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

Jul 16, 2026

Upload by Miles

Deep learning models for medical image classification can be limited by fixed sampling, multi-scale morphological complexity, and scarce annotations, which weakens performance when anatomy varies structurally across resolutions. MSA-DCNN introduces a scale-consistent deformable attention framework with adaptive multi-scale sampling, within-scale saliency refinement, learned cross-scale fusion, and auxiliary self-distillation under a unified optimisation scheme. Experiments on three public benchmarks plus an external leukemia hold-out show improved accuracy, F1, and binary AUC under distribution shift and label scarcity while using fewer parameters, supported by ablations.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

Introduction

Methodology

arXiv :2607 .06083v 1 [ cs .CV] 7 Jul 2026 MSA-DCNN: A Data-Efficient Multi-Scale Deformable CNN for Medical Image Classification Hamza Hussaini 1[0009−0005−6463−1607], Shahana Bano 1[0000−0002−6907−1777], Eyad Elyan 1[0000−0002−8342−9026], and Carlos Francisco Moreno-García 1[0000−0001−7218−9023] Robert Gordon University, School of Computing, Engineering and Technology, Aberdeen, Scotland, UK h.hussaini;s.bano;e.elyan;[c.moreno-garcia@rgu.ac.uk](c.moreno-garcia@rgu.ac.uk)[ ](c.moreno-garcia@rgu.ac.uk)[http://www.rgu.ac.uk](http://www.rgu.ac.uk) 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 MSADCNN, 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. Keywords: Multi-scale attention learning · Deformable convolutional networks · Data-efficient medical image classification · Scale-consistent feature fusion · Self-distillation learning. 1 Introduction Automated medical image classification is a key enabler of computational healthcare, delivering fast, objective, and reproducible analysis across radiography, histopathology, and microscopy [12,18] . Convolutional neural networks (CNNs) remain the workhorse due to their hierarchical feature learning and strong performance across lung, brain, oncology, and haematology tasks [29,6,20,16,17] . Yet, 2 Hussaini et al. standard CNNs inherit fixed receptive fields and uniform sampling, which are often poorly matched to the heterogeneous textures and multi-scale structural variability of medical imagery [28,24,26] . Prior remedies, including attention modulesand multi-scale designs, improve feature focus and coverage [21,8,10,30], but often rely on global aggregation that can wash out subtle cues [9], retain fixed kernels that do not adapt sampling across scales [27], and rarely provide internal supervision to strengthen shallow layers in data-limited regimes [13] . Moreover, medical image classification is frequently constrained by limited labelled data [11,14,15,3], limiting generalisation and sensitivity to minority patterns. We address these gaps with the Multi-Scale Attention Deformable Convolutional Neural Network (MSA-DCNN) . This work introduces a general principle for scale-consistent learning under structural heterogeneity and label scarcity, designed to perform effectively at low label fractions with a lean parameter budget while enforcing semantic alignment across resolution levels. Critically, attention is applied scale-specifically, and the resulting features are fused by a learned multi-scale attention mechanism, coupling within-scale recalibration with across-scale selection, an integration not explicitly studied in prior deformable or attention-based CNNs. We evaluate MSA-DCNN on three publicly available benchmark

FAQs

What problem does MSA-DCNN address in medical image classification?

It targets poor matching between fixed receptive fields and heterogeneous, multi-scale structures, alongside limited labeled data that harms generalisation and sensitivity to minority patterns.

How does MSA-DCNN perform multi-scale feature learning?

It uses adaptive multi-scale deformable sampling, then refines saliency within each scale and performs learned cross-scale fusion through multi-scale attention over projected embeddings.

How was MSA-DCNN evaluated and what were the main results?

MSA-DCNN was evaluated on three public benchmarks and an external leukemia hold-out set, achieving competitive or better performance than ViT, CNN, and a MICCAI semi-supervised baseline under distribution shift and label scarcity on accuracy, F1, and binary AUC while using fewer parameters.

Jul 16, 2026

Upload by Miles