This repository contains the pair-repeat allocation study for Paired-Acquisition Neural Factorization.
The method uses paired acquisitions of the same underlying tissue to factor frozen pathology embeddings into a scanner-suppressed tissue factor and an acquisition-specific factor.
The allocation question is whether, under a matched pair-presentation budget, using more unique biological pairs improves factor separation more than repeatedly presenting fewer pairs.
Evidence tables:
data/matched_budget_allocation_means.csvdata/matched_budget_global_allocation_contrasts.csvdata/matched_budget_doubling_effects.csv
Main research hub:
https://github.com/matthewvaishnav/computational-pathology-research