Acute ischemic stroke lesion is easily represented as hyperintense areas on Magnetic Resonance Imaging ( MRI) . Compared toMRI,Computed Tomography ( CT)?is faster,less expensive,and less susceptible to interference from metal implants. However,CT is notsensitive to acute ischemic stroke lesion and contains less information than MRI. Usually,it is difficult to determine the location of lesionson CT. Considering the improvement of speed and usability and the reduction of cost,
in order to replace real MRI with MRI generatedfrom CT for the diagnosis of acute ischemic stroke,a cross- modal image generation algorithm from CT to MRI is proposed. Firstly,radiomics is used to determine the lesion area on CT and extract radiomics features,and the feature with the largest information gain isscreened?
out and visualized,and then the feature map is used together with the real CT as the input of the generative adversarial network.The generated adversarial network introduces residual block in pix2pix generator,and the discriminator adopts PatchGAN. Finally,thelesion feature similarity loss function is introduced in the loss function,paying more attention to the similarity of the lesion area. After thesubjective judgment of two radiologists and the objective analysis of evaluation indicators,
the results show that the MRI generated by thealgorithm is very similar to the real MRI,and the location of the lesion is correct and the shape is similar,which can provide assistance fordoctors’ diagnosis and treatment.