AI Tool for Pediatric Brain Tumor Imaging
Author Information
Author(s): Gandhi Deep B, Khalili Nastaran, Familiar Ariana M, Gottipati Anurag, Khalili Neda, Tu Wenxin, Haldar Shuvanjan, Anderson Hannah, Viswanathan Karthik, Storm Phillip B, Ware Jeffrey B, Resnick Adam, Vossough Arastoo, Nabavizadeh Ali, Fathi Kazerooni Anahita
Primary Institution: The Children’s Hospital of Philadelphia
Hypothesis
We hypothesize that this pipeline can establish a standardized method for preprocessing pediatric brain MRI acquisitions.
Conclusion
Our skull-stripping models demonstrate excellent performance and include sellar/suprasellar regions, using single- or multi-parametric inputs.
Supporting Evidence
- Deep learning models for skull-stripping demonstrate robustness across various pediatric brain tumor histologies.
- The automated brain tumor segmentation models perform reliably even in limited data scenarios.
- Skull-stripping models include sellar/suprasellar regions, making them generalizable for a wider range of tumor histologies.
- Models can segment whole lesions and enhancing tumor regions, adapting to MRI sessions with missing sequences.
- Performance was evaluated using Dice scores, sensitivity, and Hausdorff distances.
Takeaway
This study created smart computer programs that help doctors see and measure brain tumors in kids better, even when they don't have all the pictures they need.
Methodology
The study used multi-institutional, multi-parametric MRI scans from pediatric patients to train deep learning models for skull-stripping and tumor segmentation.
Potential Biases
The study may be biased due to the exclusion of subjects with major neuroanatomical changes.
Limitations
Inaccuracies in model prediction can disproportionately affect performance metrics, especially for smaller tumor volumes.
Participant Demographics
The study included 527 pediatric patients with various brain tumor histologies.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website