Tackling Bias in the Dice Similarity Coefficient: Introducing nDSC for White Matter Lesion Segmentation

by   Vatsal Raina, et al.

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 .


Simultaneous lesion and neuroanatomy segmentation in Multiple Sclerosis using deep neural networks

Segmentation of both white matter lesions and deep grey matter structure...

Refined-Segmentation R-CNN: A Two-stage Convolutional Neural Network for Punctate White Matter Lesion Segmentation in Preterm Infants

Accurate segmentation of punctate white matter lesion (PWML) in infantil...

Continuous Dice Coefficient: a Method for Evaluating Probabilistic Segmentations

Objective: Overlapping measures are often utilized to quantify the simil...

A Soft STAPLE Algorithm Combined with Anatomical Knowledge

Supervised machine learning algorithms, especially in the medical domain...

Spinal Cord Gray Matter-White Matter Segmentation on Magnetic Resonance AMIRA Images with MD-GRU

The small butterfly shaped structure of spinal cord (SC) gray matter (GM...

Soft labeling by Distilling Anatomical knowledge for Improved MS Lesion Segmentation

This paper explores the use of a soft ground-truth mask ("soft mask") to...

Spatially Covariant Lesion Segmentation

Compared to natural images, medical images usually show stronger visual ...

Please sign up or login with your details

Forgot password? Click here to reset