Writing study: Graph-Semantic Guided Learning for Virtual Immunohistochemistry Staining on Consecutive Histology Sections
← AAAI 2026 oral writing corpus
Fanhao Qiu, Yangyang Zhang, Zhengxia Wang · Official AAAI 2026 oral record
Figure, table and section ledger
Editorial contribution type: Domain-grounded methods & physical systems (primary).
Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.
- Canonical PDF
- 9 physical pages
- Before references / appendix
- ≈ 7.06 page-equivalents
- Main figures / tables
- 4 / 4
- Attached appendix figures / tables
- 0 / 0
- Attached appendix span
- No boundary detected; not assumed zero
Extraction notes: No attached appendix boundary detected. Zero means no recognized caption, not independently verified absence.
Source-bound correction: Checked 2026-10-25 against the writing read and physical-page bounding boxes. Unnumbered main headings have extracted glyph height 10.7475, distinct from 9.8076-height subsections. Main figure labels 1–4 and tables 1–4 match the captions; no caption exclusion/restoration needed. Viewed full-page rasters 2, 5, 6. No attached appendix. This is the official nine-page publisher PDF only.
Section-by-section lengths and source pages
The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.
All detected figure and table locations
Main text
Attached appendix
No recognized caption labels.
Argument and organization
- Introduction; Related Works
Starts from consecutive-section acquisition and cell-level semantic mismatch, then positions virtual staining, graph modeling and semantic supervision before the proposed method.
- Method
Explains graph construction, graph-semantic rectification, the two-branch discriminator and the combined framework; ties biological descriptors and graph losses to the stated mismatch problem.
- Experiments
Moves from datasets and metrics to cross-method comparisons, component ablations, node-feature tests, a rectification example and generator-size alternatives.
- Conclusion; Acknowledgments; References
Closes the methodological claim before production back matter; no separate limitations section or attached appendix appears.
Problem, claim and contribution
The problem is not simply image translation: adjacent H&E and IHC sections contain spatially mismatched cells. The introduction turns that acquisition constraint into two design requirements—cell-level pathological representation and correction of noisy semantic correspondence—before naming the modules.
Introduction, physical pp.1–3; Method opening and Graph Construction, p.3 · Main paper
Argument and section order
Graph Construction introduces cell segmentation, contextual patch descriptors, coordinates and IHC-derived features before the rectification losses use them. Separate GSRM and discriminator explanations then converge in the full objective. The experiment subsections revisit these decisions rather than using a single undifferentiated results narrative.
Method, pp.3–5; Equation 9 and Experiments opening, p.5; ablations and generator discussion, pp.6–7 · Main paper
Figures, captions and visual explanation
Figure 1 is a full-width, four-panel mechanism diagram that expands the cell-feature construction and graph correction instead of showing only generator boxes. Figures 2–3 align input, competing outputs and reference tissue by row, with the component ablation following the same visual convention. A clearer adaptation would explicitly label training-only target-IHC signals and avoid treating adjacent-section reference images as pixel-identical truth.
Figure 1, p.2; Figure 2, p.5; Figure 3, p.6 · Main paper
Tables, comparisons and evidence
The main comparison separates MIST and BCI and gives directions for FID, KID, SSIM and DAB histogram divergence. Table 2 is a two-by-two component ablation: the full method improves several measures, but DBD-only has higher SSIM (0.251 versus 0.248). The text itself explains why SSIM is limited under weak pairing. The inspected tables do not provide repeated-run uncertainty; selected tissue images and distributional image metrics do not establish diagnostic performance.
Evaluation Metrics, p.5; Tables 1–2 and Ablation Study, p.6; Tables 3–4, p.7 · Main paper
Appendix and reproducibility
The inspected proceedings file keeps the training objective, experimental setup, feature ablation and generator-size comparison in its main body. It supplies no attached appendix to inspect. An adaptation should provide detailed preprocessing and reproducibility material without claiming that this paper's unavailable supplementary protocol was audited.
Method and Experimental Setup, pp.3–5; Tables 3–4, p.7; acknowledgments/reference boundary, p.8 · Main paper
Clarity and writing conventions
Terminology repeatedly connects the cell graph to semantic consistency, making the domain-to-method argument easy to follow. However, broad clinical-value and paradigm language exceeds what the benchmark comparisons alone establish. The useful writing pattern is the explicit acquisition constraint and metric qualification, not promotional wording or an implication of validated clinical deployment.
Introduction, pp.1–3; Evaluation Metrics, p.5; Conclusion, p.7 · Main paper
What to reuse—and what not to infer
Transferable lesson: For a medical-imaging application, make the acquisition mismatch concrete, show how domain signals become model inputs and losses, and pair aligned qualitative panels with component-level numeric evidence. Keep reference images, surrogate metrics and clinical outcomes conceptually separate.
Boundary: This study analyzes writing, not clinical validity. Consecutive sections are weakly paired, the main tables do not report uncertainty, and the full model does not win every metric. The inspected file has no attached appendix. Neither attractive staining examples nor favorable FID establish biomarker accuracy or safe diagnostic substitution.
Apply this lesson using the argument and evidence worksheet.
Inspection record and sources
- Checked
- 2026-10-25
- Reading status
- Writing-focused manuscript scan + selected visuals inspected
- Main text
- Writing-oriented manuscript-wide read of physical pp.1–7, covering the acquisition problem, related work, graph construction, losses, evaluation, discussion and conclusion; acknowledgments and reference boundary on p.8 and reference continuation through p.9 checked. Full-page rasters of pp.2, 5 and 6 were viewed. Not independent clinical validation, code inspection or experimental reproduction.
- Appendix / supplement
- No appendix is attached to this nine-page proceedings PDF, and no appendix routing statement was identified in the main-text read. No separate supplement, code or presentation asset was inspected. This is a statement about the inspected file, not proof that no other materials exist.
- Visually inspected PDF pages
- 2, 5, 6
This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.