research
Compare like with like
A collaborative narrative review of transformer-based surgical AI.
- Role
- Co-author
- Status
- Published narrative review
Context and real problem
Transformer-based models appear across surgical imaging, workflow recognition, and other tasks. Claims of advantage are difficult to interpret when studies use different datasets, metrics, and comparison conditions.
Andres’s role
I am a co-author of “Transformers in surgical artificial intelligence: A domain-stratified, study-level narrative review.” My contribution sits within a collaborative evidence-synthesis and scientific-writing effort.
Collaborators and setting
The paper is a shared research output with clinical and technical co-authors. The public records connect it to Tufts. Co-authorship is established; this summary does not assign all searching, analysis, or writing to one person.
Methods and tools
The review organizes study-reported evidence by surgical domain. Its methods distinguish comparisons made within the same study and task from broader claims across different settings. It is a narrative review, not a claim of a completed clinical trial or a substitute for prospective validation.
Responsibility and safety
External validation, appropriate comparators, latency, safety, and fairness shape the path from a technical finding toward clinical use. A result about model performance does not, by itself, demonstrate patient benefit.
Outputs and outcomes
The review is publicly recorded in JTCVS Open and linked through AATS. This website uses the verified title, co-authorship, journal, and publication year. Exact online and issue dates remain distinguished from the date of the AATS resource page.
What I learned
The conditions of a comparison are part of its conclusion. I want a synthesis to make the strongest supported claim clear while keeping the work needed for translation visible.
Current status and links
The article is published. The DOI and AATS record are linked below. This account avoids turning pooled impressions across heterogeneous tasks into a universal claim that one model family is better.