Skip to content
Computational Pathology Reading Roadmap

Writing study: SegMoTE: Token-Level Mixture of Experts for Medical Image Segmentation

← CVPR 2026 oral writing corpus

Yujie Lu, Jingwen Li, Sibo Ju, Yanzhou Su, He Yao, Yisong Liu, Min Zhu, Junlong Cheng · Official CVPR 2026 oral record

Figure, table and section ledger

Editorial contribution type: Architecture & representation (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
11 physical pages
Before references / appendix
≈ 8.05 page-equivalents
Main figures / tables
7 / 4
Attached appendix figures / tables
0 / 0
Attached appendix span
Separate file recorded below; not in this PDF total
Before references / appendix: ≈ 73.18%References / intervening material: ≈ 26.82%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Extraction notes: No attached appendix boundary detected. Zero means no recognized caption, not independently verified absence.

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.

Top-level headingStarts on PDF pageEstimated spanShare of entire PDF
1. Introduction1≈ 1.45 pp.13.18%
2. Related Work2≈ 0.56 pp.5.09%
3. Method3≈ 2.53 pp.23%
4. Experiments5≈ 2.87 pp.26.09%
5. Conclusion8≈ 0.26 pp.2.36%
References9≈ 1.02 pp.9.27%
6. Ieee, 2020. 610≈ 1.94 pp.17.64%
All detected figure and table locations

Main text

Attached appendix

    No recognized caption labels.

    Separate source files—excluded from canonical-PDF totals

    A separate author version may repeat main text and references. Whole-file lengths and counts are not extra appendix material and are never added to the canonical PDF totals. Read the study's source scope for which sections received substantive reading versus layout inspection.

    Additional source 1

    Separate PDF (or ZIP containing it): whole-file layout of 6 physical pages; 5 detected figures and 4 tables. These may include overlapping main-text material.

    Located appendix only: starts on physical p.1; ≈ 5.94 page-equivalents, 5 figure labels and 4 table labels. This interval is distinct from the whole file.

    Detected headingPDF pageEstimated span
    Supplementary Material1≈ 0.08 pp.
    6. More Details of MedSeg-HQ1≈ 1.49 pp.
    7. More Visual Results Comparison2≈ 0.41 pp.
    8. More Quantitative Comparisons3≈ 3.95 pp.

    Argument and organization

    1. 3. Method

      Explains expert tokens, routing, prompt tokenization, objective, and data curation.

    2. 4. Experiments

      Separates in-domain, out-of-domain, decoder-unfreezing, and expert-configuration evidence.

    Problem, claim and contribution

    The paper jointly identifies modality heterogeneity and unreliable heterogeneous supervision, linking model adaptation with data curation.

    Section 1 · Main paper · Additional source 2

    Argument and section order

    The data-construction subsection follows architectural/loss details, while evaluation distinguishes the frozen-decoder setting from an unfrozen extension.

    §3.6; §4.2 · Main paper · Additional source 2

    Figures, captions and visual explanation

    The wide overview marks frozen and trainable modules and displays original-token versus expert-token outputs, making the changed decoder interface visible.

    Figure 2 p. 4 · Main paper · Additional source 2

    Tables, comparisons and evidence

    Out-of-domain results include organ-level rows and preserve exceptions; a separate table explicitly labels the decoder-unfreezing experiment.

    Tables 1–2 p. 7 · Main paper · Additional source 2

    Appendix and reproducibility

    The supplement reports dataset composition and prompt conditions, and explicitly notes that nnU-Net is automatic whereas SegMoTE uses interaction.

    Supplement Table 5 and Table 8 · Main paper · Additional source 2

    Clarity and writing conventions

    Expert-selection statistics and configuration sweeps are presented as diagnostics alongside segmentation metrics rather than only a model diagram.

    Figures 6–7 p. 8 · Main paper · Additional source 2

    What to reuse—and what not to infer

    Transferable lesson: For interactive adaptation, keep prompt privileges, frozen/unfrozen parameters, dataset curation, and organ-level transfer conditions visible.

    Boundary: Interactive and automatic segmentation scores are not like-for-like without an interaction budget; low-dimensional feature plots do not prove data quality.

    Apply this lesson using the argument and evidence worksheet.

    Inspection record and sources

    Checked
    2026-09-26
    Reading status
    Writing-focused manuscript scan + selected visuals inspected
    Main text
    Manuscript-wide writing scan of main PDF pp. 1–8: framing, section structure, method exposition, figure/table captions, evaluation organization and conclusions; not an exhaustive technical proof audit or reproduction.
    Appendix / supplement
    Scanned six supplementary pages on MedSeg-HQ composition, click/box visual comparisons, loss-weight ablations, fine-tuning baselines, and a qualified comparison with automatic nnU-Net.
    Visually inspected PDF pages
    4, 7

    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.