Utilization of a Deep Learning Algorithm for Microscope-Based Fatty Vacuole Quantification in a Fatty Liver Model in Mice

Abstract

<jats:p> Quantification of fatty vacuoles in the liver, with differentiation from lumina of liver blood vessels and bile ducts, is an example where the traditional semiquantitative pathology assessment can be enhanced with artificial intelligence (AI) algorithms. Using glass slides of mice liver as a model for nonalcoholic fatty liver disease, a deep learning AI algorithm was developed. This algorithm uses a segmentation framework for vacuole quantification and can be deployed to analyze live histopathology fields during the microscope-based pathology assessment. We compared the manual semiquantitative microscope-based assessment with the quantitative output of the deep learning algorithm. The deep learning algorithm was able to recognize and quantify the percent of fatty vacuoles, exhibiting a strong and significant correlation ( r = 0.87, P &lt; .001) between the semiquantitative and quantitative assessment methods. The use of deep learning algorithms for difficult quantifications within the microscope-based pathology assessment can help improve outputs of toxicologic pathology workflows. </jats:p>

Artificial Intelligence in Toxicologic Pathology: Quantitative Evaluation of Compound-Induced Hepatocellular Hypertrophy in Rats

“…Our results add to the growing body of papers on AI-based approaches connected to the field of toxicologic pathology. 11,29 –33…”

Microscope-Based Automated Quantification of Liver Fibrosis in Mice Using a Deep Learning Algorithm

“…We have previously published a report on the use of a DL algorithm for quantification of fatty vacuoles in a fatty liver mouse model. 42 We found an excellent correlation between the manual semiquantitative evaluation and the quantitative evaluation of hepatic fatty vacuoles by the DIA. We believe that these reports and additional ones in the future can be of great benefit to ensure applicability of such novel tools in routine laboratory workflows.…”

“…To further expand the applicability of DIA in preclinical models for liver fibrosis, we used data from a study that evaluated liver fibrosis in mice using the carbon tetrachloride (CCl 4 ) model, 40,41 and we compared the traditional semiquantitative evaluation by an experienced toxicologic pathologist with a novel artificial intelligence (AI) application for identification and quantification of liver fibrosis. 42 Furthermore, this analysis was done by deploying the application into the pathologist's microscope, therefore facilitating its use, without disrupting the regular reporting workflow. [42][43][44] Nine-week-old female BALB/c mice were used in this study.…”

“…42 Furthermore, this analysis was done by deploying the application into the pathologist's microscope, therefore facilitating its use, without disrupting the regular reporting workflow. [42][43][44] Nine-week-old female BALB/c mice were used in this study. The animals were maintained in the animal facility of the Molecular Biotechnology Center of the University of Torino under a 12-hour light-dark cycle with ad libitum access to food and water.…”

Machine learning algorithm outperforms fibrosis markers in predicting significant fibrosis in biopsy‐confirmed NAFLD

“…In other machine learning studies, subjects were either from animal models or non-liver biopsy patients with NAFLD, or the sample size of liver biopsy patients with NAFLD of previous studies was smaller than that of our study and the results lacked validation. [26][27][28][29] In addition, previous MLA studies have mostly focused on NASH, while few have focused on liver fibrosis. It should be noted that the degree of fibrosis is the most important risk factor of the prognosis of NAFLD patients.…”