Automatic contrast phase classification of polyphasic CT scans

Co-authored work on identifying contrast phase from CT volumes — poster at AACR 2026.

AACR Annual Meeting 2026 · Poster

Background

Contrast-enhanced CT is the standard imaging modality for many oncology applications — but downstream pipelines (tumor segmentation, response assessment, radiomics) generally assume a known contrast phase. In real-world multi-center data, contrast phase metadata is often missing, inconsistent, or incorrect at the DICOM tag level.

This work develops a deep learning classifier that infers contrast phase directly from CT volumes, allowing downstream segmentation and analysis pipelines to either filter to a target phase or be conditioned on the inferred phase.

Contribution

Co-authored with G. Hughes, M. Patwari, Y. Wei, M. Parker, J. Parkin, and A. Filippov. Data engineering, preprocessing, and modeling contributions through the broader medical imaging ML stack at AstraZeneca.

Reference

Hughes G, Patwari M, Wei Y, Parker M, Parkin J, Zhang Z, Filippov A. “Automatic contrast phase classification of polyphasic CT scans.” Poster, AACR Annual Meeting 2026. Abstract: Cancer Research 86, Suppl. 7, 2777.

Read the abstract