The Challenge
Cancer diagnosis and treatment planning often require invasive biopsies to obtain tissue for genomic analysis. These procedures carry risks, may not fully capture tumor heterogeneity, and cannot be repeated frequently to monitor treatment response. Meanwhile, medical imaging is routinely performed but typically analyzed in a qualitative manner that doesn't fully leverage the wealth of information contained in the images.
The emerging fields of radiomics (extracting quantitative features from medical images) and radiogenomics (correlating these features with genomic data) offer the potential to develop non-invasive biomarkers. However, the relationship between imaging phenotypes and genomic characteristics is complex and not easily discernible by human perception alone.
Our Approach
Our research aims to leverage advanced AI techniques to extract and analyze quantitative imaging features and correlate them with genomic characteristics. Through deep learning and other machine learning approaches, we seek to:
- Develop algorithms that can extract thousands of quantitative features from medical images
- Identify imaging biomarkers that correlate with specific genetic mutations and molecular subtypes
- Create non-invasive tools for predicting treatment response and monitoring disease progression
- Build integrated models that combine radiomics with other clinical and molecular data for improved decision-making
Deep Radiomics
Using deep learning to automatically extract relevant imaging features beyond traditional handcrafted features, potentially revealing previously unrecognized imaging biomarkers.
Genome-Image Correlation
Developing models that establish relationships between imaging phenotypes and genomic alterations, potentially allowing for non-invasive genetic profiling of tumors.
Temporal Radiomics
Creating algorithms that track changes in imaging features over time to monitor treatment response and detect early signs of resistance or recurrence.
Multi-modality Integration
Building systems that integrate features from multiple imaging modalities (CT, MRI, PET) with genomic and clinical data for comprehensive tumor characterization.
Potential Research Projects
Radiomic Signatures for Molecular Subtypes
Developing AI algorithms that can analyze standard-of-care imaging (CT, MRI, PET) to predict molecular subtypes of cancer that currently require tissue biopsy. For example, we could create models to identify specific genomic alterations like EGFR mutations in lung cancer or HER2 status in breast cancer from imaging features alone.
Spatiotemporal Tumor Heterogeneity Mapping
Creating AI systems that map tumor heterogeneity across different regions and over time using radiomics. This could potentially identify regions of a tumor with different genetic profiles, guiding more precise biopsies and treatments targeted to the most aggressive areas.
Radiogenomic Treatment Response Prediction
Combining baseline radiomics features with genomic biomarkers to develop models that predict response to specific therapies. Such models could guide treatment selection by identifying which patients are most likely to benefit from particular treatments based on their integrated radiomic-genomic profile.
Future Directions
As our radiomics and radiogenomics research progresses, we anticipate exploring several promising future directions:
- Development of AI models that can generate synthetic biopsies from imaging data, potentially eliminating the need for invasive procedures
- Creation of radiogenomic digital twins that simulate how tumors with specific genomic profiles would appear on different imaging modalities
- Integration of radiomics with liquid biopsy data to create comprehensive non-invasive tumor monitoring systems
- Application of federated learning to train radiomics models across institutions while preserving patient privacy
Collaborations and Partnerships
We would seek partnerships with:
Medical Imaging Centers
To access large, annotated imaging datasets for model development and validation
Genomics Research Institutes
To correlate imaging features with comprehensive genomic profiling
Imaging Equipment Manufacturers
To integrate radiomic analysis tools directly into imaging workflows
Research Background
This research area contributes to the growing body of knowledge in AI-powered cancer research. We're currently developing foundational work in this space.
Related Research Areas
Stay Updated
Subscribe to receive updates on our latest research findings and breakthroughs.