A multimodal dataset for precision oncology in head and neck cancer

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Head and neck cancer presents a significant clinical challenge with poor patient prognoses.
• Personalized treatment strategies hold promise for improving outcomes but are hindered by limited public data.
• Existing research lacks comprehensive, multimodal datasets for head and neck cancer.

Purpose of the Study:

• To introduce the HANCOCK dataset, a novel multimodal resource for head and neck cancer research.
• To facilitate the exploration of head and neck cancer pathology and treatment response using integrated data.
• To advance multimodal machine learning methodologies in the field of precision oncology.

Main Methods:

• Development of the HANCOCK dataset, a monocentric collection of real-world data from 763 head and neck cancer patients.
• Inclusion of diverse data modalities: demographical, pathological, blood data, surgery reports, and histologic images.
• Application of machine learning techniques, including foundation models for imaging data, for endpoint prediction.

Main Results:

• Demonstration that combining multiple data modalities via machine learning outperforms single-modality approaches for endpoint prediction.
• Validation of the utility of integrating imaging data with foundation models to enhance predictive accuracy.
• Establishment of a low-dimensional representation for exploring the multimodal HANCOCK dataset.

Conclusions:

• The HANCOCK dataset provides a valuable resource for advancing head and neck cancer research.
• Multimodal data integration and machine learning significantly improve prediction of clinical endpoints in head and neck cancer.
• HANCOCK is poised to drive innovation in precision oncology and multimodal learning research.