MDL-CA - a multimodal deep learning approach with a cross attention mechanism for accurate brain cancer diagnosis

Sep 22, 2026

Upload by Patrick

Brain cancer diagnosis is challenged by complex molecular mechanisms and anatomical abnormalities, while conventional approaches such as invasive biopsies, isolated genomic assays, and single-modality MRI often suffer from limited sensitivity, inadequate interpretability, and incomplete characterization of tumor heterogeneity. MDL-CA is proposed as a multimodal deep learning framework that integrates genomic and MRI data using a cross-attention fusion strategy. Genomic graph embeddings from a GAT and MRI feature maps from a 3D DenseNet are combined, with Entmax sigmoid improving sparsity and interpretability. Experiments across four benchmark datasets show accuracies between 96.22% and 98.46% and F1-scores from 95.95% to 98.40%.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Introduction

Methods

Results

Discussion

Keywords

TYPE Original Research PUBLISHED 05 January 2026 DOI 10.3389/fpubh.2025.1687335 OPEN ACCESS EDITED BY Yasin Kaya, Adana Alparslan Turkes Science and Technology University, Türkiye REVIEWED BY Elif Kevser Topuz Aydur, Adana Alparslan Turkes Science and Technology University, Türkiye Najam Aziz, Capital University of Science & Technology, Pakistan *CORRESPONDENCE Asif Nawaz [asif.nawaz@uaar.edu.pk](asif.nawaz@uaar.edu.pk)[ ](asif.nawaz@uaar.edu.pk)Seung Won Lee [swleemd@g.skku.edu](swleemd@g.skku.edu)[ ](swleemd@g.skku.edu)RECEIVED 17 August 2025 REVISED 27 November 2025 ACCEPTED 28 November 2025 PUBLISHED 05 January 2026 CITATION Sarwar S, Majeed S, Nawaz A, Bibi R and Lee SW (2026) MDL-CA: a multimodal deep learning approach with a cross attention mechanism for accurate brain cancer diagnosis. Front. Public Health 13:1687335 . doi: 10.3389/fpubh.2025.1687335 COPYRIGHT © 2026 Sarwar, Majeed, Nawaz, Bibi and Lee. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. MDL-CA: a multimodal deep learning approach with a cross attention mechanism for accurate brain cancer diagnosis Sumaira Sarwar 1, Saqib Majeed 1, Asif Nawaz 1*, Ruqia Bibi 1 and Seung Won Lee 2,3,4,5,6* 1University Institute of Information Technology, PMAS-Arid Agriculture University Rawalpindi, Rawalpindi, Pakistan, 2Department of Precision Medicine, Sungkyunkwan University School of Medicine, Suwon, Republic of Korea, 3Department of Artificial Intelligence, Sungkyunkwan University, Suwon, Republic of Korea, 4Department of MetaBioHealth, Sungkyunkwan University, Suwon, Republic of Korea, 5Personalized Cancer Immunotherapy Research Center, Sungkyunkwan University School of Medicine, Suwon, Republic of Korea, 6Department of Family Medicine, Kangbuk Samsung Hospital, Sungkyunkwan University School of Medicine, Seoul, Republic of Korea Introduction: Brain cancer diagnosis poses a significant clinical challenge due to the complex interplay between molecular mechanisms and anatomical abnormalities. Traditional diagnostic techniques, including invasive biopsies, isolated genomic assays, and standalone Magnetic Resonance Imaging (MRI), often exhibit limitations such as procedural risks, inadequate sensitivity, and incomplete assessment of tumor heterogeneity. These shortcomings contribute to delayed diagnosis, inaccurate tumor grading, and suboptimal treatment planning. Furthermore, single-modality data, whether MRI or genomic profiles, frequently yield limited diagnostic accuracy and biological interpretability. Methods: To address these limitations, this study proposes MDL-CA, a Multimodal Deep Learning framework with a Cross-Attention mechanism, designed to integrate genomic and MRI modalities for enhanced brain cancer diagnosis. The framework fuses genomic graph embeddings, extracted using a Graph Attention Network (GAT), with MRI feature maps derived from a 3D DenseNet. The cross-modal attention fusion mechanism enables the model to capture intricate biological and spatial interactions, producing a biologically informed feature representation. Additionally, the Entmax sigmoid function is employed in the classification stage to promote sparsity and improve interpretability. Data were sourced from The Cancer Imaging Archive (TCIA) and The Cancer Genome Atlas (TCGA) following comprehensive preprocessing. Results: Extensive experiments conducted across four benchmark datasets demonstrated that MDL-CA achieved superior diagnostic performance, with accuracies of 96. 22%, 97. 14%, 98.46%, and 98. 21%, and F1-scores ranging from 95.95% to

FAQs

What problem does MDL-CA address in brain cancer diagnosis?

It targets the limitations of traditional and single-modality approaches that struggle with sensitivity, interpretability, and capturing tumor heterogeneity from genomic and anatomical signals.

How does MDL-CA combine genomic and MRI information?

It fuses genomic graph embeddings obtained via a Graph Attention Network (GAT) with MRI feature maps derived from a 3D DenseNet using a cross-modal attention fusion mechanism.

What performance does MDL-CA achieve on benchmark datasets?

Across four benchmark datasets, MDL-CA reports accuracies of 96.22%, 97.14%, 98.46%, and 98.21%, with F1-scores ranging from 95.95% to 98.40%.

Sep 22, 2026

Upload by Patrick