New Criteria for Histopathological Classification of Testis Based on Johnsen Score for Male Infertility Using Automated Deep Learning Software

Abstract

We examined whether a tool for determining Johnsen scores automatically using arti cial intelligence (AI) could be used in place of traditional Johnsen scoring to support pathologists' evaluations. Average precision, precision, and recall assessed by the Google Cloud AutoML vision platform. We obtained testicular tissues for the 275 patients and were able to make 264 haematoxylin and eosin (H&E)-stained glass microscope slides. In addition, we cut out of parts of the histopathology images (5.0 X 5.0 cm) for expansion of Johnsen's characteristic areas with seminiferous tubules. We de ned four labels: Johnsen score 1-3, 4-5, 6-7, and 8-10 to distinguish Johnsen scores in clinical practice. All images were uploaded to the Google Cloud AutoML vision platform. We obtained a dataset of 7155 images at magni cation X400 and a dataset of 9822 expansion images for the 5.0 X 5.0 cm cutouts. For the X400 magni cation image dataset, the average precision (positive predictive value) of the algorithm was 82.6%, precision was 80.31%, and recall was 60.96%. For the expansion image dataset (5.0 X 5.0 cm), the average precision of the algorithm was 99.5%, precision was 96.29%, and recall was 96.23%. This is the rst report of an AI-based algorithm for predicting Johnsen scores.