Clinical SPECT score validation finds learned estimates agree less well for larger defects
Source check: Source checked. Not independent scientific confirmation.
Finding: A neural scorer trained on a Japanese multicenter SPECT resource showed substantial agreement with expert perfusion scores in a small internal test, while errors increased for more severe defects. The source reports 79.6% exact segment agreement, with most source segments normal or mildly abnormal. Scan-level sample accounting and exchanged validation/test roles limit any claim of independent patient-level diagnostic performance.
Technology: U-Net segment localization with concurrent CNN features and fully connected perfusion-score estimation
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- The learned network localized myocardial segments and combined image features into scores, enabling the represented clinical expert-agreement and severity-related error analysis.
- Test and result
- The source retained 1,288 people from J-ACCESS 3/4 and analyzed stress/rest volumes separately. Expert 17-segment perfusion grades trained a U-Net/CNN scorer. Two twenty-volume validation/test sets exchanged roles, yielding forty reported test studies and 680 segment comparisons. Exact segment agreement was 79.6%; error increased with defect severity and was worse in the female dataset.
- Limitation / still unknown
- The paper inconsistently calls doubled stress/rest data patients and reports differing scan totals; patient-level isolation of paired scans across training/validation/test is not established. Exchanging validation and test roles compromises an untouched-test interpretation. Ninety-six percent of source segments were normal or mildly abnormal, so aggregate agreement does not establish sensitivity for severe disease. Expert interpretation is the reference, not invasive disease truth or measured subsequent prognosis. The test is small and no clinical benefit was demonstrated.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC11538165/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 21 Nov 2026. Review scope: Read primary clinical source/sample accounting, expert scoring reference, U-Net/CNN and validation/test Methods, agreement/severity-specific Results and prognostic-category limits. Represented test units are scan studies, without asserting forty independent held-out people. Exact original article front DOI/title and fresh PubMed primary relationship metadata checked..
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.