Efficacy of Software-Based Analysis of Mouse Liver Samples to Assess Cell Death

Introduction Histopathological assessment of histological samples is a major challenge in terms of reliability and validity. Inter-rater differences in knowledge and training can vastly alter the scoring outcome of histological samples. The emergence of software utilizing neural networks in image analysis therefore offers great opportunity to approach this issue by offering software-based analysis methods aimed at improving histopathological analysis. We present a novel approach of delineating regions in mouse liver samples affected by cell death.

Material & Methods We introduced hepatic ischemia in a mouse model, followed by different times of reperfusion (0h, 20min, 1h, 2h, 6h, 24h and 48h). Liver tissue slides were stained with hematoxylin and eosin and digitalized. We then imported the digitalized slides into QuPath. In a first step, we used a neural-network-based pixel classifier to differentiate between areas affected by cell death and healthy areas within the sample. In a second step, we trained a second pixel classifier to further stratify the areas affected by cell death into three categories (early cell death, moderate cell death, extensive cell death).

Results Both the algorithm differentiating between healthy regions and regions experiencing cell death as well as the algorithm aimed at further stratifying the stages of cell death show high correlation with results obtained by human raters.

Conclusion Software-based histopathological image analysis using neural networks shows great promise in obtaining valid and reliable results for research. However, future research is necessary to make algorithms more robust to differences in staining and morphological abnormalities.

Publication History

Article published online:
20 January 2025