PruDensNet: a parameter efficient depthwise separable CNN for MRI-based brain tumor classification

1

PruDensNet is a depthwise-separable CNN designed for efficient brain tumor classification using MRI data, with approximately 1.46 million parameters.
2

The architecture incorporates lightweight attention mechanisms and GELU activations, optimizing for cost and latency-sensitive clinical deployments.
3

PruDensNet achieved a validation accuracy of 97.27% and test accuracy of 96.05% on a four-class brain tumor benchmark, outperforming other models.
4

The study emphasizes a robust data curation pipeline that standardizes labels and mitigates data leakage to enhance reproducibility.
5

A curriculum-regularization training approach was implemented, combining techniques like MixUp and CutMix to improve model generalization.

Frontiers In Medicine

by Mithila Arman, Ahnaf Samin, A. K. M. Muzahidul Islam, Md Maruf Rusafi Arnob, Md Jahirul Islam, Ishtiak Al Mamoon
May 20, 2026

1