Synthetic patches, real images: screening for centrosome aberrations in EM images of human cancer cells
This work addresses a bottleneck in cancer cell analysis by automating a manual step in centrosome aberration screening, though it is incremental as it builds on existing deep learning methods for a specific domain.
The authors tackled the problem of automating centriole detection in electron microscopy images of human cancer cells, which is necessary for selecting cells for higher-magnification re-imaging, and achieved high accuracy by training a two-level DenseNet on a hybrid dataset of synthetic patches and real images, outperforming image-level training on real patient data.
Recent advances in high-throughput electron microscopy imaging enable detailed study of centrosome aberrations in cancer cells. While the image acquisition in such pipelines is automated, manual detection of centrioles is still necessary to select cells for re-imaging at higher magnification. In this contribution we propose an algorithm which performs this step automatically and with high accuracy. From the image labels produced by human experts and a 3D model of a centriole we construct an additional training set with patch-level labels. A two-level DenseNet is trained on the hybrid training data with synthetic patches and real images, achieving much better results on real patient data than training only at the image-level. The code can be found at https://github.com/kreshuklab/centriole_detection.