Machine Learning Models for 3-Month Outcome Prediction Using Radiomics of Intracerebral Hemorrhage and Perihematomal Edema from Admission Head Computed Tomography (CT)

2024

Machine Learning Models for Predicting Outcomes in Intracerebral Hemorrhage

Sample size: 852 publication 10 minutes Evidence: moderate

Author Information

Author(s): Dierksen Fiona, Sommer Jakob K., Tran Anh T., Lin Huang, Haider Stefan P., Maier Ilko L., Aneja Sanjay, Sanelli Pina C., Malhotra Ajay, Qureshi Adnan I., Claassen Jan, Park Soojin, Murthy Santosh B., Falcone Guido J., Sheth Kevin N., Payabvash Seyedmehdi

Primary Institution: Yale School of Medicine

Hypothesis

Combining radiomic features from intracerebral hemorrhage and perihematomal edema will improve outcome prediction.

Conclusion

The addition of perihematomal edema radiomics can enhance individual-level risk assessment in intracerebral hemorrhage patients, although it does not significantly improve overall prognostic accuracy.

Supporting Evidence

Machine learning models using clinical variables are more accurate than the widely used ICH score.
Adding PHE radiomics improves individual-level risk assessment metrics.
Models achieved prediction accuracy equal to or higher than the ICH score.

Takeaway

This study shows that using special computer models can help doctors predict how well patients with brain bleeding will do after three months, especially when they look at both the bleeding and the swelling around it.

Methodology

The study used a multicentric trial cohort of 852 patients, extracting radiomic features from CT scans and applying machine learning models to predict outcomes.