In the beginning of may the Botkin.AI team published an article “Deep Learning on Point Clouds for False Positive Reduction at Nodule Detection in Chest CT Scans” (Drokin I., Ericheva E.). The article suggests a brand-new method of false positive reduction in chest CT scans studies. This method transforms 3D CT scan data to a point cloud. Thus, it allows us to detect nodules on images much more precisely. According to this research, the brand-new method outperforms previous widely-used approaches. Analysis of CT scans as point clouds makes it possible to reduce the cost of both model training and the method inference as a more computationally efficient by radiologists. Meanwhile such approach demonstrates State-of-the-art on a LIDC-IDRI data set.
Full pipeline of data processing as it is shown in the article
To learn more about the research you may have a look the article published on Arxiv.org.