Pediatric posterior fossa tumour classifier from T2 MRI with abstention, evaluated at an outside site (PedDx)
A preprint reports a classifier for five pediatric posterior fossa tumour types from a single T2-weighted MRI, which abstains when unsure, reaching 81.4% accuracy on 129 external cases (88.7% when it did not abstain) and 92.5% on 40 cases run through the full workflow by a team at another institution, while degrading on out-of-distribution scans, sometimes confidently.
- Model
- PedDx, a frozen 3DINO vision transformer with a classification head, calibrated, with confidence-based abstention
- Field
- Medicine
- Date
- 2026-10-29
What was found
The classifier separates medulloblastoma, ependymoma, atypical teratoid/rhabdoid tumour, low-grade glioma and diffuse midline glioma without needing a tumour segmentation. It was trained on 492 treatment-naive scans from the Children's Brain Tumor Network and CHOP and tested on 114 held-out replication cases (79.8%, or 90.5% among cases where it did not abstain), and on 129 cases from Children's Healthcare of Atlanta. A team at Macquarie University ran the deployed workflow, which produces a prediction and a structured quality-control report, on 40 cases from the Women's and Children's Hospital in Adelaide. The authors also tested 236 out-of-distribution cases, including post-treatment scans, artefacts and tumour types never seen in training, and found degraded performance with confidently wrong predictions, plus sensitivity to acquisition and alignment perturbations.
Novelty check
Europe PMC, 2026-11-05: TITLE:"posterior fossa" AND ("deep learning" OR "artificial intelligence" OR "machine learning") AND (pediatric OR paediatric OR children) returned 68 records, including a multi-institutional deep learning study of detection and classification (AJNR 2020), automatic machine learning on routine MRI (AJNR 2020) and a multinational machine-assist study (Neurosurgery 2021). Classifying these tumours from MRI is established; what this preprint adds is calibrated abstention, systematic failure-mode testing and an outside team running the whole workflow.
Caveats and known objections
A medRxiv preprint, not peer-reviewed, so capped at claimed. The title calls the 40-case evaluation independent, but the data came from the Women's and Children's Hospital in Adelaide, two of whose clinicians are among the paper's 30 authors, so it is a collaborating site rather than an unaffiliated replication; it is also small. The 129-case external cohort is 102 medulloblastomas, so overall accuracy leans on one class and the other diagnoses have few cases. Retrospective; ground truth is histopathology from records, except diffuse midline glioma, which was sometimes diagnosed radiologically. The endpoint is diagnostic accuracy, a surrogate, and no clinician used the system in care. Funding, competing interests and data or code availability were not found in the medRxiv text read for this entry. Autonomy is ai-assisted, the weaker reading: the system is a diagnostic aid that defers to clinicians when it abstains, and the agentic label refers to the automated workflow, not to any decision taken without a clinician.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Claimed and AI-assisted.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded claimed for verification and ai-assisted for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). Pediatric posterior fossa tumour classifier from T2 MRI with abstention, evaluated at an outside site (PedDx). whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-29-peddx-posterior-fossa-tumoursBibTeX
@misc{whataifound-childrenshospitalofphiladelphia-2026-tumours,
title = {Pediatric posterior fossa tumour classifier from T2 MRI with abstention, evaluated at an outside site (PedDx)},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by Children's Hospital of Philadelphia. Verification: Claimed. Autonomy: AI-assisted.},
url = {https://whataifound.org/finding/2026-10-29-peddx-posterior-fossa-tumours}
}