Virtual Diffusion Weighted Imaging for Assessing Acute Ischemic Stroke from Computed Tomography Angiography Via Deep Learning

Background: Diffusion-weighted imaging (DWI) form of magnetic resonance imaging (MRI) plays a crucial role in the diagnosis and assessment of acute ischaemic stroke (AIS). However, its usefulness is limited by low accessibility worldwide. Computed tomography angiography (CTA) imaging is more commonly used, but its subtle changes of hypodensity from infarcted brain tissue have resulted in large inter-rater variations in clinical practice. Methods: A generative adversarial network (GAN) model that achieved the imaging modality transformation from CTA to the virtual DWI (VDWI) was constructed and evaluated using datasets of 167 patients with AIS. The similarity and clinical applicability of this model and VDWI images were evaluated by comparing the peak-signal-to-noise-ratio (PSNR) and mean absolute error (MAE), and annotations of the infarct area on the images by three radiologists. Findings: 3D VDWI images showed high quality and similarity compared to real DWI images as indicated by PSNR 66·55 ±3·12 and MAE 6·20±2·97 in the testing dataset. Inter-rater correlation and agreement with DWI- Alberta Stroke Program Early Computed Tomography Scores (DWI-ASPECTS) were 0·85 and 0·73 for CTA-ASPECTS and increased to 0·90 and 0·81 for VDWI-ASPECTS. No significant differences in volume and location were found between DWI and VDWI images. Interpretation: The results show higher correlation between images generate by VDWI than CTA images. By improving diagnostic accuracy and agreements, 3D VDWI generative adversarial network (VDWI-GAN) demonstrated potential improvement in clinical assessment of AIS. Funding: This research was supported by the National Natural Science Foundation of China (Grant No. 81871329, 81671673), the New Developing and Frontier Technologies of Shanghai Shen Kang Hospital Development Center (Grant No. SHDC12018117), the Shanghai Municipal Education Commission-Gaofeng Clinical Medicine Grant Support (Grant No. 2016427), Science and Technology Commission of Shanghai Municipality (19411968500). Declaration of Interest: None to declare Ethical Approval: The study protocol was approved by the Institutional Review Board of Shanghai Jiao Tong University Affiliated Sixth People’s Hospital (No.2020:212).