Robust and Annotation-Free Wound Segmentation on Noisy Real-World Pressure Ulcer Images: Towards Automated DESIGN-R\textsuperscript{\textregistered} Assessment

Yun-Cheng Tsai

REVIEW 3 major objections 3 minor 19 references

Robust and Annotation-Free Wound Segmentation on Noisy Real-World Pressure Ulcer Images: Towards Automated DESIGN-R\textsuperscript{\textregistered} Assessment

T0 review · 3 major / 3 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A detection-crop step in front of a fixed pretrained segmenter is enough to make wound masks and DESIGN-R size estimates accurate on non-foot body sites.

desk verdict A plausible detector-plus-frozen-segmenter idea, but the headline IoU and DESIGN-R numbers rest on ground-truth masks the paper itself says do not exist. read the letter →

arxiv 2505.23392 v1 pith:2ME7C4SQ submitted 2025-05-29 cs.CV

Yun-Cheng Tsai This is my paper

classification cs.CV

keywords pressureulcersegmentationwoundDESIGN-RassessmentYOLOv11nFUSegNetbounding-boxsupervisiondomaingeneralizationclinicaldeployment

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that a foot-trained segmentation model can be made to work on sacral, trochanter, and limb pressure ulcers without pixel-level retraining, as long as a lightweight detector first crops the wound. The proposed pipeline—YOLOv11n for region-of-interest detection followed by pretrained FUSegNet for segmentation—is reported to lift mean IoU by 23 percentage points over vanilla FUSegNet and to raise end-to-end DESIGN-R size estimation accuracy from 71% to 94% on real-world clinical photos. If true, this means the main barrier to cross-site wound segmentation is background noise and input variability, not the segmenter's learned wound knowledge. That would make annotation-light deployment of wound scoring tools practical in nursing documentation workflows.

What carries the argument

The load-bearing mechanism is the detector–segmenter cascade: a lightweight YOLOv11n object detector trained on 500 bounding boxes localizes the wound, crops it, and resizes the crop to 512×512 pixels before passing it to the fixed pretrained FUSegNet. The crop removes background clutter and normalizes scale and framing, which the paper argues is what lets a foot-ulcer segmenter generalize to other body sites without fine-tuning.

What would settle it

Take an independent set of sacral and trochanter wound photographs that do have expert pixel-level masks, run the released YOLOv11n+FUSegNet pipeline and vanilla FUSegNet on them, and compare mean IoU; if the non-foot IoU values do not land near the reported 0.78–0.85, the 23-point improvement and the 94% DESIGN-R size accuracy are not reproducible.

Watch

Extended reading notes

Core claim

The central claim is that feeding FUSegNet tight, normalized wound crops instead of full noisy frames lets the same pretrained weights segment pressure ulcers on the sacrum, trochanter, and foot, with reported mean IoU values of 0.85, 0.78, and 0.82 respectively. The design is deliberately annotation-efficient: only 500 manually labeled bounding boxes are needed to train the detector, and no segmentation weights are updated. The paper also reports a 99.0% end-to-end success rate across 526 clinical images and positions the masks as the size-measurement foundation for automated DESIGN-R assessment.

Load-bearing premise

The reported IoU and DESIGN-R gains rest on the premise that reliable pixel-level ground-truth masks exist for the three test sets; Section 2.2 says nurses annotated them while Sections 3.3, 4.1, and 5.3 say no pixel-wise annotations were available, so if the latter is accurate the quantitative claims have no measurement basis.

Editorial extensions

If this is right

A fixed segmentation backbone can be reused across anatomical locations by adding a small ROI detector, eliminating per-site pixel annotations.
DESIGN-R size scoring becomes computable end-to-end from noisy handheld photos, with the reported accuracy jumping from 71% to 94%.
The detector's small size and speed (about 4.5 MB, roughly 25 ms on a T4 GPU) support deployment on low-resource devices such as nursing-station terminals and bedside carts.
The 99.0% pipeline success rate implies that failures concentrate in degenerate inputs—nearly healed, occluded, or low-contrast wounds—rather than in the segmentation step itself.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the 23-point IoU gain holds under independent pixel-level ground truth, the dominant variable in cross-site wound segmentation is input normalization rather than model capacity.
A natural transfer test would be to apply the same cascade pattern to other pretrained clinical segmenters, using a few hundred boxes to move to new body sites and reduce annotation cost across medical imaging tasks.
The 71% to 94% DESIGN-R size gain is specifically the Size dimension; the paper's own mapping table implies that Exudate, Granulation, and Necrosis would need tissue classifiers inside the mask, so the current claim should not be read as full DESIGN-R automation.
A control experiment worth running is to compare the full pipeline against FUSegNet fed manually cropped ROIs; if manual crops recover the same gain, detector localization precision is not the active ingredient.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear colleague,

