CT model distinguishes recorded psychiatric diagnosis groups
Source check: Source checked. Not independent scientific confirmation.
Finding: A DenseNet-based multiple-instance model classified CT scans from patients with bipolar disorder or schizophrenia and healthy controls. Development used 685 people, with another 105 from two additional centers for testing; the paper reports 72.4% external accuracy. Cases and healthy controls came from different hospitals, so acquisition or site differences may contribute to classification. The retrospective result does not establish a diagnostic brain mechanism, screening benefit or performance beyond these selected groups.
Technology: DenseMD three-dimensional DenseNet multiple-instance learning
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- A shared three-dimensional DenseNet encoder learns representations from eight brain-CT patches per person, fuses them and classifies recorded clinical groups through multiple-instance learning.
- Test and result
- Retrospective three-group classification against DSM-V-based clinical diagnoses and healthy-control assessments. Training/validation used 506 psychiatric patients and 179 controls; external testing used 65 patients and 40 controls, with Table2 accuracy .724. Primary Fig2 explicitly shows Normal, Bipolar and Schizo outputs, resolving an inconsistent binary-task phrase in the Results.
- Limitation / still unknown
- Performance independent of hospital/scanner confounding, medication effects and selection bias remains unestablished. Patients and controls were sourced from different centers in both development and testing. No screening effectiveness or patient benefit was tested.
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/PMC12092902/
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: Primary relevant recruitment/reference, patient partition, DenseMD model Methods, Table1–3 and primary Fig2, Results and limitations; supplementary scanner table not inspected. Exact primary identity/date and fresh primary PubMed correction/retraction relationships 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 2/5 · 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.