Generalized Overlap Measures for Evaluation and Validation in Medical Image Analysis

By William R. Crum, Oscar Camara, Derek L. G. Hill. 2006

770 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Evaluation & Benchmarks, Medical Imaging

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

In Medical Imaging

Citations in the corpus

V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation (2016, methodology): “Compared to this, we introduce a novel loss layer specifically designed for segmentation tasks, which is based on Dice coefficient, one of the most common measures of region overlap in medical image analysis [4].”
Making the Most of Text Semantics to Improve Biomedical Vision-Language Processing (2022, methodology): “We report segmentation results via mean intersection over union (mIoU) and contrast-to-noise ratio (CNR), and report the Dice score [10] to compare to [56].”
Generalised Dice Overlap as a Deep Learning Loss Function for Highly Unbalanced Segmentations (2017, methodology): “As stated in [2], when choosing the GDLv weighting, the contribution of each label is corrected by the inverse of its volume, thus reducing the well known correlation between region size and Dice score.”
ResUNet-a: a deep learning framework for semantic segmentation of remotely sensed data (2019, background): “In order to demonstrate the performance difference between the Dice loss as defined in [13] and the Tanimoto loss, and the Tanimoto with dual (Eq.”
Reducing the Hausdorff Distance in Medical Image Segmentation With Convolutional Neural Networks (2019, methodology): “The performance of automatic segmentation methods is usually evaluated by computing some common objective criteria such as the Dice similarity coefficient (DSC), mean boundary distance, volume difference or overlap, and Hausdorff Distance (HD) [5], [6].”
Issues with threshold masking in voxel-based morphometry of atrophied brains (2009, methodology): “By considering the simple Jaccard Similarity coefficient (Crum et al., 2006) between the binary (maximum probability) model of grey matter B and the estimated probabilistic segmentation S after binarisation at a particular threshold T , the optimal threshold may then be found as T * = arg max T…”
Automated Volumetry and Regional Thickness Analysis of Hippocampal Subfields and Medial Temporal Cortical Structures in Mild Cognitive Impairment (2014, methodology): “Additionally, we measured the generalized Dice coefficient (GDSC), an overall measure of agreement between multilabel segmentations [Crum et al., 2006].”
Joint Optic Disc and Cup Segmentation Based on Multi-Label Deep Network and Polar Transformation (2018, methodology): “Dice coefficient is a measure of overlap widely used to assess segmentation performance when the ground truth is available [35].”
Unified Focal loss: Generalising Dice and cross entropy-based losses to handle class imbalanced medical image segmentation (2021, methodology): “…and false negatives (FN): The Dice loss ( L DSC ), can therefore be defined as: Other variants of the Dice loss include the Generalised Dice loss (Crum et al., 2006; Sudre et al., 2017) where the class weights are corrected by the inverse of their volume, and the Generalised Wasser-stein Dice…”
A survey of MRI-based brain tumor segmentation methods (2014, methodology): “In the ﬁeld of brain tumor segmentation, the Jaccard coefﬁcient and the Dice Similarity Coefﬁcient (DSC) are the most commonly used evaluation standards [128] .”
Metrics for evaluating 3D medical image segmentation: analysis, selection, and tool (2015, background): “Therefore, it does not provide additional information to select both of them together as validation metrics as done in [15–17].”
Volumetric Navigators (vNavs) for Prospective Motion Correction and Selective Reacquisition in Neuroanatomical MRI (2011, methodology): “Given two regions, A and B, and using |A| and |B| to represent the number of voxels in region A and B, respectively, the Dice coefficient is (35)”
CE-Net: Context Encoder Network for 2D Medical Image Segmentation (2019, methodology): “paper, we use the Dice coefficient loss function [57], [58] to replace the common cross entropy loss.”
U-Sleep: resilient high-frequency sleep staging (2021, positive): “The more common and simpler cross-entropy cost function was optimized instead of the generalized dice loss 53,54 used for U-Time.”
Disc-Aware Ensemble Network for Glaucoma Screening From Fundus Image (2018, background): “The other is the Dice coefﬁcient for assessing disc segmentation [32], which is deﬁned as: where N is the pixel number, p i ∈ [0 , 1] and g i ∈ { 0 , 1 } denote predicted probability and binary ground truth label for disc region, respectively.”