A hybrid framework for effective microscopic cell counting segmentation integrating Light-U-net with watershed.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma causes irreversible blindness due to retinal ganglion cell (RGC) degeneration.
• Early RGC detection is crucial for glaucoma management.
• Current RGC monitoring methods are inefficient and error-prone.

Purpose of the Study:

• To develop a lightweight and scalable deep learning model for automated RGC segmentation and counting.
• To improve the accuracy and efficiency of RGC analysis for glaucoma detection.

Main Methods:

• Proposed Light-U-Net, a deep learning model for RGC segmentation in retinal images.
• Trained and tested the model on synthetic and real retinal datasets.
• Introduced a local maxima algorithm for RGC counting using generated annotations.
• Compared Light-U-Net with the watershed algorithm for segmentation and counting.

Main Results:

• Light-U-Net demonstrated superior performance in segmenting and counting RGCs.
• The combination of Light-U-Net and the watershed algorithm yielded optimal results.
• Automated RGC analysis showed improved accuracy and efficiency over traditional methods.

Conclusions:

• Light-U-Net offers a promising automated solution for RGC segmentation and counting.
• This technology can significantly reduce errors and enhance efficiency in glaucoma diagnosis.
• Light-U-Net facilitates effective glaucoma progression tracking and treatment assessment.