AUTHOR=Liu Shuai , Wu Xiaomeng , He Shengji , Song Xiaowei , Shang Fei , Zhao Xihai 

TITLE=Identification of White Matter Lesions in Patients With Acute Ischemic Lesions Using U-net

JOURNAL=Frontiers in Neurology

VOLUME=Volume 11 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2020.01008

DOI=10.3389/fneur.2020.01008

ISSN=1664-2295

ABSTRACT=Background:White matter lesions (WML) have been proved to be significantly associated with many brain diseases. Precise evaluation of burden of WML at early stage could provide insights on the prognosis and assist in intervention. However, in the patients with both WML and acute ischemic lesions (AIL), AIL as well as WML, exhibited hyperintensities on FLAIR images. It is challenging to segment WML in such patients using traditional methods. Deep learning, a type of machine learning technique, has been validated as an efficient method for segmentation. This study aimed to evaluate the performance of U-net in identifying WML in the patients with AIL.
Methods:A total of 208 cases from Chinese Atherosclerosis Risk Evaluation (CARE II) were used in the present study. All subjects underwent FLAIR and DWI on a 3.0 Tesla scanner. The contours delineated by the observer and the scores offered by the observer were considered as gold standard. Among all 208 cases, 108 cases were randomly selected as training set, and the remaining 100 cases were used as testing set. The performance of lesion segmentation toolbox (LST) and three U-net models were evaluated on three levels: pixel, lesion and subject levels. The performance of all methods on WML identification and segmentation was also evaluated among the cases with different lesion volumes and between the cases with and without AIL.
Results:All U-net models outperformed LST on pixel, lesion and subject levels, while no differences were found among three U-net models. All segmentation methods performed best in the cases with WML volume (WMLV)>20 ml but worst in those with WMLV<5 ml. In addition, all methods showed similar performance between the cases with and without AIL. The scores calculated by U-net exhibited a strong correlation with the gold standard (all Spearman correlation coefficients>0.89, ICCs>0.88, p values<0.001). 
Conclusion: U-net performed well in identification and segmentation of WML in the patients with and without AIL on FLAIR and DWI images. The performance of U-net was validated by a dataset of multicenter study.  Our results indicate that U-net has an advantage in assessing burden of WML in the patients suffering from both WML and AIL.