Root Mean Square Minimum Distance as a Quality Metric for Stochastic Optical Localization Nanoscopy Images

摘要

A localization algorithm in optical localization nanoscopy plays an importantrole in obtaining a high-quality image. A universal and objective metric, whichis crucial and necessary to evaluate qualities of nanoscopy images andperformances of localization algorithms, has not yet been established. Themetrics of root mean square error (RMSE or accuracy), precision, recall, andJaccard index (JAC) are currently used in the literature but in certainconditions fail in distinguishing qualities of different nanoscopy images. Inthis paper, the root mean square minimum distance (RMSMD) is proposed as aquality metric for localization nanoscopy images. Its properties are analyzedand demonstrated by examples which also show the advantages over RMSE,precision, recall and JAC. As a universal and objective metric, the RMSMD canbe broadly applied to applications that need to measure average, local, andmutual fitness of two sets of points.