PedDx: An End-to-End Deep Learning Framework for Pediatric Posterior Fossa Tumor Diagnosis with Agentic AI-Enabled Independent External Evaluation
- 1Center for Data-Driven Discovery in Biomedicine (D3b), The Children's Hospital of Philadelphia, Philadelphia, PA, USA; Division of Neurosurgery, The Children's
Abstract
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.
Related
- Malnutrition Diagnosis: Validation of a Weight Independent GLIM adapted Framework in U.S. Adults (NHANES 1999 2020)
- Unsupervised machine learning identifies two distinct posterior tibial slope clusters largely independent of Coronal Plane Alignment of the Knee in osteoarthritic knees
- Unreconciled postoperative imaging limits interpretation of gross-total resection in pediatric posterior fossa dermoids
- Posterior Plating Through a Single Posterolateral Approach for a Comminuted Thurston-Holland Fragment in a Pediatric Triplane Ankle Fracture
- Shared reading and independent book engagement in early childhood: associations with the home literacy environment in a Turkish pediatric outpatient population
BibTeX
@misc{kazerooni2026peddx,
title = {{PedDx: An End-to-End Deep Learning Framework for Pediatric Posterior Fossa Tumor Diagnosis with Agentic AI-Enabled Independent External Evaluation}},
author = {A. Fathi Kazerooni and F. Batal and B. Vunnava and E. Damaraju and A. Tavallaii and D. Gandhi and A. Gottipati and P. Zhu and D. Chrysochoou and S. Varshochi and S. Adib Moradi and A. M. Familiar and A. Chiang and Y. Vaghani and G. Azemi and M. Doig and M.-S. To and A. Luck and M. C. Kim and N. Khalili and N. Khalili and P. Jain and C. Lee and J. R. Hansford and A. Di Ieva and P. B. Storm and A. C. Resnick and K. Reddy and A. Vossough and A. Nabavizadeh},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.27.26366616},
url = {https://www.medrxiv.org/content/10.64898/2026.10.27.26366616v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - PedDx: An End-to-End Deep Learning Framework for Pediatric Posterior Fossa Tumor Diagnosis with Agentic AI-Enabled Independent External Evaluation
AU - A. Fathi Kazerooni
AU - F. Batal
AU - B. Vunnava
AU - E. Damaraju
AU - A. Tavallaii
AU - D. Gandhi
AU - A. Gottipati
AU - P. Zhu
AU - D. Chrysochoou
AU - S. Varshochi
AU - S. Adib Moradi
AU - A. M. Familiar
AU - A. Chiang
AU - Y. Vaghani
AU - G. Azemi
AU - M. Doig
AU - M.-S. To
AU - A. Luck
AU - M. C. Kim
AU - N. Khalili
AU - N. Khalili
AU - P. Jain
AU - C. Lee
AU - J. R. Hansford
AU - A. Di Ieva
AU - P. B. Storm
AU - A. C. Resnick
AU - K. Reddy
AU - A. Vossough
AU - A. Nabavizadeh
PY - 2026
DA - 2026/10/27
PB - medRxiv
DO - 10.64898/2026.10.27.26366616
UR - https://www.medrxiv.org/content/10.64898/2026.10.27.26366616v1
AB - 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.
ER -