Evaluating Emerging AI and E-Nose Diagnostic Tools in Clinical Practice

The cluster is relevant to clinical nutrition primarily as a diagnostic-literacy issue: none of the reported developments establishes a nutrition intervention, dietary biomarker, or practice change. The signal is technological; its clinical applicability remains conditional.

An e-nose result is not a nutrition test

One article examines a pilot study of a quantum-dot-based electronic nose designed to detect volatile organic compounds emitted through the skin. The device used cadmium sulfide nanocrystals, and researchers reportedly identified potential biomarker patterns associated with the presence or absence of a malignant tumour.

The reported performance figure—100% accuracy and sensitivity in distinguishing cancer patients from healthy controls—requires unusually careful interpretation. This was a pilot study, not evidence that an e-nose is ready for routine screening, diagnosis, dietary assessment, or metabolic monitoring. The report also notes a potential ability to classify disease severity, but does not provide the study population, validation method, specificity, or comparison with established diagnostic pathways.

For nutrition professionals and patients, the practical boundary is straightforward:

• no diet can be selected or adjusted from this report;
• no consumer “breath,” skin-odour, or metabolite device can be treated as equivalent evidence;
• a proposed volatile-compound biomarker is not, by itself, a clinically validated endpoint.

Analytical sensitivity is not synonymous with clinical utility. The pharmacokinetic and metabolic sources of volatile compounds are complex; the available report does not establish how such signals should be interpreted in relation to food intake or nutritional status.

AI may assist rare-disease research, but has not demonstrated routine clinical value

A second JMIR article reviews machine-learning approaches for rare diseases, where sparse data and limited funding complicate diagnosis and treatment research. Newswise reports that the deep generative model popEVE has identified 123 novel genetic variants that may contribute to severe developmental disorders.

That is a research signal, not a diagnostic verdict. The report itself states that, beyond a small number of success stories, larger foundation and frontier models have not yet demonstrated their value in real-world rare-disease care.

This distinction matters in nutrition-adjacent rare conditions, where patients may encounter highly specific dietary claims before the underlying diagnosis is secure. Data suggesting a potentially pathogenic variant do not establish disease mechanism, nutritional requirements, or treatment efficacy. The evidence described concerns the potential acceleration of detection and drug discovery—not an approved clinical workflow.

The more immediate issue is AI-dependent clinical reasoning

The third article addresses the expanding use of AI in medical education. JMIR’s reporting highlights personalized, immediate feedback as a possible benefit, while educators warn of “never-skilling,” “mis-skilling,” and “de-skilling” when learners rely on AI instead of developing critical reasoning.

For clinical nutrition, the concern is concrete. A plausible-sounding automated answer can still be detached from patient-specific biochemical data, diagnostic uncertainty, and the limits of available evidence. AI may scaffold reasoning; it cannot replace it.

Verdict: the reported work is scientifically interesting, but it does not support changes to diet, supplementation, screening, or clinical decision-making. Until independent validation and real-world performance data are available, the appropriate response is methodological restraint rather than technological enthusiasm.