ConvNeXt–KAN: A Kolmogorov–Arnold Network Head for Robust Multi-class Staging of Alzheimer's Disease from Brain MRI

Abstract

Alzheimer's disease (AD) is a progressive neurodegenerative disorder for which accurate, stage-wise classification from structural magnetic resonance imaging (MRI) can support earlier intervention. Convolutional neural networks (CNNs) dominate this task, but the classifier heads they use rely on fixed activation functions and linear weights, which constrains the modeling of the subtle, highly non-linear atrophy and texture patterns that separate dementia stages. In addition, the widely used Open Access Series of Imaging Studies (OASIS) collection is severely imbalanced, and two common practices—undersampling every class to the smallest one, and splitting at the image (slice) level— respectively discard most of the data and inflate reported accuracy. We address these issues with ConvNeXt–KAN, a hybrid model that couples a transfer-learned ConvNeXt feature extractor with a Kolmogorov–Arnold Network (KAN) classification head whose edges are learnable spline activations implemented efficiently with Gaussian radial basis functions. The model is trained with a robustness-oriented recipe (MixUp and CutMix, label smoothing, decoupled weight decay, a cosine learning-rate schedule with warm-up, and test-time augmentation) and a data strategy that uses far more of the imbalanced dataset while evaluating on a fixed, balanced, leakage-controlled test set. On the held-out test set the model attains 94.5% accuracy, a macro-averaged F1 of 0.944, and a mean one-vs-rest ROC–AUC of 0.991, with an ablation confirming that pretrained features and expanded data are the main drivers of performance. The learnable-activation head improves discrimination of the difficult non-demented versus very-mild-demented boundary, and the efficient formulation keeps the approach practical in standard deep-learning frameworks. To assess out-of-distribution robustness we further evaluate the trained model, without any fine-tuning, on an independent four-class Alzheimer's MRI collection distributed on Kaggle; test-time-augmented accuracy is 61.6% with a macro-averaged F1 of 0.565, and a class-wise analysis shows that moderate dementia continues to be recognized perfectly while the two clinically intermediate stages absorb most of the domain-shift error. We further quantify the impact of evaluation protocol and argue that subject-level validation is essential before clinical claims.