Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Site Class-Incremental Learning with Weighted Experts in Echocardiography

Kit M. Bransby, Woo-jin Cho Kim, Jorge Oliveira, Alex Thorley, Arian Beqiri, Alberto Gomez, Agisilaos Chartsias

TL;DR

A class-incremental learning method which learns an expert network for each dataset, and combines all expert networks with a score fusion model, which minimises the influence of ``unqualified experts'' and promotes transparency during inference.

Abstract

Building an echocardiography view classifier that maintains performance in real-life cases requires diverse multi-site data, and frequent updates with newly available data to mitigate model drift. Simply fine-tuning on new datasets results in "catastrophic forgetting", and cannot adapt to variations of view labels between sites. Alternatively, collecting all data on a single server and re-training may not be feasible as data sharing agreements may restrict image transfer, or datasets may only become available at different times. Furthermore, time and cost associated with re-training grows with every new dataset. We propose a class-incremental learning method which learns an expert network for each dataset, and combines all expert networks with a score fusion model. The influence of ``unqualified experts'' is minimised by weighting each contribution with a learnt in-distribution score. These weights promote transparency as the contribution of each expert is known during inference. Instead of using the original images, we use learned features from each dataset, which are easier to share and raise fewer licensing and privacy concerns. We validate our work on six datasets from multiple sites, demonstrating significant reductions in training time while improving view classification performance.

Multi-Site Class-Incremental Learning with Weighted Experts in Echocardiography

TL;DR

A class-incremental learning method which learns an expert network for each dataset, and combines all expert networks with a score fusion model, which minimises the influence of ``unqualified experts'' and promotes transparency during inference.

Abstract

Building an echocardiography view classifier that maintains performance in real-life cases requires diverse multi-site data, and frequent updates with newly available data to mitigate model drift. Simply fine-tuning on new datasets results in "catastrophic forgetting", and cannot adapt to variations of view labels between sites. Alternatively, collecting all data on a single server and re-training may not be feasible as data sharing agreements may restrict image transfer, or datasets may only become available at different times. Furthermore, time and cost associated with re-training grows with every new dataset. We propose a class-incremental learning method which learns an expert network for each dataset, and combines all expert networks with a score fusion model. The influence of ``unqualified experts'' is minimised by weighting each contribution with a learnt in-distribution score. These weights promote transparency as the contribution of each expert is known during inference. Instead of using the original images, we use learned features from each dataset, which are easier to share and raise fewer licensing and privacy concerns. We validate our work on six datasets from multiple sites, demonstrating significant reductions in training time while improving view classification performance.
Paper Structure (11 sections, 2 equations, 4 figures, 3 tables)

This paper contains 11 sections, 2 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Problem Setting
  4. Score Fusion
  5. Weighted Score Fusion
  6. Multi-Site Training
  7. Experiments & Results
  8. Datasets
  9. Implementation & Training
  10. Comparison to Existing Methods & Ablation Study
  11. Conclusion

Figures (4)

  • Figure 1: Examples of different views from WASE, CAMUS, Medstar, StG datasets.
  • Figure 2: Network architecture: predictions from expert branches are re-weighted by an in-distribution score to minimise the influence of unqualified experts.
  • Figure 3: Distribution of attention scores across a selection of test sets using attn-wSF (top row) and nmd-wSF (bottom row). Note when using nmd, the attention scores are pushed towards 0 and 1 which may reduce generalisability
  • Figure 4: Efficiency analysis: Cumulative training time (a) and inference time (b).