Background and Aims
    Automatic determination of hemoglobin species inside intracerebral hemorrhage in emergency radiology may improve efficiency and speed of radiologist diagnosis. Characterization of hemorrhage using quantification MRI and modern methods of machine learning, such as gradient boosting with categorical features (CatBoost) support can be helpful.

Method
     Laboratory constructed phantom was used for modeling of different hemoglobin species and statements of erythrocyte membrane. T1, T2, T2* relaxation time and apparent diffusion coefficient were measured. Spectrophotometry was used to control of hemoglobin species and erythrocytes membrane status. Additionally, hemorrhage stroke was modeled in rat with subsequent dynamic MRI. examination At the first stage, we trained model on experimental data. After that on rat with intracerebral hemorrhage the form of hemoglobin and condition of clot were determined.

Results
     Obtained data suggests that methemoglobin appears from the first hour and sequentially increased up to day 21, the blood clot rapidly lyse between day 8-10, while erythrocyte lysis is not clearly observed. The presence of oxyhemoglobin for up to 21 days in raw data can be explained by its diamagnetic nature and poor classification against the background tissue.

Conclusion
     Using of machine learning in assessment of hemoglobin species may help to understand some aspects of hemoglobin biodegradation and may improve diagnostic of intracerebral hemorrhage.

GRADIENT BOOSTING IN AUTOMATIC DETERMINATION OF HEMOGLOBIN SPECIMEN IN INTRACEREBRAL HEMORRHAGE