A deep learning model for prognosis prediction after intracranial hemorrhage

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Neurology

Background:

• Intracranial hemorrhage (ICH) is a critical condition requiring prompt diagnosis and treatment.
• Prognostic uncertainty in ICH management hinders treatment consensus.
• Existing prediction models primarily focus on ICH detection, not prognosis.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting intracranial hemorrhage prognosis.
• To assess the efficacy of a hybrid input model combining CT images and clinical data.
• To compare the hybrid model's performance against image-only and data-only models.

Main Methods:

• A custom deep learning model was trained on 262 ICH patient cases.
• The model utilized a hybrid input: brain CT images and clinical variables.
• Performance was evaluated against models trained solely on images (I-model) or data (D-model).

Main Results:

• The hybrid model achieved an AUC of 0.924 and an accuracy of 0.861.
• Image-only and data-only models showed lower performance (AUCs of 0.763 and 0.746, respectively).
• The hybrid model demonstrated superior accuracy in classifying patients into good and poor prognosis groups.

Conclusions:

• Deep learning, particularly with hybrid inputs, shows significant promise for ICH prognosis prediction.
• This novel model can accurately classify patient prognosis, potentially impacting clinical decisions.
• This represents the first deep learning model for ICH prognosis prediction, highlighting a new avenue in medical imaging AI.