Machine learning-based model for predicting outcomes in cerebral hemorrhage patients with leukemia
Lu Shi1, Ping Yin1, Cancan Chen2
1Department of Radiology, Peking University People's Hospital, 11 Xizhimen Nandajie, Xicheng District, Beijing 100044, China.
European Journal of Radiology
|June 21, 2024
Summary
This study developed combined models using non-contrast computed tomography (NCCT) imaging to predict mortality in leukemia patients with intracranial hemorrhage (ICH). The models accurately forecast 7-day and short-term survival, highlighting the importance of radiomic shape features.
Area of Science:
- Radiology
- Oncology
- Medical Imaging
Background:
- Intracranial hemorrhage (ICH) in leukemia patients is associated with rapid progression and high mortality.
- Limited data exist on imaging predictors for ICH outcomes in this specific patient population.
Purpose of the Study:
- To develop and evaluate prediction models for 7-day and short-term mortality risk in leukemia patients with ICH.
- To utilize non-contrast computed tomography (NCCT) image features for prognostic modeling.
Main Methods:
- Retrospective analysis of NCCT images from 135 leukemia patients with ICH.
- Extraction of image features using manual assessment and radiomics.
- Development of prediction models (clinical, radiomics, combined) using logistic regression and random forest algorithms after data imputation.
Main Results:
- Combined models demonstrated good predictive efficacy for 7-day (AUC=0.84, AUPRC=0.83) and short-term mortality (AUC=0.87, AUPRC=0.89).
- Clinical decision curve analysis indicated superior benefit of combined models for predicting 7-day and 30-day mortality risk.
- Radiomic shape features were significant contributors to the predictive power of the models.
Conclusions:
- Combined models integrating NCCT features effectively predict 7-day and short-term mortality in leukemia patients with ICH.
- Radiomic analysis, particularly shape features, provides crucial prognostic markers for ICH in this cohort.
