Integrating CT Radiomics and Clinical Factors to Predict Hematoma Growth in HICH

Overview

This study developed and validated a hybrid machine learning model combining CT radiomics, clinical data, and imaging signs to predict hematoma expansion (HE) in hypertensive intracerebral hemorrhage (HICH). The model demonstrated improved predictive accuracy over conventional methods, offering a reliable tool for early identification of patients at risk of HE.

Background

Intracerebral hemorrhage (ICH) accounts for 10–15% of all strokes, with hypertensive ICH (HICH) being the most common subtype. Hematoma expansion (HE) occurs in 13–38% of cases and significantly worsens outcomes, increasing mortality and disability risk. Accurate early prediction of HE is critical for timely intervention. Traditional prediction relies on clinical expertise and imaging signs, but recent advances in radiomics and machine learning offer quantitative, objective approaches to improve prognostication.

Data Highlights

Key Findings

• HE was defined as >33% relative or >6 mL absolute hematoma volume increase within 24 hours.
• Radiomics features extracted from CT images significantly improved HE prediction accuracy compared to clinical and imaging signs alone.
• Previous clinical-imaging nomogram models achieved an AUC of 0.762, while radiomics-based models reached up to 0.892–0.960 in prior studies.
• The hybrid model integrating radiomics, clinical data, and imaging signs was constructed and visualized as a nomogram for intuitive clinical use.
• Data quality was ensured by excluding poor-quality images and using multiple imputation for missing values below 20%.

Clinical Implications

The integration of CT radiomics with clinical and conventional imaging factors enhances the early prediction of hematoma expansion in HICH patients. This approach facilitates timely clinical decision-making and targeted interventions to reduce hematoma growth, potentially improving patient outcomes. The nomogram tool provides an accessible method for clinicians to assess HE risk objectively.

Conclusion

Combining CT radiomics with clinical and imaging features yields a robust predictive model for hematoma expansion in hypertensive intracerebral hemorrhage. This integrated approach supports improved prognostication and personalized patient management.

References

1. Ma et al 2021 -- Radiomics for Predicting Hematoma Expansion in HICH
2. Song et al 2022 -- CT Radiomics and Machine Learning in HE Prediction
3. Previous Clinical Nomogram Study 2020 -- HE Prediction in HICH