Analyzing the Effect of $k$-Space Features in MRI Classification Models

TL;DR

This paper presents an explainable AI approach for MRI classification that combines image and frequency domain features, utilizing UMAP for visualization to improve interpretability and diagnostic trust.

Contribution

The study introduces a novel CNN-based method integrating $k$-space features and UMAP visualization to enhance interpretability in MRI classification models.

Findings

Improved early training efficiency with combined features

Enhanced understanding of feature impact on predictions

Increased interpretability of MRI classification models

Abstract

The integration of Artificial Intelligence (AI) in medical diagnostics is often hindered by model opacity, where high-accuracy systems function as "black boxes" without transparent reasoning. This limitation is critical in clinical settings, where trust and reliability are paramount. To address this, we have developed an explainable AI methodology tailored for medical imaging. By employing a Convolutional Neural Network (CNN) that analyzes MRI scans across both image and frequency domains, we introduce a novel approach that incorporates Uniform Manifold Approximation and Projection UMAP] for the visualization of latent input embeddings. This approach not only enhances early training efficiency but also deepens our understanding of how additional features impact the model predictions, thereby increasing interpretability and supporting more accurate and intuitive diagnostic inferences