Topics of Interest
We invite high-quality submissions that advance the field of cancer detection and prevention through the use of advanced AI. The scope of this workshop is centered on multi-modal data and foundation models, including but not limited to the following areas:
Multi-Modal Data Integration in Cancer Care:
Combining radiology (e.g., CT/MRI scans), pathology (e.g., histopathology slides), genomics (e.g., somatic mutations), spatial omics (e.g., transcriptomics, proteomics), and clinical data for holistic cancer analysis.
Techniques such as transformers for multi-modal fusion or graph neural networks (GNNs) for relational data modeling.
Case studies on multi-modal AI applications in cancer detection and prognosis.
Data processing techniques like data pipelines, cleaning, sampling/subset selection.
Multi-Modal Foundation Models for Cancer Analysis
Architectures and training strategies for foundation models using diverse data modalities: radiology (CT, MRI), pathology (WSI), genomics, proteomics, and electronic health records (EHRs).
Multi-modal learning techniques for fusing complementary information to improve diagnostic performance.
Image-to-text models for automatically generating findings for radiology or pathology reports.
Methods for longitudinal studies, tracking lesion changes and patient status over time using multi-modal history.
Efficient voxel representations and 3D architectures for volumetric data (CT, MRI) within foundation models.
Foundation Models in Clinical Pathology:
Pre-trained models designed for large-scale medical imaging tasks (e.g., tumor segmentation or classification).
Applications in rare cancer detection, biomarker discovery, and generalization across diverse patient populations.
Challenges in adapting foundation models to domain-specific oncology tasks.
AI-Driven Early Detection and Risk Prediction:
Leveraging AI to identify early-stage cancers through imaging biomarkers and circulating tumor DNA (ctDNA).
Risk stratification using genetic predispositions (e.g., BRCA mutations) and lifestyle factors.
Public health implications of predictive modeling for large-scale screening programs.
Personalized Oncology:
Predictive modeling for therapeutic response based on tumor microenvironment characteristics by developing digital biomarkers for targeted and immunotherapy.
Integration of imaging-derived phenotypes with genomic alterations to guide precision oncology.
Data-Efficient Learning and Adaptation in Oncology
Un/semi/weakly-supervised learning for pre-training and fine-tuning on medical data where annotations are scarce or incomplete.
Domain adaptation techniques to address performance degradation across different hospitals, imaging devices, or patient demographics (e.g., race, ethnicity).
Active learning strategies to intelligently select the most informative data for annotation, reducing cost and maximizing model improvement for cancer-specific tasks.
Methods for label refinement and handling noisy labels, especially for diseases with subtle visual cues or high inter-observer variability among physicians.
Robust and Interpretable AI for Clinical Translation
Out-of-distribution (OOD) detection to ensure model safety by identifying and rejecting inputs that the model cannot reliably analyze.
Development of explainable AI (XAI) systems tailored for oncologists and pathologists to build trust and aid in decision-making.
Core computer vision tasks (classification, detection, segmentation) specifically benchmarked for robustness and accuracy in clinical cancer workflows.
Image enhancement techniques (e.g., noise reduction in low-dose CT) to improve the quality and reliability of input data for diagnostic models.
AI-Driven Applications in Personalized Cancer Care
Predictive modeling for therapeutic response based on tumor microenvironment characteristics.
Development of novel digital biomarkers from imaging and other data for guiding targeted and immunotherapy.
AI-powered risk stratification using genetic predispositions, imaging biomarkers, and lifestyle factors for large-scale screening programs.
Future Directions in AI-Driven Cancer Care:
Real-time decision-making during interventions using intra-operative imaging, pathology analysis, and any existing source of patient history.
Development of explainable AI systems tailored for oncologists.
Scaling multi-modal datasets through international collaborations to improve model robustness.