Hierarchical Spatio-Channel Clustering for Efficient Model Compression in Medical Image Analysis
For practitioners deploying CNNs in resource-constrained medical imaging settings, this method offers a more effective compression technique that improves both efficiency and accuracy over standard low-rank methods.
The paper introduces a hierarchical spatio-channel clustering method for compressing CNNs, which partitions feature maps into spatial regions and groups channels by co-activation patterns before applying rank-adaptive SVD. On an AlexNet-based brain tumor MRI classifier, it achieves 81.1% FLOPs reduction, 1.38x speed-up, and accuracy improvement from 87.76% to 89.80% under 3x and 6x compression budgets.
Convolutional neural networks (CNNs) have become increasingly difficult to deploy in resource-constrained environments due to their large memory and computational requirements. Although low-rank compression methods can reduce this burden, most existing approaches compress spatial and channel redundancy independently and therefore do not fully exploit the localised structure within convolutional feature maps. This paper proposes a hierarchical spatio-channel low-rank compression framework for CNNs that exploits redundancy across spatial regions and channel activations. Unlike conventional methods, which apply a uniform decomposition across an entire layer, the proposed approach first partitions feature maps into spatial regions, then groups channels according to their co-activation patterns within each region, and finally applies rank-adaptive SVD to each resulting spatio-channel cluster. The method is evaluated on an AlexNet-based brain tumour MRI classification model and compared with Global SVD and Tucker decomposition under \(3\times\) and \(6\times\) compression budgets. Our method outperforms both baselines, reducing FLOPs from \(8.21\,\mathrm{G}\) to \(1.55\,\mathrm{G}\) (\(81.1\%\) reduction), achieving a \(1.38\times\) inference speed-up, and increasing classification accuracy from \(87.76\%\) to \(89.80\%\). The method also improves the macro \(F_1\)-score and performance on challenging classes such as meningioma. A hyper-parameter trade-off analysis demonstrates that the framework provides Pareto-optimal configurations, enabling control over the balance between compression and predictive performance. Moderate clustering with adaptive rank selection yields strong results. Bootstrap standard errors are reported for all classification metrics.