Evaluation of a Novel Automated Method for the Quantification of Inflammation in Renal Biopsies

The development of robust software-based technologies for objective image interpretation has facilitated the integration of Artificial Intelligence (AI) tools into digital pathology workflows for diagnostic applications. In this work, a novel automated method for quantifying inflammation in renal tissue biopsies is presented. To this purpose, the open-source QuPath software is exploited to design a custom classifier specifically trained to handle images stained with a novel combination of Periodic Acid-Schiff (PAS) and Leukocyte Common Antigen (LCA) techniques. This dual-staining approach enhances the visualization of both extracellular structures and inflammatory cells, enabling a more accurate assessment of the inflammation status. The performance of the presented quantification method is compared against measurements performed by two expert professionals, also referred to as observers, employing statistical metrics including the correlation coefficient and the normalized Mean Squared Error (MSE). For a thorough inspection, a Passing-Bablok regression analysis is also performed, showing a good agreement between outcomes obtained from the proposed method and one of the two observers. Discrepancies between the measurements performed by the two observers are also highlighted.