Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Divergence Learning for Missing-Modality Segmentation

Runze Cheng, Zhongao Sun, Ye Zhang, Chun Li

TL;DR

A novel single-modality parallel processing network framework based on Hölder divergence and mutual information is introduced and demonstrates that this method outperforms existing techniques in handling missing modalities and validates each component's effectiveness.

Abstract

Multimodal Magnetic Resonance Imaging (MRI) provides essential complementary information for analyzing brain tumor subregions. While methods using four common MRI modalities for automatic segmentation have shown success, they often face challenges with missing modalities due to image quality issues, inconsistent protocols, allergic reactions, or cost factors. Thus, developing a segmentation paradigm that handles missing modalities is clinically valuable. A novel single-modality parallel processing network framework based on Hölder divergence and mutual information is introduced. Each modality is independently input into a shared network backbone for parallel processing, preserving unique information. Additionally, a dynamic sharing framework is introduced that adjusts network parameters based on modality availability. A Hölder divergence and mutual information-based loss functions are used for evaluating discrepancies between predictions and labels. Extensive testing on the BraTS 2018 and BraTS 2020 datasets demonstrates that our method outperforms existing techniques in handling missing modalities and validates each component's effectiveness.

Robust Divergence Learning for Missing-Modality Segmentation

TL;DR

A novel single-modality parallel processing network framework based on Hölder divergence and mutual information is introduced and demonstrates that this method outperforms existing techniques in handling missing modalities and validates each component's effectiveness.

Abstract

Multimodal Magnetic Resonance Imaging (MRI) provides essential complementary information for analyzing brain tumor subregions. While methods using four common MRI modalities for automatic segmentation have shown success, they often face challenges with missing modalities due to image quality issues, inconsistent protocols, allergic reactions, or cost factors. Thus, developing a segmentation paradigm that handles missing modalities is clinically valuable. A novel single-modality parallel processing network framework based on Hölder divergence and mutual information is introduced. Each modality is independently input into a shared network backbone for parallel processing, preserving unique information. Additionally, a dynamic sharing framework is introduced that adjusts network parameters based on modality availability. A Hölder divergence and mutual information-based loss functions are used for evaluating discrepancies between predictions and labels. Extensive testing on the BraTS 2018 and BraTS 2020 datasets demonstrates that our method outperforms existing techniques in handling missing modalities and validates each component's effectiveness.

Paper Structure

This paper contains 13 sections, 5 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Knowledge Distillation for Segmentation Using Hölder Divergence
  4. High Mutual Information Knowledge Transfer Learning
  5. Overall Framework
  6. Experiments and Analysis
  7. Datasets and Evaluation Metrics
  8. Training Details
  9. Compare Experimental Models
  10. Exploration of the Superiority of Hölder Divergence
  11. Exploring the Impact of Hölder Conjugate Exponents on Experimental Results
  12. Ablation Study
  13. Conclusion

Figures (2)

  • Figure 1: The Framework of Robust Divergence Learning for Missing-Modality Segmentation. This figure illustrates the overall structure of the proposed robust divergence learning approach, specifically designed to address segmentation challenges in scenarios where certain modalities are missing.
  • Figure 2: This figure presents the segmentation results of three models on the BraTS 2018 dataset using different modality inputs. The second row shows the reproduced results of the M3AE, the third row shows the reproduced results of the GGMD, and the fourth row displays the results of ours. Each column represents different input settings: the first four columns show the results for single modality inputs (T1, T1ce, T2, and Flair, respectively), the fifth column displays the results using all four modalities as input simultaneously, and the last column shows the corresponding ground truth.

Theorems & Definitions (1)

  • Definition 1