International Journal of Science, Engineering and TechnologyAuthors: Mansi, Assistant Professor Dr. Grace M. BasanShrieh
Abstract: Artificial intelligence (AI) tools for detecting depression and anxiety process some of the most sensitive information a person can share: their thoughts, voice, movements and health history. Technical accuracy is therefore necessary but not sufficient for responsible use. This paper, derived from a doctoral study on AI-assisted diagnosis of depression and anxiety, proposes a lifecycle governance framework that translates four ethical principles (autonomy, beneficence and non-maleficence, justice, and explicability) into concrete safeguards at five stages: data collection, model development, validation, deployment and monitoring. Eight principal risks are identified and prioritised on a likelihood–impact matrix: privacy breach, consent failure, subgroup bias, automation bias, dataset shift, opaque output, miscalibration, and stigma or misuse. An analytic fairness demonstration shows how a sensitivity drop from 0.85 to 0.70 in a minority-language subgroup doubles missed cases from 22.5 to 45.0 per 1,000 people screened, even though the overall model appears acceptable. A privacy-preserving data pipeline and a checklist of minimum safeguards are proposed. The framework positions AI as accountable, clinician-supervised decision support and gives researchers, ethics committees and clinicians a practical basis for reviewing mental health AI projects.
