AI dental reports match human readability but need standardized evaluation before routine clinical implementation
This scoping review analyzed 1,265 records to identify seven studies meeting inclusion criteria regarding AI applications for automated dental report generation. The intervention involved NLP-based models, including GPT variants and fine-tuned large language models, compared against human-authored reports. The review scope covered accuracy for common findings, readability, patient-rated clarity, hallucination, structural validity, response latency, output length, readability indices, clinician ratings, and patient questionnaires.
Key synthesized findings show high accuracy for common findings and readability comparable to human-authored reports. Patient-rated clarity improved with simplified AI-adapted versions. Specific effect sizes, absolute numbers, p-values, or confidence intervals were not reported for these outcomes. Safety data such as adverse events, serious adverse events, discontinuations, or tolerability were not reported.
The authors note significant limitations including heterogeneity in report types, datasets, languages, and evaluation metrics. These factors contribute to uncertainty in the evidence. The review does not describe a specific study population or setting as these details were not reported. Funding or conflicts of interest were not reported.
Practice relevance suggests that standardized evaluation frameworks, larger multilingual datasets, and assessment of patient comprehension are needed before routine clinical implementation. The review explicitly advises against overstating the readiness for routine clinical implementation given the current evidence gaps.
View Original Abstract ↓
Intraosseous route for out-of-hospital cardiac arrest linked to lower outcomes than intravenous access
Using a bone needle instead of a vein line lowers survival chances for cardiac arrest patients
Related in Emergency Medicine
From Other Specialties
