Predicting hematoma expansion in ICH: from traditional signs to AI multimodal fusion

Review: Imaging prediction of hematoma expansion after ICH has progressed from CTA spot sign and NCCT markers to AI-driven radiomics and multimodal fusion, but interpretability and generalizability remain challenges.

• Narrative review covering traditional markers (spot sign, NCCT signs, clinical-imaging scores) and AI methods (radiomics, deep learning, multi-task learning).
• Discusses precision prediction via multimodal data fusion and subgroup analyses by hemorrhage location and onset time.
• Major limitation: most AI models lack external validation and interpretability, hindering clinical translation.

RadPigeon summaries are original and for information only. They are not clinical advice.