The thing to know: this is a plausible pipeline idea—lightweight detector crops an ROI, frozen FUSegNet segments it—but the headline numbers don't survive contact with the paper's own text. The ground-truth situation is contradictory: Section 2.2 and Table 2 say two nurses produced pixel-level masks with Dice 0.93 on all three test sets, while Sections 3.3, 4.1, and 5.3 say no pixel-wise ground truth exists. Table 5's IoU values require those masks. You can't have both, and the quantitative contribution falls apart until that's resolved.

What the paper does well: the application is timely and the method is genuinely lightweight. Training a YOLO detector on 500 boxes and leaving FUSegNet frozen is a sensible way to test whether cropping alone fixes cross-body-site generalization. The random-crop ablation is the right control, and the deployment-speed discussion is grounded in concrete hardware numbers. The specific combination on sacrum and trochanter wounds is not something I've seen in the cited prior work.

The soft spots, in proportion:

1. The contradiction above is load-bearing, not cosmetic. If pixel masks do not exist, Table 5, the 23-pp mean-IoU claim, and the abstract's 94% DESIGN-R accuracy have no measurable foundation. The paper's own limitations section concedes evaluation was primarily visual inspection.

2. The abstract overstates the body. The 23-pp gain is trochanter-only; averaged over Table 5's three sites it's roughly +14 pp. The 71-to-94 DESIGN-R accuracy does not appear anywhere in the results, and the pointer to "Table 3" lands on the YOLO hyperparameter table.

3. Reproducibility is promised, not delivered. No images, code, or weights are attached; full source is withheld for commercial reasons. For a clinical application whose numbers can't be checked, that's a real obstacle.

Who this is for: a reader building an annotation-light wound-segmentation deployment could learn from the architecture and the ablation design. But the preprint, as written, doesn't provide trustworthy numbers to build on. I'd tell the authors to resolve the ground-truth contradiction, report the DESIGN-R result in the body, and actually release the artifacts. If they do that, it becomes a reasonable peer-review candidate. As it stands, I would not spend referee time on this version.

Referee Report

3 major / 3 minor

Summary. The paper proposes a two-stage wound-segmentation pipeline: a YOLOv11n object detector trained on 500 manually labeled bounding boxes crops wound ROIs, which are then fed to a pre-trained FUSegNet segmentation model without fine-tuning. The goal is to generalize from foot-ulcer training data to sacral, trochanter, and other body-site wounds in noisy real-world clinical images. The central claims are a 99.0% end-to-end pipeline success rate on 526 images, mean IoU improvements of up to 23 percentage points over vanilla FUSegNet, and an increase in DESIGN-R size-estimation accuracy from 71% to 94%.

Significance. If the central claims were sound, the paper would make a practically valuable contribution: a lightweight detector front-end that normalizes input ROIs before a fixed, pre-trained segmenter is a cheap and deployment-friendly way to address domain shift in wound segmentation. The authors also make commendable gestures toward reproducibility (planned release of weights, training config, Colab notebook) and include real-world noisy images, failure-case analysis, and IRB approval. However, the paper's own text contains a direct contradiction about the existence of pixel-level ground truth, and the numerical claims in the abstract are not consistent with the tables. As a result, the main quantitative findings are unsubstantiated in the current manuscript.

major comments (3)

[§2.2, Table 2 vs. §3.3, §4.1, §5.3] The 23-percentage-point mean-IoU improvement claimed in the abstract is not supported by Table 5. Table 5 shows per-site improvements over vanilla FUSegNet of +0.04 (foot), +0.16 (sacrum), and +0.23 (trochanter), which averages to roughly +0.14, not +0.23. The +23 pp figure applies only to the trochanter site. Additionally, the abstract refers to 'Table 3' for the DESIGN-R size accuracy increase from 71% to 94%, but Table 3 is the YOLOv11n hyperparameter table and contains no DESIGN-R results; no table or equation in the paper reports the 71% or 94% accuracy figures. These discrepancies require correction before any quantitative conclusion can be drawn.
[§4.1–§4.3, Table 6] The relationship between the 280 test images in Table 2 and the 526 images used for the pipeline success rate in Table 6 is unclear. Section 2.2 and Table 2 describe three test sets of 102 + 86 + 92 = 280 images, all with pixel-wise masks. Sections 4.1–4.3 and Table 6 evaluate a different set of 526 images and report 521/526 successes. If Table 5 is based on the 280-image subset, that should be stated explicitly; if it is based on the 526-image set, then Table 2 and Section 2.2 describe only a subset, and the IoU evaluation in Table 5 must indicate which images it covers. As written, the reader cannot reconcile the two dataset descriptions, which further undermines the reported numbers.
[Title, Abstract, §3.2] The claim of 'annotation-free' segmentation is contradicted by the method itself, which requires 500 manually labeled bounding boxes to train the YOLOv11n detector (§3.2). While the segmentation backbone FUSegNet is not fine-tuned, the overall pipeline is not annotation-free; it is weakly supervised at the box level. The title and abstract should be reworded to 'annotation-light' or 'ROI-label-only' to accurately reflect the supervision required. This is a presentation issue, but it affects the paper's central framing.

