Advancements in Artificial Intelligence (AI) has paved way for solving the problems in different domains. In healthcare industry, brain stroke incidence is in alarming rate across the globe. According to WHO, it is one of the leading causes of disability and depth which needs sustainable research effort. It is believed that bringing AI into Clinical Decision Support System (CDSS) has potential to have unprecedented strides in healthcare units for improving Quality of Service (QoS). The existing MRI scans based research with deep learning has limitations as it uses pre-defined CNN based models. The CNN models such as VGG16, ResNet50 and DenseNet121 have their built in functionalities. However, for a given problem, their performance may not be adequate and they are to be optimized to meet requirements of the problem in hand. To overcome this problem, in this paper, we proposed a framework and underlying mechanisms to optimize and exploit the deep CNN models aforementioned for efficient brain stroke detection. Apart from the improving the models, an algorithm named Optimized Deep Learning for Brain Stroke Detection (ODL-BSD) is proposed. MRI scans are used to evaluate the proposed framework. The empirical results revealed that the proposed methodology has caused improvement in the performance of the deep CNN models significantly.