EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Multimodal Deep Learning Framework for Identifying Discriminative Gene Expression Biomarkers and Enhancing ASD Diagnosis

Sarita Gajanan Rathod, Mansing Rathod

Journal of Intelligent Decision Making and Information Science · 27.08.2026 · 3 (9s) · 2263-2287

DOI: 10.17762/jidmis.v3.2544 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Intelligent Decision Making and Information Science, online erschienen am 27.08.2026. Verfasst von Sarita Gajanan Rathod und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Autism Spectrum Disorder is a complicated neurodevelopmental disorder that has both genetic and neurological factors, nevertheless, the current diagnosis is still mainly performed through behavioural assessments, which are subjective and hence, limit the possibility of early and precise detection. Current computational studies usually analyse either gene expression or neuroimaging data separately which leads to having limited diagnostic accuracy and a poor understanding of the biological processes involved. Additionally, previous models did not have solid biomarker refinement and failed to provide significant genetic–brain relationships. Our study aims to fill these gaps by presenting a multi-faceted deep learning framework that brings together gene expression markers from the GSE28521 dataset and structural and functional MRI features from the ABIDE dataset. One of the main principles behind this research is the dual-branch CNN-MLP architecture that allows the integration of gene profiles at the molecular level and brain imaging patterns at the brain level, which is backed by SHAP and Grad-CAM based interpretability tools. The model presented has really good performance metrics that include 96.67% accuracy, 93.33% precision, 100% recall, 0.9655 F1 score, and ROC-AUC above 0.92. SHAP analysis recognized the top-ranked gene biomarkers as the most discriminative, while Grad-CAM indicated the cortical and subcortical regions involved in ASD and hence confirmed the biological relevance of the model. By merging two complementary modalities, ASD research has reached a new height with this study as it presents a more accurate and biologically grounded diagnostic model as well as proves the potential of multimodal AI systems in early ASD detection and clinical use

Zitieren

APA

Rathod, S. G., & Rathod, M. (2026). Multimodal Deep Learning Framework for Identifying Discriminative Gene Expression Biomarkers and Enhancing ASD Diagnosis. Journal of Intelligent Decision Making and Information Science, 3(9s), 2263-2287. https://doi.org/10.17762/jidmis.v3.2544

Vancouver

Rathod SG, Rathod M. Multimodal Deep Learning Framework for Identifying Discriminative Gene Expression Biomarkers and Enhancing ASD Diagnosis. Journal of Intelligent Decision Making and Information Science. 2026;3(9s):2263-2287. doi:10.17762/jidmis.v3.2544

BibTeX
@article{Rathod2026,
  author = {Rathod, Sarita Gajanan and Rathod, Mansing},
  title = {Multimodal Deep Learning Framework for Identifying Discriminative Gene Expression Biomarkers and Enhancing ASD Diagnosis},
  journal = {Journal of Intelligent Decision Making and Information Science},
  year = {2026},
  volume = {3},
  number = {9s},
  pages = {2263--2287},
  doi = {10.17762/jidmis.v3.2544}
}
Kostenlos starten →