Research Focus

AI for Multimodal Data Integration

We fuse imaging, genomics, and clinical data to build full‑stack patient profiles that drive precise decisions.

The Challenge

Cancer is an extraordinarily complex disease that manifests itself across multiple biological scales and data modalities. Clinicians and researchers now have access to an unprecedented wealth of patient data—from molecular profiles and medical images to electronic health records and wearable sensor data.

However, these different data types typically exist in silos, analyzed separately with modality-specific tools and frameworks. This compartmentalized approach fails to capture the complex interrelationships between different aspects of cancer biology and patient health, limiting our ability to develop truly personalized treatment strategies.

Our Approach

Our research develops AI frameworks that integrate diverse data modalities into unified analytical models. Our approaches include:

  • Creating deep learning architectures specifically designed to process and align heterogeneous data types
  • Developing methods to handle differences in scale, resolution, and dimensionality across data modalities
  • Building interpretable models that reveal meaningful cross-modal relationships and patterns
  • Implementing transfer learning techniques that leverage knowledge across different data domains
  • Designing federated learning systems that enable multimodal analysis while preserving data privacy and security

Imaging-Genomics Integration

Combining radiomics features from medical images with genomic data to identify novel biomarkers and predictive signatures that neither modality alone could reveal.

Clinical-Molecular Fusion

Integrating structured clinical data with molecular profiles to create comprehensive patient models that inform treatment decisions and risk assessment.

Temporal Data Synthesis

Developing methods to align and analyze longitudinal data across modalities, capturing how different aspects of cancer evolve over time and in response to treatment.

Digital Biomarker Integration

Combining data from wearables, mobile health applications, and other digital sources with traditional clinical and molecular data to monitor patient status continuously.

Current Research Projects

Integrated Cancer Phenotyping Platform

We're developing a platform that integrates radiomics features from imaging with genomic, transcriptomic, and proteomic data to create detailed cancer phenotypes, with the aim of identifying novel cancer subtypes.

Multimodal Treatment Response Prediction

This project focuses on models that integrate baseline imaging, molecular profiles, and early on‑treatment data from multiple modalities to forecast patient responses to specific therapies.

Spatial Multi-omics Analysis

We're developing methods to integrate spatially resolved multi‑omics data, including spatial transcriptomics, proteomics, and metabolomics, to map cellular ecosystems within tumors.

Patient Digital Twin Framework

Our team is building a "digital twin" framework that creates computational models of individual patients by integrating clinical, molecular, imaging, and monitoring data to simulate responses to treatment options.

Technical Innovations

Our multimodal integration research involves several technical innovations:

  • Cross-modal attention mechanisms: Neural network architectures that can selectively focus on relevant information across different data types
  • Multimodal representation learning: Techniques to create unified data representations that preserve the unique characteristics of each modality while enabling joint analysis
  • Missing data handling: Advanced methods to address the common challenge of incomplete data across modalities in real-world clinical settings
  • Uncertainty quantification: Frameworks to estimate and communicate the confidence of predictions based on multimodal inputs
  • Explainable AI approaches: Tools that provide interpretable insights into how different data types contribute to model predictions

Clinical Applications

Our multimodal integration research aims to transform several aspects of cancer care:

Precision Diagnostics

Enhancing cancer diagnosis through comprehensive integration of pathology, imaging, and molecular data to identify specific disease subtypes.

Treatment Selection

Guiding therapy decisions by analyzing how multiple patient characteristics collectively influence treatment outcomes.

Disease Monitoring

Tracking disease progression and treatment response through integrated analysis of longitudinal data from multiple sources.

Survivorship Care

Improving follow-up care by monitoring diverse indicators of recurrence risk and quality of life after primary treatment.

Future Directions

As our research progresses, we plan to explore several exciting directions:

  • Expanding our frameworks to incorporate emerging data types, such as single-cell multi-omics and spatial proteomics
  • Developing multimodal AI systems that can continuously learn and adapt as new patient data becomes available over time
  • Creating interactive visualization tools that allow clinicians to explore integrated multimodal data intuitively
  • Extending our integration approaches to include social determinants of health and environmental exposure data
  • Building multimodal knowledge graphs that capture complex relationships between different aspects of cancer biology and patient health

Collaborations and Partnerships

Our multimodal integration research thrives on collaborative partnerships. We're actively seeking to work with:

Cancer Centers

For access to diverse patient data across multiple modalities to develop and validate our integration approaches

Technology Companies

To leverage advanced computing infrastructure and data management platforms for handling complex multimodal datasets

Healthcare Systems

To implement and evaluate our multimodal approaches in real-world clinical settings

Research Background

This area is in active scoping and pilot design. We share open artifacts as they mature.

Stay Updated

Subscribe to receive updates on our latest research findings and breakthroughs.

Explore More Research

Discover our other research initiatives and ongoing projects.