Deep learning-enabled assessment of cardiac allograft rejection from endomyocardial biopsies
Jana Lipkova, Tiffany Y. Chen, Ming Y. Lu, Richard J. Chen, Maha Shady, Mane Williams, Jingwen Wang, Zahra Noor, Richard N. Mitchell, Mehmet Turan, Gulfize Coskun, Funda Yilmaz, Derya Demir, Deniz Nart, Kayhan Basa, Nesrin Turhan, Selvinaz Ozkara, Yara Banz, Katja E. Odening, and Faisal Mahmood*
ArXiv | GitHub
TL;DR: CRANE is a high-throughput, interpretable, multi-task framework that simultaneously address the main diagnostic tasks in endomyocardial biopsy assesment: acute cellular rejection,antibody-mediated rejection and quilty B lesions, as well as their concurrent appearances (e.g.cellular rejection with quilty lesions). For the detected rejection, the model estimates also the rejection grade.
Understanding Attention - This demo demonstrates the High attention regions corresponds to the regions used by the model to make classification determinations. In principle, CRANE models can make accurate predictions from a few patches. For visualization, attention scores are normalized for the entire slide.