Among the various brain diseases, ischemic stroke is considered to be a major cause behind death and disability in the developed countries. The segmentation of lesion serves to perceive the level of the influenced tissues for effective diagnosis and treatment. At present, manual delineation of lesion structures is the standard method to localize its presence. However, it is operator-dependent and time-consuming. Hence, developing a fully-automatic approach to detect the abnormal structures from brain images is considered to be a challenging issue in medical image analysis. In this research, we utilize the discrete curvelet transformation to extract features on multiple scales and show its relevance for detection of ischemic stroke. An extensive investigation is done on the multidirectional statistical features obtained for different datasets. This resulted in a significant pattern in the scales of higher dimension. These features are trained using Support Vector Machine (SVM) with radial basis function kernel and the developed classifier is able to discriminate significantly between 20 patients affected from ischemic stroke and 25 healthy controls. The accuracy of the proposed approach was 99.1%. © 2017 Elsevier Ltd