Delwedd Flaen

Wedi'i Gadw mewn:
Manylion Llyfryddiaeth
Prif Awdur: Musharu, Timothy
Fformat: Recurso digital
Iaith:Saesneg
Cyhoeddwyd: Zenodo 2026
Pynciau:
Medical Informatics
Alopecia
Dermatology/methods
Latent Diffusion
Negative Results/standards
clinical image synthesis
Reproducibility of Results
QA-Gated Evaluation
Severity Synthesis
generative prognosis
conditional diffusion
proxy baseline
computational reproducibility
python
Mynediad Ar-lein:https://doi.org/10.5281/zenodo.20376566
Tagiau: Ychwanegu Tag
Dim Tagiau, Byddwch y cyntaf i dagio'r cofnod hwn!
Tabl Cynhwysion:
  • <p>This software archive provides reproducibility code and computational artefacts for the MDPI Informatics manuscript "When Severity Response Is Not Enough: A Systematic Evaluation of Latent Diffusion Approaches for Computational Alopecia Prognosis Synthesis" (submitted May 2026).</p> <p>Software Components:<br>• Experiment scripts for Experiments 01–06 (proxy baseline, residual editor, latent editor, LoRA adaptation, severity-delta diffusion)<br>• QA-gate implementation (auto-pass, monotonicity, span, drift metrics)<br>• Evaluation pipeline (LPIPS, FID, Guard L1, contact-sheet generation)<br>• Data preprocessing utilities (manifest builder, LCH augmentation, non-IID partitioning)<br>• Proxy-generation utilities (mask parameter control, severity labeling)</p> <p>Data Artifacts:<br>• Proxy-pair metadata manifest (78 QA-gated supervision pairs with mask parameters, severity labels, case IDs)<br>• Fixed split definitions (validation: sparse2, test: sparse4)<br>• Evaluation summaries (per-case results for all experiments in JSON format)<br>• Supplementary Table S1 (per-case auto-pass, monotonicity, span, drift results)</p> <p>Manuscript Materials:<br>• LaTeX source and compiled PDFs<br>• Supplementary materials</p> <p>Exclusions (see RESTRICTED_ASSETS.md):<br>• Original source scalp images from UniDataPro (third-party licensing restrictions; access instructions in SOURCE_DATA_ACCESS.md)<br>• Full-resolution generated contact sheets (contain third-party source images)<br>• Trained model checkpoints (>5 GB; available on request from corresponding author)</p> <p>This archive enables full reproduction of Tables 1–3, Figures 1–3, and Supplementary Table S1 from the manuscript, given access to the original UniDataPro source images and a Python 3.9+ environment (see requirements.txt).</p> <p>License: Apache-2.0 (software and code); third-party data subject to UniDataPro terms (see DATA_AVAILABILITY.md).</p>

Eitemau Tebyg