Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Distance Map Regression Network with Shape-aware Loss for Imbalanced Medical Image Segmentation

Huiyu Li, Xiabi Liu, Said Boumaraf, Xiaopeng Gong, Donghai Liao, Xiaohong Ma

TL;DR

This paper tackles the challenge of segmenting small medical objects under severe class imbalance by introducing a distance-map ground truth and a tandem network that regresses this map. The main segmentation network (M-Net) is paired with a light-weight regression network (LR-Net) to convert segmentation into a distance-map regression task, guided by a shape-aware MapDice loss that emphasizes boundaries. Experiments on the LiTS dataset and a clinical CT tumor dataset show improved performance over baseline imbalance methods and state-of-the-art approaches, with better boundary fidelity and phase-consistent shapes. The approach provides differentiable distance-map computation, enabling mutual reinforcement between segmentation and shape priors, and demonstrates practical impact for liver tumor analysis.

Abstract

Small object segmentation, like tumor segmentation, is a difficult and critical task in the field of medical image analysis. Although deep learning based methods have achieved promising performance, they are restricted to the use of binary segmentation mask. Inspired by the rigorous mapping between binary segmentation mask and distance map, we adopt distance map as a novel ground truth and employ a network to fulfill the computation of distance map. Specially, we propose a new segmentation framework that incorporates the existing binary segmentation network and a light weight regression network (dubbed as LR-Net). Thus, the LR-Net can convert the distance map computation into a regression task and leverage the rich information of distance maps. Additionally, we derive a shape-aware loss by employing distance maps as penalty map to infer the complete shape of an object. We evaluated our approach on MICCAI 2017 Liver Tumor Segmentation (LiTS) Challenge dataset and a clinical dataset. Experimental results show that our approach outperforms the classification-based methods as well as other existing state-of-the-arts.

Deep Distance Map Regression Network with Shape-aware Loss for Imbalanced Medical Image Segmentation

TL;DR

This paper tackles the challenge of segmenting small medical objects under severe class imbalance by introducing a distance-map ground truth and a tandem network that regresses this map. The main segmentation network (M-Net) is paired with a light-weight regression network (LR-Net) to convert segmentation into a distance-map regression task, guided by a shape-aware MapDice loss that emphasizes boundaries. Experiments on the LiTS dataset and a clinical CT tumor dataset show improved performance over baseline imbalance methods and state-of-the-art approaches, with better boundary fidelity and phase-consistent shapes. The approach provides differentiable distance-map computation, enabling mutual reinforcement between segmentation and shape priors, and demonstrates practical impact for liver tumor analysis.

Abstract

Small object segmentation, like tumor segmentation, is a difficult and critical task in the field of medical image analysis. Although deep learning based methods have achieved promising performance, they are restricted to the use of binary segmentation mask. Inspired by the rigorous mapping between binary segmentation mask and distance map, we adopt distance map as a novel ground truth and employ a network to fulfill the computation of distance map. Specially, we propose a new segmentation framework that incorporates the existing binary segmentation network and a light weight regression network (dubbed as LR-Net). Thus, the LR-Net can convert the distance map computation into a regression task and leverage the rich information of distance maps. Additionally, we derive a shape-aware loss by employing distance maps as penalty map to infer the complete shape of an object. We evaluated our approach on MICCAI 2017 Liver Tumor Segmentation (LiTS) Challenge dataset and a clinical dataset. Experimental results show that our approach outperforms the classification-based methods as well as other existing state-of-the-arts.
Paper Structure (15 sections, 2 equations, 3 figures, 2 tables)

This paper contains 15 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Distance Map
  4. Network Architecture
  5. The Main Segmentation Network.
  6. The Light Weight Regression Network.
  7. The Training and Inference Pipeline.
  8. Loss Functions
  9. Experiments
  10. Dataset and Preprocessing.
  11. Implementation Details.
  12. Evaluation Criteria.
  13. Comparison of Various Distance Maps Regression
  14. Effectiveness of the M-Net with LR-Net
  15. Conclusion

Figures (3)

  • Figure 1: Illustration of the main segmentation network with the LR-Net.
  • Figure 2: An illustration of different kinds of distance maps.
  • Figure 3: Fig. 3. Liver tumor segmentation results of different methods on the LiTS validation dataset. From left to right, ground truth, segmentation results by MNet+${L_{Dice}}$, MNet+${L_{MapDice}}$, MNet+LR-Net+${L_{smoothL1}}$, MNet+${L_{Dice}}$+LR-Net+${\alpha\cdot{L_{smoothL1}}}$, MNet+${L_{MapDice}}$+LR-Net+${\alpha\cdot{L_{smoothL1}}}$ are shown respectively.