@inproceedings{554,
  author = {Y. Li},
  title = {SRNet: A spatial-relationship aware point-set classification method for multiplexed pathology images.},
  abstract = {Point-set classification for multiplexed pathology images aims to distinguish between the spatial configurations of cells within multiplexed immuno-fluorescence (mIF) images of different diseases. This problem is important towards aiding pathologists in diag- nosing diseases (e.g., chronic pancreatitis and pancreatic ductal adenocarcinoma). This problem is challenging because crucial spa- tial relationships are implicit in point sets and the non-uniform distribution of points makes the relationships complex. Manual morphologic or cell-count based methods, the conventional clinical approach for studying spatial patterns within mIF images, is limited by inter-observer variability. The current deep neural network methods for point sets (e.g., PointNet) are limited in learning the representation of implicit spatial relationships between categorical points. To overcome the limitation, we propose a new deep neural network (DNN) architecture, namely spatial-relationship aware neural networks (SRNet), with a novel design of representation learning layers. Experimental results with a University of Michigan mIF dataset show that the proposed method significantly outperforms the competing DNN methods, by 80%, reaching 95% accuracy.},
  year = {2021},
  journal = {In Proceedings of DeepSpatial’21: 2nd ACM SIGKDD Workshop on Deep Learning for Spatiotemporal Data, Applications, and Systems},
  volume = {10},
  month = {07},
  url = {https://par.nsf.gov/biblio/10350530},
}