PedDx: An End-to-End Deep Learning Framework for Pediatric Posterior Fossa Tumor Diagnosis with Agentic AI-Enabled Independent External Evaluation
Reliable clinical translation of diagnostic artificial intelligence (AI) requires evaluation beyond classification accuracy. We developed PedDx, a segmentation-free framework for pediatric posterior fossa tumor diagnosis using T2-weighted MRI, pretrained 3DINO representations, calibrated predictions, and confidence-based abstention. Evaluation included 492 discovery, 114 replication, 129 external validation, and 40 independently tested cases, alongside 236 out-of-distribution (OOD) cases and controlled imaging perturbations. Classification accuracy was 79.8% in replication and 81.4% in external validation, increasing to 90.5% and 88.7%, respectively, among non-abstained cases. OOD analyses revealed performance degradation and confidently incorrect predictions, while perturbation analyses identified acquisition and alignment vulnerabilities. An external institution independently executed the complete agentic workflow, successfully generating predictions and structured quality-control reports for all 40 cases, with classification accuracy of 92.5%, increasing to 97.3% among non-abstained cases. PedDx demonstrates how uncertainty assessment, failure-mode characterization, external validation, and independently reproducible, traceable reporting can collectively inform the reliability and limitations of diagnostic AI systems.
Authors
- Anahita Fathi Kazerooni (ORCID: https://orcid.org/0000-0001-7131-2261)
- Mikaela Doig (ORCID: https://orcid.org/0000-0002-9912-4489)
- Kartik M. Reddy (ORCID: https://orcid.org/0000-0001-7435-1864)
- Ariana Familiar (ORCID: https://orcid.org/0000-0002-9315-8653)
- Arastoo Vossough (ORCID: https://orcid.org/0000-0003-1346-427X)
- Nastaran Khalili (ORCID: https://orcid.org/0000-0001-8078-3591)
- Amin Tavallaii (ORCID: https://orcid.org/0000-0003-2486-2919)
- Jordan R. Hansford (ORCID: https://orcid.org/0000-0001-7733-383X)
- Sahand Adib Moradi
- Ghasem Azemi (ORCID: https://orcid.org/0000-0002-8675-6858)
- Phillip B. Storm (ORCID: https://orcid.org/0000-0002-7964-2449)
- Deep B. Gandhi (ORCID: https://orcid.org/0000-0002-9479-5427)
- Eswar Damaraju (ORCID: https://orcid.org/0000-0002-5345-3551)
- Ali Nabavizadeh (ORCID: https://orcid.org/0000-0002-0380-4552)
- Meen Chul Kim (ORCID: https://orcid.org/0000-0002-0308-783X)
- Andrew Chiang
- Adam C. Resnick (ORCID: https://orcid.org/0000-0003-0436-4189)
- Sanaz Varshochi (ORCID: https://orcid.org/0000-0003-0771-0729)
- Minh‐Son To (ORCID: https://orcid.org/0000-0002-8060-6218)
- Paarth Jain
- Fadel Batal
- Anurag Gottipati (ORCID: https://orcid.org/0000-0003-3871-8133)
- Dimosthenis Chrysochoou
- Neda Khalili (ORCID: https://orcid.org/0000-0001-8415-8996)
- Yog Vaghani
- Bhavyasri Vunnava (ORCID: https://orcid.org/0009-0000-5995-0498)
- Amanda Luck
- Antonio Di Ieva
- Pengjia Zhu
- Celine Lee
Institutions
- Children's Hospital of Philadelphia (US)
- Emory University (US)
- Flinders University (AU)
- Emory University Hospital (US)
- Women's and Children's Hospital (AU)
- Nepean Hospital (AU)
- Flinders Medical Centre (AU)
- South Australian Health and Medical Research Institute (AU)
- Children's Healthcare of Atlanta (US)
- University of Pennsylvania (US)
- Macquarie University (AU)
Publication Details
- Journal
- medRxiv
- Published
- 2026-10-29
- DOI
- https://doi.org/10.64898/2026.10.27.26366616
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- preprint
Funders
- Tour de Cure
- National Institutes of Health