To use machine learning for early detection of malignant brain edema in patients with MCA ischemia

Malignant cerebral edema following large ischemic strokes account for up to 10% of all ischemic strokes. Mortality rates are high and most of the survivors are left severely disabled. Although decompressive craniectomy has been shown to significantly decrease mortality, high morbidity rates among survivors are reported. The optimal timepoint when neurosurgical decompression should be performed in the individual patient varies and is a subject of debate.

Early prediction of malignant brain edema to identify those patients who benefit from surgical treatment is a clinical challenge. The aim of this study is to use machine learning for comprehensive analysis of CT images as well as clinical data from 1500 patients with large ischemic MCA strokes in oder to develop a model for early prediction of malignant brain edema. In a first step algorithms automatically identify characteristic imaging features and clinical data of 1400 retrospective data sets to create a multistage model (learning phase). This is followed by a validation phase where the model is tested with 100 other retrospective data sets.

Inclusion Criteria:

• Acute ≥ subtotal MCA infarct (M1-M2 occlusion)
• with or without malignant brain swelling
• with or without reperfusion therapy
• with or without neurosurgical decompression
• with or without death following malignant brain edema

Exclusion Criteria:

• Non-acute MCA infarct
• < subtotal MCA infarct

1500 retrospective datasets will be collected from 5 large German stroke units. Data sets include imaging data and clinical data from patients with subtotal MCA infarcts (M1-M2 occlusion), with or without malignant brain swelling, with or without reperfusion therapy, with or without neurosurgical decompression, and with or without death following malignant brain edema. Data sets from patients who have died following malignant brain edema will be included. Each data set consists of initial NCCT, CTA, (DSA if available), and follow-up NCCT until 14 days after stroke onset as well as clinical data.