Unlocking Robust Segmentation Across All Age Groups via Continual Learning

TL;DR

This paper addresses the generalization gap between adult-trained CT organ segmentation and pediatric data. It systematically evaluates TotalSegmentator on pediatric CT, analyzes simple domain adaptation via preprocessing, and introduces continual learning with rehearsal to preserve adult performance while learning pediatric anatomy. The rehearsal approach achieves the best cross-age accuracy (mean DSC of 0.84 in pediatrics and 0.90 in adults), outperforming adult-only baselines and basic augmentation. The results demonstrate that a continual learning strategy can deliver robust, age-inclusive segmentation for whole-body CT with 19 organ labels, enabling safer and more equitable clinical deployment.

Abstract

Most deep learning models in medical imaging are trained on adult data with unclear performance on pediatric images. In this work, we aim to address this challenge in the context of automated anatomy segmentation in whole-body Computed Tomography (CT). We evaluate the performance of CT organ segmentation algorithms trained on adult data when applied to pediatric CT volumes and identify substantial age-dependent underperformance. We subsequently propose and evaluate strategies, including data augmentation and continual learning approaches, to achieve good segmentation accuracy across all age groups. Our best-performing model, trained using continual learning, achieves high segmentation accuracy on both adult and pediatric data (Dice scores of 0.90 and 0.84 respectively).

Figures (2)

• Figure 1: (a) Age distribution of datasets. (b) Continual Learning with Rehearsal.
• Figure 2: (a) Performance across age groups. (b) We observed clear segmentation errors using TotalSegmentator (TS) errors markedly improved using our proposed approach (CL).