Progressive Growing of Patch Size: Curriculum Learning for Accelerated and Improved Medical Image Segmentation

2025-10-27

AI Summary

To address inefficiency in training, sensitivity to class imbalance, and limited generalization caused by fixed patch sizes in 3D medical image segmentation, this paper proposes a progressive patch-size enlargement strategy based on automatic curriculum learning. The method dynamically adjusts input patch dimensions during training—without manual curriculum design—and is compatible with diverse architectures including UNet, UNETR, and SwinUNETR. Evaluated across 15 public benchmarks, it achieves two operational modes: under *resource mode*, training time reduces to 44% of baseline while maintaining performance; under *performance mode*, mean Dice improves by 1.28% with only 89% training time, yielding greater gains for long-tail tasks such as lesion segmentation. The core innovation lies in the first formulation of patch-size growth as an adaptive curriculum learning process, which accelerates convergence, mitigates class imbalance, significantly reduces result variance, and enhances cross-task comparability.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience

Abstract

In this work, we introduce Progressive Growing of Patch Size, an automatic curriculum learning approach for 3D medical image segmentation. Our approach progressively increases the patch size during model training, resulting in an improved class balance for smaller patch sizes and accelerated convergence of the training process. We evaluate our curriculum approach in two settings: a resource-efficient mode and a performance mode, both regarding Dice score performance and computational costs across 15 diverse and popular 3D medical image segmentation tasks. The resource-efficient mode matches the Dice score performance of the conventional constant patch size sampling baseline with a notable reduction in training time to only 44%. The performance mode improves upon constant patch size segmentation results, achieving a statistically significant relative mean performance gain of 1.28% in Dice Score. Remarkably, across all 15 tasks, our proposed performance mode manages to surpass the constant patch size baseline in Dice Score performance, while simultaneously reducing training time to only 89%. The benefits are particularly pronounced for highly imbalanced tasks such as lesion segmentation tasks. Rigorous experiments demonstrate that our performance mode not only improves mean segmentation performance but also reduces performance variance, yielding more trustworthy model comparison. Furthermore, our findings reveal that the proposed curriculum sampling is not tied to a specific architecture but represents a broadly applicable strategy that consistently boosts performance across diverse segmentation models, including UNet, UNETR, and SwinUNETR. In summary, we show that this simple yet elegant transformation on input data substantially improves both Dice Score performance and training runtime, while being compatible across diverse segmentation backbones.

Problem

• Improves medical image segmentation via progressive patch size growth
• Accelerates training convergence while enhancing class balance
• Achieves higher Dice scores with reduced computational costs

Innovation

• Progressively increases patch size during training
• Improves class balance and accelerates convergence
• Applicable across diverse segmentation model architectures