The slice level predictions from RADnet were aggregated to infer CT level predictions. A 3D CT scan was declared positive if the deep learning model predicted hemorrhage in consecutive three or more slices. This corresponds to minimum cross-sectional dimension of approximately 1.5 mm for hemorrhagic region. Further, we calculated accuracy, recall, precision, and F1 score of the CT level predictions by RADnet and compared with the annotations from individual radiologists.