Evaluating White Matter Lesion Segmentations with Refined Sørensen-Dice Analysis
Aaron Carass1,2, Snehashis Roy3, Adrian Gherman4
1Department of Electrical and Computer Engineering, The Johns Hopkins University, Baltimore, MD, 21218, USA. aaron_carass@jhu.edu.
Scientific Reports
|May 20, 2020
Summary
The Sørensen-Dice index (SDI) offers a refined method for analyzing medical image segmentation, especially when the number of objects is unknown. This approach improves the evaluation of segmentations for conditions like multiple sclerosis (MS) white matter lesions.
Area of Science:
- Medical Imaging
- Computer Vision
- Neurology
Background:
- The Sørensen-Dice index (SDI) is a standard metric for evaluating medical image segmentation accuracy.
- However, SDI provides limited insight when segmenting an unknown number of objects, such as white matter lesions in multiple sclerosis (MS).
Purpose of the Study:
- To present a refined method for analyzing SDI results in scenarios with an unknown number of objects.
- To demonstrate the utility of this refined analysis through case studies in MS lesion segmentation.
Main Methods:
- Developed a refined approach for finer-grained parsing of SDI results.
- Applied the refined method to two case studies: inter-rater comparison and algorithm fusion for MS lesion segmentation.
Main Results:
- The first case study revealed that smaller lesions are difficult to identify reliably.
- The second case study showed that fusing multiple segmentation algorithms, guided by the refined analysis, improved overall performance.
- Demonstrated that detailed SDI analysis yields significant insights into segmentation quality.
Conclusions:
- Refined SDI analysis enhances the understanding of medical image segmentation, particularly for unknown object counts.
- This method is valuable for evaluating and improving segmentation algorithms in clinical applications like MS lesion detection.
