Tackling Bias in the Dice Similarity Coefficient: Introducing nDSC for White Matter Lesion Segmentation
| Autor: | Raina, Vatsal, Molchanova, Nataliia, Graziani, Mara, Malinin, Andrey, Muller, Henning, Cuadra, Meritxell Bach, Gales, Mark |
|---|---|
| Rok vydání: | 2023 |
| Předmět: | |
| Druh dokumentu: | Working Paper |
| Popis: | The development of automatic segmentation techniques for medical imaging tasks requires assessment metrics to fairly judge and rank such approaches on benchmarks. The Dice Similarity Coefficient (DSC) is a popular choice for comparing the agreement between the predicted segmentation against a ground-truth mask. However, the DSC metric has been shown to be biased to the occurrence rate of the positive class in the ground-truth, and hence should be considered in combination with other metrics. This work describes a detailed analysis of the recently proposed normalised Dice Similarity Coefficient (nDSC) for binary segmentation tasks as an adaptation of DSC which scales the precision at a fixed recall rate to tackle this bias. White matter lesion segmentation on magnetic resonance images of multiple sclerosis patients is selected as a case study task to empirically assess the suitability of nDSC. We validate the normalised DSC using two different models across 59 subject scans with a wide range of lesion loads. It is found that the nDSC is less biased than DSC with lesion load on standard white matter lesion segmentation benchmarks measured using standard rank correlation coefficients. An implementation of nDSC is made available at: https://github.com/NataliiaMolch/nDSC . Comment: 5 pages, 5 figures, accepted at ISBI 2023 |
| Databáze: | arXiv |
| Externí odkaz: |
|