Abstract
Pathological remodeling of the extracellular matrix is a hallmark of cardiovascular disease. Abnormal fibrosis causes cardiac dysfunction by reducing ejection fraction and impairing electrical conductance, leading to arrhythmias. Hence, accurate quantification of fibrosis deposition in histological sections is of extreme importance for preclinical and clinical studies. Current automatic tools do not perform well under variant conditions. Moreover, users do not have the option to evaluate data from staining methods of their choice according to their purpose. To overcome these challenges, we underline a novel machine learning-based tool (FibroSoft) and we show its feasibility in a model of cardiac hypertrophy and heart failure in mice. Our results demonstrate that FibroSoft can identify fibrosis in diseased myocardium and the obtained results are user-independent. In addition, the results acquired using our software strongly correlate to those obtained by Western blot analysis of collagen 1 expression. Additionally, we could show that this method can be used for Masson's Trichrome and Picosirius Red stained histological images. The evaluation of our method also indicates that it can be used for any particular histology segmentation and quantification. In conclusion, our approach provides a powerful example of the feasibility of machine learning strategies to enable automatic analysis of histological images.