Indian Journal of Science and Technology

Abstract

Objective: Automated segmentation is an active research for medical images. Accuracy of automated segmentation methods plays a vital role during brain image analysis. Segmentation being an important area of research, determining its performance is also important. Gold Standard is required for comparison during segmentation evaluation. Method: The Gold Standard for segmentation of medical images is the manual drawing of region of interest. This manual tracing is performed by experts (radiologists). The deviation of segmentation when compared with the experts and the quality of segmentation are inversely proportional. Analysis: The quantitative methods indicate the performance of the segmentation methods when compared with Gold Standard. Evaluation metrics mostly fall into three categories: Area Based Evaluation method (Dice coefficient, Jaccard Coefficient, Relative Volume Difference, Volume Overlap error), Surface Evaluation type (Average Symmetric Surface Distance, Root Mean Square Symmetric Surface Distance, Scatter Plot) and Specificity, Sensitivity and Accuracy.

Keywords: Gold Standard, Segmentation, MRI, Manual Segmentation, Automated Segmentation, Evaluation Metrics