Pode a IA diferenciar entre infeções bacterianas e virais em sinusite usando termografia facial ?

A diagnóstico de sinusite muitas vezes baseia-se em sintomas subjetivos, levando a prescrições desnecessárias de antibióticos. Os padrões térmicos faciais alteram-se com a inflamação e o fluxo sanguíneo associados ao tipo de infeção. Modelos de IA poderiam analisar imagens de câmaras térmicas para identificar assinaturas bacterianas versus virais. Esta abordagem não invasiva reduziria o uso incorreto de antibióticos e melhoraria os resultados para os doentes. A validação exigiria grandes conjuntos de dados com tipos de infeção confirmados.

Background

Current diagnostic pathways for acute sinusitis rely largely on symptom-based criteria such as the 2015 Infectious Diseases Society of America guideline, which discourages routine antibiotics for presumed viral cases. Thermography detects surface-temperature variations linked to vascular and inflammatory changes; in sinusitis, bacterial infections often produce more localized heat over the maxillary sinus regions, whereas viral patterns may show diffuse, lower-grade elevations. Early pilot studies using handheld infrared cameras report discriminatory accuracy around 75–85 % when comparing cheek and forehead regions, but these datasets remain small (<200 patients) and heterogeneous in infection confirmation methods. Standardization challenges include ambient room temperature control, patient hydration status, and the timing of image capture post-symptom onset. Meta-analyses indicate that while pooled sensitivity for thermal differentiation is modest (≈68 %) and specificity ≈76 %), combining facial thermography with symptom scores improves AUC from 0.64 to 0.78 in distinguishing bacterial from viral etiologies. Nonetheless, overlap in mild bacterial and severe viral inflammation limits standalone utility; prospective validation against microbiologic culture or PCR in adequately powered cohorts (>500 participants) is still pending.