Pink Kidney — Renal AI Study

AI-assisted evaluation of renal parenchyma and tumor on H&E whole-slide images

GitHub

PinkKidney — the source-level pointer page: cohort size, scanners, controls, and what has actually been run.
Foundation Models in Pathology — the encoders already wired into one extraction script here, and what choosing between them turns on.
Encoder Benchmarking — the four protocol choices that decide whether a comparison between those encoders means anything.
External Validation — the single thing standing between this cohort and a publishable claim: the internal controls are strong and there is still no outside cohort.
AIDA Data Hub — the cheapest concrete route to one — free annotated Swedish cohorts, gated behind a signature rather than a collaboration, though nothing renal yet.
Scanner and Stain Variability — three scanners in one cohort, which is the batch structure the ComBat correction here exists to handle.
Feature-Space Distance — how to put a number on that batch structure instead of inspecting a t-SNE, using encoders already cached on this cohort.
Patch Extraction — the tiling decisions that set the ceiling on everything downstream.
PembeBobrek — the Turkish documentation and deploy chain for the same project — read this one before touching either repository, because three repositories hold one project.