minor comments (3)

[§2.2, Table 2] The tick marks in Table 2 for 'Pixel-wise masks' are undercut by the statements in §3.3 and §4.1 that no such masks exist; if the table is retained after revision, it must be accompanied by a description of who created the masks and how they were used.
[§3.2, footnote 1] The model is called YOLOv11n but is described as a custom variant derived from Ultralytics YOLOv8.3.129. Please clarify whether this is a new architecture or a renamed YOLOv8 configuration, and avoid implying a version that does not match the framework cited.
[Table 4] The inference times in Table 4 are a mix of measured values (NVIDIA T4, Raspberry Pi 4) and estimates (laptop, Android, Jetson). Please mark which values are measured and which are estimated, and provide the measurement conditions.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the claimed improvements are empirical comparisons rather than quantities defined by construction; the ground-truth inconsistency is a correctness issue, not circularity.

full rationale

No circular derivation chain is present. The paper is an empirical engineering study: a YOLOv11n detector is trained on bounding boxes (Section 3.2), a pre-trained external FUSegNet backbone is used without fine-tuning, and the reported IoU gains are comparisons on evaluation images rather than consequences of a definitionally linked quantity. No load-bearing self-citation occurs: the only model citation is to FUSegNet [6], an external published network, and the paper's contribution is the detector-crop front end, not an imported uniqueness theorem or ansatz. The most serious issue is internal consistency: Section 3.3 states 'Since pixel-wise ground truth annotations were unavailable for non-foot wounds, we adopted a two-level evaluation strategy'; Section 4.1 repeats 'No pixel-wise ground truth annotations were available'; and Section 5.3 concedes 'the lack of pixel-wise ground truth annotations for non-foot wounds hindered the use of standard quantitative metrics such as the Dice coefficient or Intersection over Union (IoU). Consequently, model performance was evaluated primarily through visual inspection.' Yet Table 5 reports mean IoU and standard deviations for three test sets, and the abstract reports a 23-point IoU gain and 94% DESIGN-R size accuracy. If pixel masks do not exist, those numbers and the DESIGN-R accuracy have no measurable foundation. That is a correctness, reproducibility, and reporting deficiency, but it is not circularity under the specified patterns: the claimed numbers are not identical by construction to any fitted input or to the paper's own definitions; they are simply unsupported by the stated data. There is also no analytic derivation whose conclusion is equivalent to a premise. Therefore the circularity score is 0.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim is empirical; there is no formal derivation, so the ledger contains the assumptions hidden inside the evaluation protocol. The most important entries are the existence and reliability of pixel-level masks for the IoU table, the assumed disjointness between detector training boxes and the evaluation set, and the reliability of the pretrained FUSegNet weights. No invented entities are introduced.

free parameters (1)

Manual bounding-box count = 500
The number of labeled ROIs used to train YOLOv11n is a hand-chosen quantity; the paper claims it suffices but does not report sensitivity to this number.

assumptions (4)

domain assumption Pixel-level ground-truth masks exist for the T-Foot, T-Sacrum, and T-Troch test sets used to compute Table 5 IoU values.
Needed for the IoU numbers; contradicted by Sections 3.3 and 4.1, which state no pixel-wise annotations are available. Section 2.2 and Table 2 are the only support.
domain assumption The 500 bounding boxes used to train YOLOv11n are disjoint from the 526-image evaluation set.
Never stated explicitly; if the test images overlap with detector training, the reported success and IoU numbers are inflated.
domain assumption The two nurses' manual annotations are reliable enough to serve as ground truth (reported inter-rater Dice 0.93).
Section 2.2; without this, the human baseline itself is unvalidated.
domain assumption The publicly released FUSegNet weights from reference [6] behave as described in its own paper and were not modified.
The pipeline treats FUSegNet as a frozen oracle; the paper cites [6] but does not re-verify the segmenter on foot ulcers.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Robust and Annotation-Free Wound Segmentation on Noisy Real-World Pressure Ulcer Images: Towards Automated DESIGN-R\textsuperscript{\textregistered} Assessment." pith.science (2026). https://pith.science/paper/2ME7C4SQ

