<p>Batch Normalisation (BN) is a widely adopted method in deep learning, but it suffers from severe limitations in scenarios involving small batch sizes, such as in medical image classification tasks. BN’s reliance on batch statistics leads to unstable training and unreliable inference, especially in high-resolution image domains with memory constraints. To overcome this, we propose a novel application of Group Normalisation (GN) tailored for small-batch medical imaging, such as Alzheimer’s disease classification. Unlike BN, GN divides channels into groups and performs normalisation within each group, making it independent of batch size and more robust across training and inference. Furthermore, compared to Instance Normalization (IN), which normalizes each instance separately, our GN method maintains better generalization while being computationally efficient and scalable. We evaluate our GN enhanced ResNet-50 model on an Alzheimer’s dataset and demonstrate superior performance, achieving a 95.5% classification accuracy. Our method significantly improves sensitivity, specificity, and Matthews Correlation Coefficient (MCC) compared to existing normalization approaches. These results affirm the novelty and effectiveness of GN in challenging low-batch, high-resolution medical imaging tasks.</p>