MSA-DCNN: A Data-Efficient Multi-Scale Deformable CNN for Medical Image Classification

2026-07-07

AI Summary

This work addresses the challenges of multi-scale morphological variations and scarce annotated data in medical image classification, where existing methods struggle to simultaneously achieve adaptive sampling, effective multi-scale fusion, and label-efficient learning. We propose the MSA-DCNN framework, which uniquely integrates deformable convolutions with a multi-scale attention mechanism within a unified optimization paradigm. This integration enables adaptive multi-scale sampling, intra-scale saliency refinement, cross-scale feature fusion, and auxiliary self-distillation regularization. Evaluated on multiple public medical imaging benchmarks and an external leukemia test set, MSA-DCNN achieves significantly superior performance over Vision Transformers, conventional CNNs, and MICCAI semi-supervised baselines—using fewer parameters—while demonstrating exceptional generalization and data efficiency under distribution shifts and limited labeling scenarios.

Abstract

Existing deep learning methods perform well in medical image classification but struggle with multi-scale morphology and limited annotations due to fixed sampling and data-hungry training. Existing approaches address these challenges in isolation: DCN-based models provide adaptive sampling but lack explicit multi-scale attention fusion and label-efficient regularisation; multi-scale architectures typically rely on static fusion; and semi-supervised methods target label scarcity without jointly modelling adaptive cross-scale representations. We propose MSA-DCNN, a scale-consistent deformable attention learning framework that introduces adaptive multi-scale sampling, within-scale saliency refinement, learned cross-scale fusion, and auxiliary self-distillation within a unified optimisation scheme, with potential to generalise to structurally heterogeneous anatomy. We evaluate on three public benchmarks and an external hold-out set for leukaemia. MSA-DCNN demonstrates competitive and often better performance against ViT baselines, CNN baselines, and a MICCAI semi-supervised baseline under distribution shift and label scarcity in accuracy, F1, and AUC (binary), while using fewer parameters. Ablations confirm complementary component contributions, supporting MSA-DCNN as a practical foundation for data-efficient medical image classification.

Problem

• multi-scale morphology
• limited annotations
• medical image classification
• data efficiency
• adaptive sampling

Innovation

• multi-scale deformable CNN
• adaptive sampling
• cross-scale fusion
• self-distillation
• data-efficient learning