Abstract 4147122: Automated quantification of amyloid burden in cardiac amyloidosis

Background: Cardiac amyloidosis (CA) involves the accumulation of amyloid fibrils in the extracellular cardiac spaces, increasing the extracellular volume (ECV) and correlating with poorer outcomes. Accurate quantification of cardiac amyloid is prognostically valuable but challenging with current manual methods using Congo-Red (CR)-stained tissue. We developed a digital pathology pipeline for automated quantification of cardiac amyloid on CR-stained endomyocardial biopsies (EMBs). Methods: We analyzed 20 consecutive left ventricular EMBs with CA (AL, n=12; ATTR, n=8) and 10 EMBs with cardiac hypertrophy (CH) and CA excluded. Amyloid was semi-quantitatively assessed as focal, multi-focal or diffuse by a cardiac pathologist. Polarized light images were converted into negative images using MATLAB to obtain bright field images, normalized using modified Reinhard normalization, and imported into QPath software. Images underwent color deconvolution to isolate the negative of the apple-green color. An intensity threshold on the amyloid channel detected positive pixels. Total tissue area was calculated by intensity threshold of the bright field image. Results: Standardized staining and image acquisition using automated machines and fully closed polarizer filters resulted in consistent histological staining. Residual staining variabilities were removed by applying a modified Reinhard normalization, equalizing the median of the image channels in the L*a*b color space. Conversion of dark field polarized images to negative allowed for the use of standard color deconvolution algorithms to isolate the negative apple-green channel and to use thresholding to quantify positive pixel. Amyloid quantification showed significantly higher amyloid percentages in positive controls (mean 6.21%) compared to negative controls (mean 2.29%) with a 0.05 threshold. The method was applied to six independent acquisition of the same section to assess reproducibility. The variation coefficient of the percentage of amyloid-positive pixels in the replicates was 11.6%, confirming method reproducibility. Conclusions: A digital pathology pipeline quantifying amyloid birefringence areas and, in most cases, differentiated it from collagen birefringence. Automated amyloid quantification simplifies EMB sample analysis and provides valuable information for risk prediction.