Home / Current Issue / Paper 1711765
A Unified Multi-Modal Transformer Framework for Synergistic Cancer Diagnosis
Subject area: Science,Engineering and Technology · Area of research: Multi-modal Learning, Precision Medicine
Abstract
Early cancer diagnosis is critical for improving patient outcomes but is challenged by the disease?s profound heterogeneity. This paper introduces a unified, AI-powered framework that synergistically integrates histopathology, genomics, and proteomics data to enhance early cancer detection. Our architecture features a novel multi-transformer model with dedicated Vision and Genomic Transformers to encode modality-specific features, which are then fused by a cross-modal attention transformer. This intermediate fusion strategy enables the model to learn intricate genotype-phenotype correlations often missed by traditional methods. Validated on cohorts from The Cancer Genome Atlas (TCGA), our framework demonstrates a significant improvement in diagnostic performance over single-modality baselines. We also incorporate Explainable AI (XAI) techniques to ensure model transparency, a crucial step for clinical adoption. The framework serves as both a powerful diagnostic tool and a hypothesis- generation engine, uncovering novel biomarkers from complex multi-modal data and advancing computational pathology and personalized medicine.
Keywords
Multi-Modal Learning, Transformers, Histopathology, Genomics, Proteomics, Explainable AI, Early Cancer Diagnosis
How to cite this paper
@article{1711765,
author = {Ayush Mishra, Anadi Mishra, Adarsh Tiwari, Uttam Sharma, Nikhil Raj},
title = {A Unified Multi-Modal Transformer Framework for Synergistic Cancer Diagnosis},
journal = {Iconic Research And Engineering Journals},
year = {2025},
volume = {9},
number = {5},
pages = {197-205},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1711765.pdf},
abstract = {Early cancer diagnosis is critical for improving patient outcomes but is challenged by the disease?s profound heterogeneity. This paper introduces a unified, AI-powered framework that synergistically integrates histopathology, genomics, and proteomics data to enhance early cancer detection. Our architecture features a novel multi-transformer model with dedicated Vision and Genomic Transformers to encode modality-specific features, which are then fused by a cross-modal attention transformer. This intermediate fusion strategy enables the model to learn intricate genotype-phenotype correlations often missed by traditional methods. Validated on cohorts from The Cancer Genome Atlas (TCGA), our framework demonstrates a significant improvement in diagnostic performance over single-modality baselines. We also incorporate Explainable AI (XAI) techniques to ensure model transparency, a crucial step for clinical adoption. The framework serves as both a powerful diagnostic tool and a hypothesis- generation engine, uncovering novel biomarkers from complex multi-modal data and advancing computational pathology and personalized medicine.},
keywords = {Multi-Modal Learning, Transformers, Histopathology, Genomics, Proteomics, Explainable AI, Early Cancer Diagnosis},
month = {November},
}