Biomarker quantification
We build software that measures stained cells and signal intensity on whole slide images and produces a score for pathologist review.
A breast cancer scored HER2 positive is a candidate for HER2 targeted therapy, and the 0 to 3+ score decides eligibility.
A newer target that therapies are now directed at.
Marks cytotoxic T cells. CD8 density in and around a tumor indicates immune activity and informs immunotherapy decisions.
Hematoxylin and eosin, the standard general stain. It shows tissue structure and is the baseline for most morphological assessment.
Each marker has its own scoring convention, based on the proportion of positive cells, on staining intensity, or on both. Interrater agreement matters here, because the score is only useful if different pathologists apply the same convention.
Scoring disagreement
Pathologists do not always score the same slide the same way. Fatigue, slide quality and staining variation add to the disagreement, and it is more noticeable in borderline HER2 cases, where the score sits near a treatment threshold.
In those cases a small difference can put the score on the wrong side of the threshold and lead to the wrong therapy.
Measurement steps
The software segments tissue from background and separates tumor regions from stroma and other tissue.
Individual cells are detected and their nuclei segmented.
For each cell, staining is measured: intensity for a membrane or nuclear marker, presence for a cell type marker.
The cell results are rolled up into the score the clinical convention calls for, such as proportion positive, intensity grade or density per area.
Every step can introduce an error. Staining can vary between labs and scanners and shift intensity measurements, so the pipeline has to correct for that variation or it gets read as a biological difference. Overlapping cells make segmentation harder. Artifacts, tissue folds and out of focus regions have to be found and excluded. Before relying on the results, we compare the pipeline with expert scores on held out slides from more than one source.
The role of the pathologist
The pathologist still makes the call. The software gives them a finer, repeatable measurement that they can review, question, override, and sign off on.
Routine scoring should come out consistent, and anything that needs judgment stays with the specialist.
SequoiaAT's digital pathology platform uses computational models to analyze HER2, Trop2, CD8 and H&E staining. Its outputs support high accuracy scoring and help detect and characterize cancer subtypes.
The pipeline automates slide scoring and the validation checks around it. Scoring in routine assessment became more consistent and quicker to turn around, and manual errors went down.
Read the case study →