The Unobserved Anatomy: Negotiating the Plausibility of AI-Based Reconstructions of Missing Brain Structures in Clinical MRI Scans

Abstract

Abstracts (EN, DE) Vast archives of fragmentary structural brain scans that are routinely acquired in medical clinics for diagnostic purposes have so far been considered to be unusable for neuroscientific research.Yet, recent studies have proposed that by deploying machine learning algorithms to fill in the missing anatomy, clinical scans could, in future, be used by researchers to gain new insights into various brain disorders.This chapter focuses on a study published in 2019, whose authors developed a novel unsupervised machine learning algorithm for synthesising missing anatomy in extremely sparse clinical MRI scans of thousands of stroke patients.By approaching the study from a media-theoretical perspective, I analyse how its authors discursively negotiated the anatomical and operative plausibility of the unobserved anatomy that their black-boxed algorithm reconstructed from the existing sparse data.My analysis foregrounds the processual, relational, context-dependent, and essentially unstable character of the thus established plausibility of the algorithmically synthesised neuroanatomy.Immense Archive fragmentarischer struktureller Gehirnscans, die routinemäßig in

Ähnliche Arbeiten