Automated MRI pipeline with U-Net accurately stages intracerebral hemorrhage

Automated MRI pipeline using U-Net segmentation and Random Forest classified intracerebral hemorrhage temporal stages with an accuracy of 0.807 in an independent validation cohort of 114 patients.

• In an external validation set of 115 patients, one case was excluded due to automated segmentation failure.
• Hematoma segmentation achieved Dice 0.794, sensitivity 0.799, and precision 0.842.
• The Random Forest classifier was selected after comparison with XGBoost and multinomial logistic regression.

Automated summary

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