Phase 1 Classifier Diagnosis and Plan v2.2 (pre-registration document)
Pre-registration document for a seed-controlled, pre-registered evaluation protocol for multi-label classification of canine and feline thoracic radiographs. Frozen before any reporting-split readout; specifies four conservative gates (macro mAP ≥ 0.56, calibrated macro F1 ≥ 0.52, marginal-tier AP ≥ 0.38, research-tier AP ≥ 0.28), six per-class floors with derivation max(0.60, AP_base − 0.02), a pre-registered STOP rule, and an 18→15 run budget. Version 2.2; document amendments are logged in §11.3. SHA-256 of the canonical markdown source in this record is 0cab8596fbf4fd561c7b1de3c975f0708f403cccfdaca652803dce926db21843; the accompanying PDF is a rendered copy. The source was scanned for local filesystem paths and machine identifiers; none occur, so no substitutions were made, and the accompanying substitution log records zero redactions in every class.
Authors
- Lian He (ORCID: https://orcid.org/0000-0002-9909-6107)
Institutions
- Taizhou University (CN)
- Taizhou University (CN)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24201579
- Primary Topic
- COVID-19 diagnosis using AI
- Type
- preprint