| Summary: | Scalable image data analysis is widely demanded in biomedical
diagnosis by leveraging rapidly developed optical technology
and advanced machine learning algorithm. However, bio-image
obtained for single molecular or cell always have additive and
multiplicative noise and requires denoising with better resolution
in diagnosis. This dissertation proposed a high-throughput bioimage
denoising method for different kinds of threedimensional
microscopy cell images. Using a convolutional encoderdecoder
network, one can provide a scalable bio-image platform,
called NucleiNet, to automatically segment, classify and track cell
nuclei. Using a benchmark of 2480 nuclei images, the experiment
results show that the network achieves a 0.98 F-score and 0.99
pixel-wise accuracy, which means that over 95% of nuclei were
successfully detected with no merging nuclei found.
Key words: Image denoising, Convolutional neural network, Machine
learning, Bio-image processing
|
|---|