@misc{pith2026250523392,
  author       = {Pith},
  title        = {Pith review of: Robust and Annotation-Free Wound Segmentation on Noisy Real-World Pressure Ulcer Images: Towards Automated DESIGN-R\textsuperscript\textregistered Assessment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2ME7C4SQ}},
  note         = {Machine review of arXiv:2505.23392}
}

read the original abstract

Purpose: Accurate wound segmentation is essential for automated DESIGN-R scoring. However, existing models such as FUSegNet, which are trained primarily on foot ulcer datasets, often fail to generalize to wounds on other body sites. Methods: We propose an annotation-efficient pipeline that combines a lightweight YOLOv11n-based detector with the pre-trained FUSegNet segmentation model. Instead of relying on pixel-level annotations or retraining for new anatomical regions, our method achieves robust performance using only 500 manually labeled bounding boxes. This zero fine-tuning approach effectively bridges the domain gap and enables direct deployment across diverse wound types. This is an advance not previously demonstrated in the wound segmentation literature. Results: Evaluated on three real-world test sets spanning foot, sacral, and trochanter wounds, our YOLO plus FUSegNet pipeline improved mean IoU by 23 percentage points over vanilla FUSegNet and increased end-to-end DESIGN-R size estimation accuracy from 71 percent to 94 percent (see Table 3 for details). Conclusion: Our pipeline generalizes effectively across body sites without task-specific fine-tuning, demonstrating that minimal supervision, with 500 annotated ROIs, is sufficient for scalable, annotation-light wound segmentation. This capability paves the way for real-world DESIGN-R automation, reducing reliance on pixel-wise labeling, streamlining documentation workflows, and supporting objective and consistent wound scoring in clinical practice. We will publicly release the trained detector weights and configuration to promote reproducibility and facilitate downstream deployment.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

19 extracted references · 6 canonical work pages

[1]

write newline

" write newline " cite write " FUNCTION editor.postfix editor num.names #1 > "( )" "( )" if FUNCTION editor.trans.postfix editor num.names #1 > "( )" "( )" if FUNCTION trans.postfix translator num.names #1 > "( )" "( )" if FUNCTION authors.editors.reflist.apa5 'field := 'dot := field num.names 'numnames := numnames 'format.num.names := format.num.names na...
[2]

sn-aps.bst

FUNCTION identify.aps.version "sn-aps.bst" " [2024/07/19 v1.1 APS bibliography style]" * top ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year eprint archive archivePrefix primaryClass adsurl adsnote version lab...

2024
[3]

write newline

" write newline "" before.all 'output.state := FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap duplicate "7" = swap duplicate "8" = swap "9" = or or or or or or or or or FUNCTION n.separate 't := "" #0 'numnames := t empty not t #-1 #1 subs...
[4]

sn-basic.bst

FUNCTION identify.basic.version "sn-basic.bst" " [2024/07/19 v1.1 bibliography style]" * top ENTRY address archive author booktitle chapter doi edition editor eid eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year archivePrefix primaryClass adsurl adsnote version lab...

2024
[5]

write newline

" write newline "" before.all 'output.state := FUNCTION add.period duplicate empty 'skip "." * add.blank if FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap duplicate "7" = swap duplicate "8" = swap "9" = or or or or or or or or or FUNCTION ...
[6]

write newline

" write newline "" before.all 'output.state := FUNCTION output.doi doi empty skip "doi:" doi * "" * output if FUNCTION format.archive archivePrefix empty "" archivePrefix ":" * if FUNCTION format.primaryClass primaryClass empty "" " [" primaryClass * "] " * if FUNCTION format.eprint eprint empty "" archive empty " https://arxiv.org/abs/" eprint * " " * " ...
[7]

write newline

" write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTION find.integer 't := #0 'int := int not t empty not and t #1 #1 substring 's :=...
[8]

sn-nature.bst

FUNCTION identify.nature.version "sn-nature.bst" " [2024/07/19 v1.1 bibliography style]" * top ENTRY address archive author booktitle chapter edition editor eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url doi volume year archivePrefix primaryClass eid adsurl adsnote version l...

2024

Show all 19 references

[9]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...
[10]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...
[11]

sn-vancouver-num.bst

FUNCTION identify.vancouver.version "sn-vancouver-num.bst" " [2024/07/19 v1.1 Vancouver bibliography style]" * top ENTRY address assignee author booktitle chapter cartographer day edition editor howpublished institution inventor journal key keywords month note number organizat...

