Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Selecting the Best Sequential Transfer Path for Medical Image Segmentation with Limited Labeled Data

Jingyun Yang, Jingge Wang, Guoqing Zhang, Yang Li

TL;DR

A novel sequential transfer scheme with a task affinity metric tailored for medical images, which develops an effective sequential transfer strategy by incorporating intermediate source tasks to gradually narrow the domain discrepancy and minimize the transfer cost.

Abstract

The medical image processing field often encounters the critical issue of scarce annotated data. Transfer learning has emerged as a solution, yet how to select an adequate source task and effectively transfer the knowledge to the target task remains challenging. To address this, we propose a novel sequential transfer scheme with a task affinity metric tailored for medical images. Considering the characteristics of medical image segmentation tasks, we analyze the image and label similarity between tasks and compute the task affinity scores, which assess the relatedness among tasks. Based on this, we select appropriate source tasks and develop an effective sequential transfer strategy by incorporating intermediate source tasks to gradually narrow the domain discrepancy and minimize the transfer cost. Thereby we identify the best sequential transfer path for the given target task. Extensive experiments on three MRI medical datasets, FeTS 2022, iSeg-2019, and WMH, demonstrate the efficacy of our method in finding the best source sequence. Compared with directly transferring from a single source task, the sequential transfer results underline a significant improvement in target task performance, achieving an average of 2.58% gain in terms of segmentation Dice score, notably, 6.00% for FeTS 2022. Code is available at the git repository.

Selecting the Best Sequential Transfer Path for Medical Image Segmentation with Limited Labeled Data

TL;DR

A novel sequential transfer scheme with a task affinity metric tailored for medical images, which develops an effective sequential transfer strategy by incorporating intermediate source tasks to gradually narrow the domain discrepancy and minimize the transfer cost.

Abstract

The medical image processing field often encounters the critical issue of scarce annotated data. Transfer learning has emerged as a solution, yet how to select an adequate source task and effectively transfer the knowledge to the target task remains challenging. To address this, we propose a novel sequential transfer scheme with a task affinity metric tailored for medical images. Considering the characteristics of medical image segmentation tasks, we analyze the image and label similarity between tasks and compute the task affinity scores, which assess the relatedness among tasks. Based on this, we select appropriate source tasks and develop an effective sequential transfer strategy by incorporating intermediate source tasks to gradually narrow the domain discrepancy and minimize the transfer cost. Thereby we identify the best sequential transfer path for the given target task. Extensive experiments on three MRI medical datasets, FeTS 2022, iSeg-2019, and WMH, demonstrate the efficacy of our method in finding the best source sequence. Compared with directly transferring from a single source task, the sequential transfer results underline a significant improvement in target task performance, achieving an average of 2.58% gain in terms of segmentation Dice score, notably, 6.00% for FeTS 2022. Code is available at the git repository.

Paper Structure

This paper contains 17 sections, 9 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Problem Definition
  4. Task Relatedness Assessment
  5. Image Similarity Analysis.
  6. Label Similarity Analysis.
  7. Task Affinity Estimation.
  8. Graph Clustering.
  9. Sequential Transfer Path Selection
  10. Experiments and Results
  11. Datasets
  12. Training Setup
  13. Performance Evaluation
  14. Effectiveness of Sequential Transfer.
  15. When does sequential transfer work?
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Challenges in transfer learning for medical image processing tasks: available source tasks are either large-scale but distant from the target task or closely related to the target task but limited in volume. (a) depicts the traditional direct transfer. (b) shows the proposed sequential transfer.
  • Figure 2: Illustraintion of the proposed sequential transfer framework. (a) depicts the graph we construct on source tasks, where edges signify affinity between these medical image processing tasks. (b) groups closely related nodes into clusters. The circled nodes are the representatives of each cluster, which indicate the most generalizable source datasets. (c) illustrates the most effective sequential transfer path we select for a given target task, represented by the bold red line.
  • Figure 3: Diagram of task relatedness assessment. When a task is represented by a node, the node size indicates the task dataset size while edges signify task affinity.