Related Experiment Video
Updated: Jul 1, 2026
Whole-body PET/MRI of Pediatric Patients: The Details That Matter
Published on: December 19, 2017
AI-powered segmentation and prognosis with missing MRI in pediatric brain tumors
Dimosthenis Chrysochoou1,2, Deep B Gandhi2,3, Sahand Adib2
1University of Pennsylvania, Department of Bioengineering, Philadelphia, PA, USA.
Abstract:
Brain MRI is the primary imaging modality for pediatric brain tumors, yet incomplete acquisitions are common, hindering the clinical utility of existing deep learning models for tumor segmentation and prognosis. These models are typically trained on complete MRI sequences and exhibit performance degradation when MRI sequences are missing at test time. In this retrospective study of 715 patients from the Children's Brain Tumor Network and BraTS-PEDs, and 43 patients with 157 longitudinal MRIs from PNOC003/007 clinical trials, we developed strategies for handling missing sequences. Methods included a dropout-trained segmentation model that randomly omitted FLAIR and/or T1w inputs during training, a generative model for image synthesis, copy-substitution heuristics, and zeroed inputs. The dropout model achieved robust segmentation under missing MRI, with ≤0.04 Dice drop relative to complete-input and stable prognostic accuracy in survival analysis using model-derived tumor volumes and clinical covariates. Generative synthesis achieved high image quality (SSIM > 0.90) and removed artifacts, benefiting visual interpretability. Together, these approaches can facilitate broader deployment of AI tools in real-world pediatric neuro-oncology settings.
Insights
Deep learning models for pediatric brain tumor segmentation struggle with incomplete MRI data. A new dropout-trained model maintains segmentation accuracy and prognostic performance even with missing MRI sequences.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Pediatric Oncology
Background:
- Brain MRI is crucial for pediatric brain tumor diagnosis and treatment planning.
- Deep learning models for tumor segmentation and prognosis require complete MRI sequences.
- Incomplete MRI acquisitions are frequent, limiting AI tool clinical utility.
Purpose of the Study:
- To develop and evaluate strategies for handling missing MRI sequences in pediatric brain tumor analysis.
- To improve the robustness and clinical applicability of AI models for pediatric neuro-oncology.
Main Methods:
- Retrospective study of 715 patients from Children's Brain Tumor Network and BraTS-PEDs, plus 43 patients from PNOC003/007 trials.
- Development of a dropout-trained segmentation model, generative image synthesis, copy-substitution, and zeroed inputs.
- Evaluation of segmentation performance (Dice score) and prognostic accuracy using model-derived tumor volumes.
Main Results:
- The dropout model demonstrated robust segmentation with minimal performance drop (≤0.04 Dice) on incomplete MRI data.
- Generative synthesis produced high-quality images (SSIM > 0.90) and reduced artifacts.
- The dropout model maintained stable prognostic accuracy in survival analysis.
Conclusions:
- AI models trained with input dropout can effectively handle missing MRI sequences in pediatric brain tumors.
- Generative synthesis enhances image quality and interpretability.
- These methods support the broader implementation of AI in pediatric neuro-oncology settings.
