Ethical dangers of facial phenotyping through photography in psychiatric genomics studies

Abstract

Psychiatric genomics research protocols are increasingly incorporating tools of deep phenotyping to observe and examine phenotypic abnormalities among individuals with neurodevelopmental disorders. In particular, photography and the use of two-dimensional and three-dimensional facial analysis is thought to shed further light on the phenotypic expression of the genes underlying neurodevelopmental disorders, as well as provide potential diagnostic tools for clinicians. In this paper, I argue that the research use of photography to aid facial phenotyping raises deeply fraught issues from an ethical point of view. First, the process of objectification through photographic imagery and facial analysis could potentially worsen the stigmatisation of persons with neurodevelopmental disorders. Second, the use of photography for facial phenotyping has worrying parallels with the historical misuse of photography to advance positive and negative eugenics around race, ethnicity and intellectual disability. The paper recommends ethical caution in the use of photography and facial phenotyping in psychiatric genomics studies exploring neurodevelopmental disorders, outlining certain necessary safeguards, such as a critical awareness of the history of anthropometric photography use among scientists, as well as the exploration of photographic methodologies that could potentially empower individuals with disabilities.

Democratising or disrupting diagnosis? Ethical issues raised by the use of AI tools for rare disease diagnosis

“…Like other commentators, they regard the use of AI tools (CP) as potentially widening access to health care by providing non-specialists with diagnostic capabilities ( Lai et al., 2020 ; Drogt et al., 2022 ) and, as a result, more patients with a diagnosis (of rare disease). This has been seen as important in low - and middle-income countries (LMICs) ( Kerasidou, 2021 ; Weissglass, 2022 ), particularly in places where genotyping methods are unavailable ( Kong, 2019 ). Similar findings are reported in a recent literature review, which suggests that the majority of healthcare professionals whose work involves image interpretation (e.g.…”

“…However, a minority of our interviewees noted that managing expectations about the use CP in the clinic is not so straightforward, for the idea that AI can and/or will produce definitive or infallible answers overlooks the fact that algorithms are socially constructed tools, which, although badged as more scientific and objective, are also subject to (human) biases. As has been argued elsewhere, the incorporation of fallacious assumptions about the representativeness of training datasets in CP may lead to misdiagnosis and consequently, disadvantage certain groups who may suffer from stigmatisation and discrimination and an inability to access care and support as a result ( Hallowell et al., 2019 ; Kong, 2019 ). While we may try to avoid such outcomes by designing more transparent AI systems ( AI HLEG, 2019 ; Morley, Floridi, et al., 2020 ), and ensure on-going dialogue between developers and users during the validation process ( Winter & Carusi, 2022 ), the problem of the representativeness of training datasets is more intractable, particularly the curation of training datasets of rare disease phenotypes.…”

“…While we may try to avoid such outcomes by designing more transparent AI systems ( AI HLEG, 2019 ; Morley, Floridi, et al., 2020 ), and ensure on-going dialogue between developers and users during the validation process ( Winter & Carusi, 2022 ), the problem of the representativeness of training datasets is more intractable, particularly the curation of training datasets of rare disease phenotypes. As Kong (2019) notes, if we are to leverage the benefits of CP technology for diagnosis, then we need to access data from all human populations and curate training datasets of facial images that are as ethnically diverse as possible, otherwise there is the potential that CP will just exacerbate pre-existing global health inequalities.…”

Facial imaging to screen for fetal alcohol spectrum disorder: A scoping review

“…This was an objective of the review, but the topic was not thoroughly covered in the literature included in this review. A recent report by Kong (2019) argues that using photography to aid phenotyping in psychiatric studies allows for the objectification of those affected, worsens stigmatization and draws parallels to the advancement of eugenics related to race, ethnicity and intellectual ability. Similarly, Bell et al (2016) have cautioned against the inadvertent generation of stigma through public health interventions regarding FASD.…”