Human-Validated\nNeural\nNetworks\nfor Precise Amastigote Categorization and Quantification to Accelerate\nDrug Discovery in Leishmaniasis

Leishmaniases present a significant global health challenge\nwith\nlimited and often inadequate treatment options available. Traditional\nmicroscopic methods for detecting Leishmania amastigotes are time-consuming\nand error-prone, highlighting the need for automated approaches. This\nstudy aimed to implement and validate the YOLOv8 deep learning model\nfor real-time detection, quantification, and categorization of Leishmania\namastigotes to enhance drug screening assays. YOLOv8 was trained on\n470 images from two microscopes, classifying them into categories\nsuch as “infected cells,” “intracellular amastigotes,”\n“uninfected cells,” and “edge cells.”\nThe model’s performance was compared to human operators using\nPearson and Spearman correlation analyses. YOLOv8 achieved strong\nperformance in detecting “infected cells” (AUC = 0.934)\nand “intracellular amastigotes” (AUC = 0.846). However,\nchallenges remained in differentiating extracellular amastigotes from\nbackground noise (AUC = 0.672). Despite these challenges, the YOLOv8\nmodel effectively minimized human variability in drug screening, providing\na reliable and efficient tool for the quantification and categorization\nof Leishmania amastigotes in drug discovery efforts. While further\nrefinements are required to resolve misclassification issues, the\nmodel demonstrates significant potential in enhancing both accuracy\nand throughput in preclinical assays.