Abstract

Functional neurological disorder (FND), historically referred to as hysteria, is a contested illness characterised by heterogeneous and often co-occurring neurological symptoms, such as seizures, abnormal movements, and paralysis. Its diagnosis remains challenging due to the disorder’s complexity and lack of standardised procedures. Recent neuroimaging research has sought to link FND symptoms to brain dysfunction, with three pioneering studies using resting-state functional magnetic resonance imaging (fMRI) to train machine learning (ML) classifiers for diagnostic purposes. These studies have aimed to identify biomarkers for future clinical application, while also providing new insights into FND’s neuropathophysiology that have eluded traditional data analyses. Combining perspectives from science and technology studies and media studies, this paper critically examines how researchers navigate the technical constraints of resting-state fMRI when constructing the datasets to train ML classifiers for diagnosing FND. Through a close reading of the published studies, it analyses operations performed during patient selection, resting-state data acquisition, preprocessing, and connectivity modelling to translate FND’s clinical complexity into ML-ready datasets. Drawing on Annemarie Mol’s concept of ontological politics, it argues that these operations enact specific versions of FND as a condition diagnosable through individual patients’ brain connectivity patterns, while excluding multiple key aspects of the disorder as algorithmically illegible. Whereas the studies present ML classifiers as tools to support future clinical decision-making, they overlook the political implications of inadvertent diagnostic exclusions these systems enact. The paper concludes that the development of neuroimaging diagnostic ML classifiers for FND and related conditions demands critical scrutiny of the new diagnostic realities they enact and their implications for patients.