Severe hemorrhagic transformation (HT) is an important complication of thrombolytic therapy. A method to identify stroke victims destined to severe HT could improve the patient selection and thus the safety of such treatment. In this study, we investigated whether very early serial diffusion weighted magnetic resonance imaging (DWI) could predict the occurrence of HT in an embolic model of experimental stroke. We tested the hypothesis that the ischemic brains with very low initial apparent diffusion coefficients (ADC) are destined to severe early (<or=5.5 h) HT. We retrospectively analyzed DWI scans of 45 New Zealand white rabbits subjected to thromboembolic stroke and treated with thrombolysis. DWI was obtained 0.5, 2, 3 and 5 h after embolization. Various thrombolytics were administered 1 h post embolization. The percentage of pixels within the ischemic hemisphere with ADC values below 550 x 10(-6) mm(2)/s was calculated and then compared to the severity of HT observed on gross brain sections at 5.5 h. As early as 30 min after embolization, ischemic brains destined to severe HT exhibited a significantly greater percentage of pixels below the cut-off value compared to those without HT: severe HT: 25%, 18.75-37.25% vs. no HT: 12%, 5.00-16.00% (median, 25th-75th %, P<0.001). Petechial HT when percentages were in the intermediate range. Quantitative analysis of initial ADC value might identify individual stroke patients at risk of severe HT.