AI in Point-of-Care Imaging for Clinical Decision Support: Systematic Review of Diagnostic Accuracy, Task-Shifting,...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Clinical Decision Support

Background:

• Point-of-care imaging integrated with AI offers potential for expanded healthcare access in resource-limited settings.
• Existing systematic reviews lack comprehensive evaluation of AI-assisted decision support across diverse imaging modalities, explainability, and clinical impact.

Purpose of the Study:

• To systematically evaluate and synthesize evidence on AI-based clinical decision support systems utilizing point-of-care imaging.
• To assess the implementation of explainability and quantify evidence gaps in clinical impact.

Main Methods:

• Systematic search of major databases (PubMed, Scopus, IEEE Xplore, Web of Science) from January 2018 to November 2025.
• Inclusion of studies on AI/machine learning for point-of-care imaging with clinical decision support outputs.
• Data extraction across 15 domains and quality assessment using QUADAS-2, with narrative synthesis due to heterogeneity.

Main Results:

• Twenty studies (78,000 patients) evaluated AI in TB, breast cancer, DVT, and other conditions using ultrasound, X-ray, photography, fundus, microscopy, and dermoscopy.
• Median sensitivity and specificity were high (93.6% and 90.6%, respectively). Task-shifting was successful in 65% of studies, with non-specialists achieving expert performance after minimal training.
• Significant gaps were found in explainable AI (XAI) implementation (75% studies omitted it) and clinical impact measurement (none reported patient outcomes). Methodological quality was poor, with 70% of studies at high risk of bias.

Conclusions:

• AI-assisted point-of-care imaging demonstrates diagnostic promise and facilitates task-shifting.
• Critical evidence gaps include lack of patient outcome data, inadequate XAI evaluation, and need for multicontext validation.
• Further research focusing on patient outcomes, rigorous XAI assessment, and cross-context validation is essential for safe adoption.