Surgical outcomes depend on operative precision, making instrument tracking essential for assessing surgeon performance. Sensor and marker-based approaches raise sterilization and ergonomic concerns, whereas vision-based markerless approaches enable tracking directly from surgical videos. However, markerless instrument tracking has predominantly focused on minimally invasive surgery, where video capture is inherently available, unlike in open surgery. Comprehensive reviews on markerless instrument tracking in open surgery, particularly motion tracking and pose estimation beyond conventional 2D analyses, are lacking. To address this gap, we present a scoping review following PRISMA-ScR guidelines, searching five databases (PubMed, Web of Science, IEEE Xplore, Scopus, and ACM Digital Library) for studies reporting markerless tool motion tracking or pose estimation during open surgery. Of 2084 articles screened, 14 were included in the review. Deep learning (DL) was used in 12 studies, while classical computer vision (CV) and hybrid DL-CV approaches were used in one study each. Outputs included pose estimation (n=8) and motion tracking (n=6). Orthopedic surgery using cadaver models was commonly reported (n=4), while skill assessment and infrastructure development/benchmarking were the most frequent applications (n=6 each). Surgical drills were the most tracked instruments (n=6), with majority of the articles utilizing simulated datasets (n=10). Most algorithms, among the 14 articles, relied on frame-wise estimation, suggesting that temporal dynamics may be underutilized. Overall, markerless tool tracking in open surgery remains at an early stage, with studies focusing largely on low-fidelity phantoms and cadaveric models. Further research is needed to improve robustness and applicability for real surgeries.
Data-driven and machine-learning polymer science increasingly depends on reusing openly deposited datasets, yet the metadata quality of polymer deposits in generalist repositories has never been measured at population scale. We harvest 53 038 polymer-related records (2020–2025) across eight generalist repositories and score each on an 11-criterion binary FAIR-aligned reuse score.Repository identity dominates score variance (Kruskal-Wallis η² = 0.379, large), with mean scores from 7.0 (Dryad) to 1.5 (OSF) — an order of magnitude larger than year, sustainability framing, or sub-domain effects. No single criterion dominates: under a corrected item-total correlation (each criterion vs. the sum of the remaining ten) ρ ≤ 0.41 throughout, and method-description (ρ = 0.716 at 55.7 % fulfilment) is a part-whole artefact of its mid-range prevalence. Annual deposit volume rose +143 % while mean score rose only +7.7 %: volume has decoupled from quality. Sustainable / bio-based datasets are modestly better documented (Cliff’s δ = −0.314, small) but only 6.8 % of the corpus. Non-isocyanate-polyurethane (NIPU) records appear under-documented (δ = −0.448) but collapse to non-significant (δ = +0.042, p = 0.21) once the two low-tier repositories are excluded — they are platform-mediated, not behavioural. A structural OSF file-level FAIR gap (53 % of OSF records) is closed by a parent-inheritance workaround that lifts coverage from 0 % to 96.8 %. The full pipeline is byte-reproducible from the deposited raw harvest.
