Fully Convolutional DenseNets for Segmentation of Microvessels in Two-photon Microscopy.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Deep Learning

Background:

• Microvessel segmentation in two-photon microscopy is challenging due to optical imaging artifacts like uneven intensities and shadowing.
• Existing methods have shown limited success in accurately segmenting microvasculature from these complex images.

Purpose of the Study:

• To develop and evaluate a customized fully convolutional neural network (FC-DenseNet) for enhanced microvessel segmentation.
• To address the limitations of current segmentation techniques in two-photon microscopy.
• To compare the performance of the proposed method against a state-of-the-art deep learning approach.

Main Methods:

• A customized version of the FC-DenseNet architecture was employed.
• The network was trained and validated using manual annotations from 8 two-photon microscopy angiograms.
• Segmentation results were benchmarked against a contemporary deep learning-based segmentation scheme.

Main Results:

• The customized FC-DenseNet demonstrated improved performance in microvessel segmentation.
• The method provided accurate and end-to-end segmentation, outperforming the comparative state-of-the-art scheme.
• Successfully mitigated issues related to uneven intensities and shadowing in image data.

Conclusions:

• The developed FC-DenseNet offers a robust solution for accurate microvessel segmentation in two-photon microscopy.
• This approach advances the analysis of microvasculature in complex biological imaging scenarios.
• Highlights the potential of customized deep learning models for overcoming specific challenges in scientific image analysis.