Generalised Dice overlap as a deep learning loss function for highly unbalanced segmentations

Abstract

Deep-learning has proved in recent years to be a powerful tool for image analysis and is now widely used to segment both 2D and 3D medical images. Deep-learning segmentation frameworks rely not only on the choice of network architecture but also on the choice of loss function. When the segmentation process targets rare observations, a severe class imbalance is likely to occur between candidate labels, thus resulting in sub-optimal performance. In order to mitigate this issue, strategies such as the weighted cross-entropy function, the sensitivity function or the Dice loss function, have been proposed. In this work, we investigate the behavior of these loss functions and their sensitivity to learning rate tuning in the presence of different rates of label imbalance across 2D and 3D segmentation tasks. We also propose to use the class re-balancing properties of the Generalized Dice overlap, a known metric for segmentation assessment, as a robust and accurate deep-learning loss function for unbalanced tasks.

References in corpus (2)

• On the Compactness, Efficiency, and Representation of 3D Convolutional Networks: Brain Parcellation as a Pretext Task
• Deep Learning for Medical Image Segmentation

Cited by in corpus (41)

• Modified U-Net (mU-Net) with Incorporation of Object-Dependent High Level Features for Improved Liver and Liver-Tumor Segmentation in CT Images
• Association of genomic subtypes of lower-grade gliomas with shape features automatically extracted by a deep learning algorithm
• Attention-Guided Version of 2D UNet for Automatic Brain Tumor Segmentation
• Relational Modeling for Robust and Efficient Pulmonary Lobe Segmentation in CT Scans
• UNet++: Redesigning Skip Connections to Exploit Multiscale Features in Image Segmentation
• On the influence of Dice loss function in multi-class organ segmentation of abdominal CT using 3D fully convolutional networks
• Segmentation Loss Odyssey
• A comparative study of 2D image segmentation algorithms for traumatic brain lesions using CT data from the ProTECTIII multicenter clinical trial
• DARTS: DenseUnet-based Automatic Rapid Tool for brain Segmentation
• Brain Tumor Segmentation using 3D-CNNs with Uncertainty Estimation
• Generative Model-Based Ischemic Stroke Lesion Segmentation
• FU-net: Multi-class Image Segmentation Using Feedback Weighted U-net
• AdvSPADE: Realistic Unrestricted Attacks for Semantic Segmentation
• SegTHOR: Segmentation of Thoracic Organs at Risk in CT images
• Deep Residual 3D U-Net for Joint Segmentation and Texture Classification of Nodules in Lung
• LAMP: Large Deep Nets with Automated Model Parallelism for Image Segmentation
• Breast mass segmentation based on ultrasonic entropy maps and attention gated U-Net
• Globally Optimal Segmentation of Mutually Interacting Surfaces using Deep Learning
• Leishmaniasis Parasite Segmentation and Classification using Deep Learning
• DeepHIPS: A novel Deep Learning based Hippocampus Subfield Segmentation method
• OGNet: Salient Object Detection with Output-guided Attention Module
• Deep Learning for Automated Medical Image Analysis
• Flood Detection On Low Cost Orbital Hardware
• Confidence-guided Lesion Mask-based Simultaneous Synthesis of Anatomic and Molecular MR Images in Patients with Post-treatment Malignant Gliomas
• Learn distributed GAN with Temporary Discriminators
• Segmentation task for fashion and apparel
• AutoML Segmentation for 3D Medical Image Data: Contribution to the MSD Challenge 2018
• A Comparison of Deep Learning Convolution Neural Networks for Liver Segmentation in Radial Turbo Spin Echo Images
• On-the-fly Prediction of Protein Hydration Densities and Free Energies using Deep Learning
• Learning Non-Unique Segmentation with Reward-Penalty Dice Loss
• Disentangled Representations for Domain-generalized Cardiac Segmentation
• Lesion Mask-based Simultaneous Synthesis of Anatomic and MolecularMR Images using a GAN
• Segmenting Ships in Satellite Imagery With Squeeze and Excitation U-Net
• Efficient Scene Text Detection with Textual Attention Tower
• Organ At Risk Segmentation with Multiple Modality
• CSSegNet: Fine-Grained Cardiac Structures Segmentation Using Dilated Pyramid Pooling in U-net
• Real-time Segmentation and Facial Skin Tones Grading
• PGD-UNet: A Position-Guided Deformable Network for Simultaneous Segmentation of Organs and Tumors
• Multi-lane Detection Using Instance Segmentation and Attentive Voting
• Soft labeling by Distilling Anatomical knowledge for Improved MS Lesion Segmentation
• Technical report: Kidney tumor segmentation using a 2D U-Net followed by a statistical post-processing filter