Related Experiment Video
Updated: Aug 29, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Identifying clinically applicable machine learning algorithms for glioma segmentation: recent advances and
Niklas Tillmanns1,2, Avery E Lum1, Gabriel Cassinelli1
1Brain Tumor Research Group, Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, Connecticut, USA.
Neuro-Oncology Advances
|September 8, 2022
Summary
Machine learning (ML) algorithms for glioma segmentation face systemic limitations, hindering clinical translation. Research often lacks reproducibility and generalizability due to dataset biases, preventing FDA clearance.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuro-oncology
Background:
- Numerous machine learning (ML) algorithms exist for glioma segmentation.
- No ML-based glioma segmentation product has yet received US FDA clearance.
- Systemic limitations in research algorithms impede clinical translation.
Purpose of the Study:
- To explore systemic limitations preventing ML glioma segmentation algorithms from becoming FDA-cleared products.
- To review current research literature for insights into translational barriers.
Main Methods:
- Systematic literature review across 4 databases.
- Inclusion of 58 articles meeting specific criteria from 11,727 initial articles.
- Data extraction and screening using TRIPOD guidelines.
Main Results:
- Many ML-based glioma segmentation studies report high accuracy.
- Substantial limitations in methodology and results reporting hinder reproducibility.
- Difficulty in translating research findings into clinical practice was observed.
Conclusions:
- Over a third of reviewed articles utilized common public datasets (BRaTS, TCIA).
- Over-reliance on shared datasets limits generalizability and increases risk of overfitting and bias.
- These factors contribute to the lack of FDA-cleared ML products for glioma segmentation.
