Addressing Pre-Analytic Variation in Digital Pathology: U-M and Kenya Collaboration

This project tackles the pre-analytic sources of variation that undermine artificial intelligence in pathology when tools trained elsewhere are deployed in new settings. Working from bottlenecks identified by Aga Khan University pathologists in Nairobi, including inconsistent fixation and staining across referring sites, scanner and monitor color variance, slide focus artifacts and a shortage of Kenya-representative digital datasets, the team pairs Aga Khan and Michigan investigators to build quality control, stain normalization and multi-scanner validation into the workflow. A joint steering committee governs the work, and fairness audits at months six and twelve check performance across patient subgroups. The goal is to enable Aga Khan Pathology to deploy digital pathology for primary diagnosis while producing openly documented methods that other institutions can adapt.