Related Experiment Video
Updated: Mar 6, 2026
Diffuse Optical Spectroscopy for the Quantitative Assessment of Acute Ionizing Radiation Induced Skin Toxicity Using a Mouse Model
Published on: May 27, 2016
Intelligent model for stratification of acute radiation syndrome severity in humans
Yuriy Dmitrievich Udalov1, Aleksey Sergeevich Umnikov1, Irina Alekseevna Galstyan1
1State Research Center - Burnasyan Federal Medical Biophysical Center of the Federal Medical-Biological Agency of Russia, Moscow, Russia.
Purpose:
The aim of this study was to develop and compare two intelligent model for stratifying the severity of acute radiation syndrome (ARS) in humans based on clinical, hematological, and radiobiological parameters.
Method:
The high inter-individual variability of clinical parameters and the infeasibility of conducting randomized studies in radiation medicine necessitated the use of machine learning approaches for model development. The study utilized retrospective data from 1519 patients affected by radiation incidents with outcome-compared clinical follow-up. To construct the stratification model, algorithms of linear discriminant analysis (LDA) and gradient boosting were employed, enabling multilevel classification based on 15 clinical and radiobiological indicators.
Results:
The gradient boosting-based model demonstrated the highest accuracy in determining ARS severity. The area under the ROC curves (AUC) reached 0.92, confirming the high predictive value of the proposed approach. To improve interpretability, SHapley Additive exPlanations (SHAP) were applied, identifying the most informative features: cytogenetically estimated radiation dose, exposure duration, markers of inflammatory response, and indicators of radionuclide incorporation.
Conclusions:
The developed model is adapted for clinical use and can be implemented as a standalone software tool under resource-limited conditions. Its deployment enables improved medical triage accuracy and reduces the risk of decision-making errors in emergency situations associated with human radiation exposure.
