radiologist interaction.
• To assess the accuracy and efficiency of the interactive segmentation algorithm.

Main Methods:

• A computer simulation utilized the BraTS dataset (285 patients, multi-spectral MR) with ground-truth tumor delineations.
• Radiologist interaction (clicks and drags) was mimicked in real-time, guided by segmentation deviations from the ground-truth.
• Accuracy and time efficiency were measured across different interaction levels.

Main Results:

• The algorithm demonstrated varying levels of accuracy depending on the degree of radiologist interaction.
• The average total time for study loading to 3D contour confirmation was 46 seconds.
• Interactive segmentation significantly improved efficiency compared to traditional methods.

Conclusions:

• The developed algorithm offers an efficient and accurate approach to brain tumor volumetric segmentation.
• Real-time interaction allows for tailored radiologist involvement, optimizing both accuracy and speed.
• This method has the potential to streamline radiological workflows for brain tumor analysis.