A CNN transfer learning‐based approach for segmentation and classification of brain stroke from noncontrast CT images

Abstract

Imaging is needed in stroke cases in order to understand what the type of stroke (ischemic, hemorrhagic) is, to rule out bleeding, to determine the infarct area and to plan treatment. Noncontrast CT is the primary imaging protocol used in the initial evaluation of patients with suspected stroke. As apart from studies in the literature, this paper proposes novel automated classification and segmentation approaches which are capable of extracting hemorrhage and ischemic lesions (infarcts) simultaneously from the noncontrasts brain CT images during the treatment of brain stroke patients. It is aimed to automate the detection of stroke lesions with a high accuracy rate using the U‐Net model for segmentation. In the experiments performed on the real data set, a precision value of 95.06% is obtained for the classification model. For segmentation, the IoU coefficient values from the experiments are 92.01% for hemorrhagic and 82.22% for ischemic, respectively.

Fusing feature and output space for unsupervised domain adaptation on medical image segmentation

“…Deep neural networks (DNN) [1][2][3] have achieved excellent results in a variety of computer vision tasks. 4 However, models trained on the training data (source domain) cannot perform well on the testing data (target domain) when the data are drawn from different distributions.…”