Abstract: In this study, we deal with the question of diagnosing brain bleeding that radiologists consider a laborious procedure, particularly in the early stages of bleeding. The problem is solved by means of a deep learning approach where the CNN, the famous AlexNet neural network as well as a newly modified version of AlexNet with a support vector machine (AlexNet-SVM) classification classifier has been developed to classify brain computer (CT) pictures into haemorrhaetic and non-hemorrhaging pictures. In medical image analytics and classification the objective of using the model of deep learning is to answer the key problem: can the requirement to develop CNN be removed by a suitable smoothing of a pre-trained model (transfer learning)? In addition, this study will examine the benefits of employing SVM as a classifier rather than a threelayer neural network. The one we have built in three deep networks is a pretrained model, which is tailor-made to the brain classification task CT, and our modified AlexNet model using the SVM class. We are working on the same classification problem in three deep networks. The three networks have been trained with the same number of CT brain pictures. Experiments have demonstrated that information may be transferred from natural pictures to medical visuals. Moreover, our findings have shown that the proposed modified “AlexNet-SVM” model may be used to identify the brain bleeding through an overall neural network developed from scratch, and by the original AlexNet.