HER2 in Gastric / GEJ Carcinoma

HER2 testing performance, heterogeneity, and reproducibility in gastric and gastroesophageal cancers

← Home · Onboarding

What this project is

HER2 scoring in gastric and gastroesophageal carcinoma is known to be harder than in breast — there is more intratumoral heterogeneity, more variation in fixation, and the scoring rubric is different. This project characterizes HER2 interpretation at Memorial across these dimensions.

Objectives

• Quantify inter-observer agreement on HER2 IHC scoring among department pathologists.
• Map the intratumoral heterogeneity of HER2 expression using WSI-level annotations.
• Describe the correlation (and discordance) between IHC and ISH.
• Identify the subset of cases where biopsy-vs-resection scoring diverges.

Status

Stream	Status
Cohort identification	Complete
IHC / ISH database	Curated
Agreement study	Pilot scoring round complete
Heterogeneity annotation (QuPath)	Active
Manuscript	Outline drafted

Data

• Gastric / GEJ resections and biopsies with HER2 IHC and reflex ISH.
• De-identified scanned slides on the Memorial share.
• Annotations in QuPath, exported as GeoJSON per the group SOP.

Tools and repositories

• HER2-intratumoral-heterogeneity — primary repo.
• Annotation SOP: see Home → Protocols and SOPs.
• Agreement statistics: meddecide (kappa, weighted kappa, ICC).

How to contribute

1. Attend an annotation calibration session.
2. Score the pilot batch and compare to the reference scorer.
3. Contribute GeoJSON annotations following the naming convention in the repo README.

Pitfalls

• HER2 scoring differs between biopsy (focal positivity accepted) and resection rules — never aggregate them without the specimen-type column.
• Slide-level tumor percentage matters for downstream heterogeneity metrics. Don’t skip the tumor-area annotation even when the HER2 score alone is the question.

In the research wiki

• HER2 Gastric Cohort — the curated cohort: cases, IHC, reflex ISH, slides and annotations in one place.
• Intratumoral Heterogeneity — the scientific reason the WSI-level annotation exists at all — the answer depends on which piece was sampled.
• Interobserver Agreement — the method behind the agreement study, and the ceiling it sets on any reference standard for scoring.
• QuPath Annotation Workflow — the GeoJSON conventions the annotation SOP describes.
• Diagnostic Test Accuracy — IHC read against ISH is an index test against a reference standard, with the prevalence caveats that implies.
• Cross-Stain Registration — what it would take to move from case-level discordance to asking whether the same regions are positive on IHC and amplified on ISH — including the serial-section ceiling that bounds any such claim.
• VALIS — the tool for that question if it is ever asked: it registers a whole series of stains rather than a pair, and can carry an existing annotation from one stain onto another instead of redrawing it.
• HER2 Intratumoral Heterogeneity — this project’s own pointer page in the wiki, with its objectives and open questions kept beside the concepts they depend on.