Multi-modal digital pathology for colorectal cancer diagnosis by high-plex immunofluorescence imaging and traditional histology of the same tissue section
Precision oncology still leans on H&E histology to diagnose and stage disease; this AACR CRC 2022 poster reads high-plex Orion immunofluorescence and traditional H&E from the same colorectal cancer tissue section, and shows that automated image-feature models can predict progression-free survival in a research cohort.
Presented by Harvard Medical School / RareCyte.
Read or download to see:
- High-plex immunofluorescence and traditional H&E were read from the same colorectal cancer tissue section, cell for cell. Each specimen was imaged in one shot with the Orion method across 16 to 20 channels — 15 to 20 fluorophore-labelled antibodies together with a Hoechst nuclear stain and tissue autofluorescence — then stained with hematoxylin and eosin and rescanned in brightfield, across 40 human colorectal cancer resections totalling roughly 60 million cells.
- Automated image-feature models built from the imaging predicted progression-free survival in this research cohort. An immune-infiltration model that recapitulates the Immunoscore test (IFM1) reached a hazard ratio of 0.209 for progression-free survival, and a tumor-intrinsic model (IFM2) reached 0.0785; when the two models were combined the hazard ratio fell to approximately 0.045. Both models were ranked automatically against all 14,950 parameter combinations the pipeline evaluated.
- The immunofluorescence agreed with an established method and carried information complementary to the histology. On neighbouring sections of 16 colorectal specimens, Orion immunofluorescence tracked cyclic immunofluorescence (CyCIF) with Pearson correlations from 0.83 to 0.97 across CD8, CD163, pan-cytokeratin and Ki-67; the authors report that immunofluorescence and H&E give human experts and machine-learning models complementary information.