2024
[12]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...
[13]

, Sanada , H

barticle Iizaka , S. , Sanada , H. , Matsui , Y. , Furue , M. , Tachibana , T. , Nakayama , T. , Sugama , J. , Furuta , K. , Tachi , M. , Tokunaga , K. , Miyachi , Y. : Clinical wound assessment using design-r total score can predict pressure ulcer healing: a prospective cohor...

2011
[14]

, Halfens , R

barticle Kottner , J. , Halfens , R. , Dassen , T. : An interrater reliability study of the assessment of pressure ulcer risk using the braden scale and the classification of pressure ulcers in a home care setting . International Journal of Nursing Studies 46 ( 10 ), 1307 -- 1...

2009
[15]

: DESIGN-R Scoring Manual

botherref Pressure Ulcers , J.S. : DESIGN-R Scoring Manual. https://www.jspu.org/english/publication/docs/DESIGN-R_manual_eng.pdf. Accessed: 2025-05-08 (2008) botherref

2008
[16]

, Wang , H

barticle Liu , T.J. , Wang , H. , Christian , M. , Chang , C.-W. , Lai , F. , Tai , H.-C. : Automatic segmentation and measurement of pressure injuries using deep learning models and a lidar camera . Scientific Reports 13 ( 1 ), 680 ( 2023 ) 10.1038/s41598-022-26812-9 barticle

2023 doi
[17]

, Anisuzzaman , D

barticle Wang , C. , Anisuzzaman , D. , Williamson , V. , Dhar , M.K. , Rostami , B. , Niezgoda , J. , Gopalakrishnan , S. , Yu , Z. : Fully automatic wound segmentation with deep convolutional neural networks . Scientific Reports 10 ( 1 ), 21854 ( 2020 ) barticle

2020
[18]

, Zhang , T

botherref Dhar , M.K. , Zhang , T. , Patel , Y. , Gopalakrishnan , S. , Yu , Z. : FUSegNet: A Deep Convolutional Neural Network for Foot Ulcer Segmentation (2024). https://arxiv.org/abs/2305.02961 botherref

2024 arXiv
[19]

write newline

" write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTI...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

write newline

" write newline " cite write " FUNCTION editor.postfix editor num.names #1 > "( )" "( )" if FUNCTION editor.trans.postfix editor num.names #1 > "( )" "( )" if FUNCTION trans.postfix translator num.names #1 > "( )" "( )" if FUNCTION authors.editors.reflist.apa5 'field := 'dot := field num.names 'numnames := numnames 'format.num.names := format.num.names na...

[2] [2]

sn-aps.bst

FUNCTION identify.aps.version "sn-aps.bst" " [2024/07/19 v1.1 APS bibliography style]" * top ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year eprint archive archivePrefix primaryClass adsurl adsnote version lab...

2024

[3] [3]

write newline

" write newline "" before.all 'output.state := FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap duplicate "7" = swap duplicate "8" = swap "9" = or or or or or or or or or FUNCTION n.separate 't := "" #0 'numnames := t empty not t #-1 #1 subs...

[4] [4]

sn-basic.bst

FUNCTION identify.basic.version "sn-basic.bst" " [2024/07/19 v1.1 bibliography style]" * top ENTRY address archive author booktitle chapter doi edition editor eid eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year archivePrefix primaryClass adsurl adsnote version lab...

2024

[5] [5]

write newline

" write newline "" before.all 'output.state := FUNCTION add.period duplicate empty 'skip "." * add.blank if FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap duplicate "7" = swap duplicate "8" = swap "9" = or or or or or or or or or FUNCTION ...

[6] [6]

write newline

" write newline "" before.all 'output.state := FUNCTION output.doi doi empty skip "doi:" doi * "" * output if FUNCTION format.archive archivePrefix empty "" archivePrefix ":" * if FUNCTION format.primaryClass primaryClass empty "" " [" primaryClass * "] " * if FUNCTION format.eprint eprint empty "" archive empty " https://arxiv.org/abs/" eprint * " " * " ...

[7] [7]

write newline

" write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTION find.integer 't := #0 'int := int not t empty not and t #1 #1 substring 's :=...

[8] [8]

sn-nature.bst

FUNCTION identify.nature.version "sn-nature.bst" " [2024/07/19 v1.1 bibliography style]" * top ENTRY address archive author booktitle chapter edition editor eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url doi volume year archivePrefix primaryClass eid adsurl adsnote version l...

2024