Abstract
The logical assessment formula (LAF) is a new theory proposed for evaluationswith inaccurate ground-truth labels (IAGTLs) to assess the predictive modelsfor artificial intelligence applications. However, the practicability of LAFfor evaluations with IAGTLs has not yet been validated in real-world practice.In this paper, we applied LAF to two tasks of tumour segmentation for breastcancer (TSfBC) in medical histopathology whole slide image analysis (MHWSIA)for evaluations with IAGTLs. Experimental results and analysis show that theLAF-based evaluations with IAGTLs were unable to confidently act like usualevaluations with accurate ground-truth labels on the one easier task of TSfBCwhile being able to reasonably act like usual evaluations with AGTLs on theother more difficult task of TSfBC. These results and analysis reflect thepotential of LAF applied to MHWSIA for evaluations with IAGTLs. This paperpresents the first practical validation of LAF for evaluations with IAGTLs in areal-world application.