Machine Learning-Supported Analyses Improve Quantitative Histological Assessments of Amyloid-β Deposits and Activated Microglia

Abstract

Background: Detailed pathology analysis and morphological quantification is tedious and prone to errors. Automatic image analysis can help to increase objectivity and reduce time. Here, we present the evaluation of the DeePathology STUDIO™ for automatic analysis of histological whole-slide images using machine learning/artificial intelligence. Objective: To evaluate and validate the use of DeePathology STUDIO for the analysis of histological slides at high resolution. Methods: We compared the DeePathology STUDIO and our current standard method using macros in AxioVision for the analysis of amyloid-β (Aβ) plaques and microglia in APP-transgenic mice at different ages. We analyzed density variables and total time invested with each approach. In addition, we correlated Aβ concentration in brain tissue measured by ELISA with the results of Aβ staining analysis. Results: DeePathology STUDIO showed a significant decrease of the time for establishing new analyses and the total analysis time by up to 90%. On the other hand, both approaches showed similar quantitative results in plaque and activated microglia density in the different experimental groups. DeePathology STUDIO showed higher sensitivity and accuracy for small-sized plaques. In addition, DeePathology STUDIO allowed the classification of plaques in diffuse- and dense-packed, which was not possible with our traditional analysis. Conclusion: DeePathology STUDIO substantially reduced the effort needed for a new analysis showing comparable quantitative results to the traditional approach. In addition, it allowed including different objects (categories) or cell types in a single analysis, which is not possible with conventional methods.