Volumetric measurement of cerebral white matter hyperintensities on fluid-attenuated inversion recovery (FLAIR) magnetic resonance images using artificial intelligence

This study aimed to develop a new artificial intelligence software that can automatically extract and measure the volume of white matter hyperintensities (WMHs) in head magnetic resonance (MR) images using only non-thin slice fluid-attenuated inversion recovery (FLAIR) sequences. We enrolled 1,092 subjects in Japan comprising this non-thin slice Private Dataset. Based on 207 randomly selected subjects, neuroradiologists annotated WMHs using predefined guidelines. The annotated subjects were divided into training (n = 138) and test (n = 69) datasets. The WMH segmentation model, consisting of a U-Net ensemble, was trained using the Private Dataset. For validation, two other models were trained using either both thin and non-thin slice MRI datasets or only the thin slice dataset. The voxel-wise Dice similarity coefficient (DSC) was used as the evaluation metric. The model trained using only non-thin slice MRI showed a DSC of 0.820 for the test dataset, which is comparable to the accuracy of human readers. The model trained using additionally the thin-slice dataset showed an only slightly improved DSC of 0.822. This automatic WMH segmentation model consisting of a U-Net ensemble trained on a non-thin slice FLAIR MRI dataset is a promising new method. Despite some limitations, this model may be applicable in clinical practice.