The Challenge
Medical imaging plays a critical role in cancer diagnosis and treatment planning. However, interpreting these images is complex, time-consuming, and subject to variability between radiologists. Traditional image analysis often fails to capture subtle patterns and features that could provide valuable diagnostic and prognostic information.
Furthermore, the increasing volume and complexity of medical imaging data—including multiple modalities such as CT, MRI, PET, and ultrasound—create challenges for clinical workflows and limit the extraction of comprehensive insights from this rich data source.
Our Approach
Our research investigates deep learning architectures for medical image analysis. We are developing systems that can:
- Detect and segment tumors with high precision across multiple imaging modalities
- Identify radiomic features that correlate with genetic profiles and treatment outcomes
- Reconstruct and enhance medical images to improve diagnostic quality
- Track subtle changes in tumor characteristics over time to evaluate treatment response
- Generate synthetic images to augment training data and improve model performance
Our models incorporate oncology and radiology knowledge and are designed to provide interpretable results that can support clinical decision-making.
Multimodal Fusion
Integrating information from various imaging modalities (CT, MRI, PET) to provide a more comprehensive view of tumor characteristics and environment.
Radiomics & Radiogenomics
Extracting quantitative features from medical images and correlating them with genomic data to uncover new biomarkers and predictive signatures.
Image Reconstruction
Applying deep learning techniques to improve image quality, reduce artifacts, and enable lower-dose imaging protocols without sacrificing diagnostic information.
Real-time Analysis
Developing systems capable of providing instantaneous analysis and feedback during image acquisition or interventional procedures.
Current Research Projects
4D Tumor Evolution Mapping
We're developing a system that integrates temporal sequences of multi-parametric MRI to create dynamic, four-dimensional maps of tumor evolution for visualization of growth patterns, vascular changes, and treatment effects over time.
AI-Enhanced Interventional Radiology
This project explores real-time guidance for interventional procedures such as tumor ablations and biopsies. Our models aim to provide feedback during procedures to support targeting accuracy and minimize damage to surrounding tissue.
Low-Field MRI Enhancement
We're applying generative models to enhance images from low-field MRI systems, which are more accessible and affordable than high-field systems, particularly in resource-constrained settings.
Whole-Body Tumor Burden Assessment
Our team is developing algorithms for automated quantification of total tumor burden from whole-body imaging studies to support treatment planning and response monitoring.
Technical Innovations
Our research incorporates several cutting-edge technical approaches:
- Self-supervised learning: Leveraging vast amounts of unlabeled medical imaging data to pre-train models that can then be fine-tuned with limited labeled data
- Few-shot and zero-shot learning: Developing models that can generalize to new cancer types or imaging protocols with minimal or no additional training
- Uncertainty quantification: Implementing techniques that provide confidence estimates with predictions, critical for clinical decision-making
- Federated learning: Creating collaborative training frameworks that enable model improvement across institutions without sharing sensitive patient data
- Physics-informed neural networks: Incorporating physical models of imaging processes to improve reconstruction and analysis
Clinical Applications
Our medical imaging research addresses key clinical needs across the cancer care continuum:
Screening & Detection
Enhancing cancer screening programs through automated detection of suspicious findings, risk stratification, and prioritization of cases for radiologist review.
Diagnosis & Staging
Providing detailed characterization of tumors and assessment of disease extent to guide treatment planning and prognostic assessment.
Treatment Planning
Supporting precise targeting for radiation therapy, surgical planning, and optimal intervention strategies based on comprehensive tumor mapping.
Response Assessment
Quantifying changes in tumor characteristics over time to evaluate treatment efficacy and guide adaptive therapy approaches.
Collaborations and Partnerships
Our medical imaging research thrives on collaborative partnerships. We're actively seeking to work with:
Radiology Departments
For access to diverse imaging data and clinical expertise to guide our algorithm development
Medical Imaging Companies
To implement our algorithms in commercial imaging platforms and accelerate clinical adoption
Cloud Computing Providers
To develop scalable infrastructure for processing and analyzing large medical imaging datasets
Research Background
This area is in active scoping and pilot design. We share open artifacts as they mature.
Related Research Areas
Stay Updated
Subscribe to receive updates on our latest research findings and breakthroughs.