Admission images and records classify COVID outcomes across six Italian hospitals
Source check: Source checked. Not independent scientific confirmation.
Finding: A December 2020 preprint studied admission chest X-rays and records from 820 hospitalized patients. A combined image-feature and clinical classifier reported mean accuracy 0.743 in leave-one-center-out evaluation for a composite severe outcome. This retrospective first-wave study does not establish prospective benefit or superiority over clinical records alone.
Technology: Convolutional image features with classical supervised classifiers
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Neural features extracted from admission radiographs were combined with clinical variables by supervised classifiers to estimate the recorded composite outcome.
- Test and result
- The six-hospital cohort had 820 admission examinations, with outcomes classified as mild versus non-invasive ventilation, intensive care or death. Training-fold normalization, feature selection and parameter fitting were described. In original v1 Table 5, GoogleNet image features plus clinical data and logistic regression achieved mean accuracy .743±.061, sensitivity .769±.189 and specificity .685±.155 across held-out centers; clinical-data SVM accuracy was .734±.044. The later August 2021 journal tables retain these values.
- Limitation / still unknown
- Many configurations were compared, and displayed best performances do not constitute an untouched locked-model evaluation. Center-held-out evaluation reuses one first-wave study rather than independent replication. Severe outcome combines several care decisions and mortality; the results do not establish time-to-event prognosis, treatment effect, capacity improvement or clinically significant added value over records alone. The v1 archive says submitted December 11, 2020, while its PDF carries a December 14 version header; this entry uses the dated PDF version and does not assert a first-public-day priority. Journal publication was August 28, 2021.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2012.06531v1 · https://arxiv.org/pdf/2012.06531v1 · https://pmc.ncbi.nlm.nih.gov/articles/PMC8401374/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: Original v1 archive metadata and PDF clinical-data/model-validation Methods and Tables4/5; exact results compared with later journal full text. Version-header and submission dates remain distinguished..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 1/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.