Training Convolutional Neural Networks and Compressed Sensing End-to-End for Microscopy Cell Detection

Area of Science:

• Biomedical image analysis
• Computational biology
• Machine learning in medicine

Background:

• Accurate cell detection and localization are crucial for biomedical research and clinical diagnostics.
• Existing methods may face limitations in handling the complexity and scale of microscopy data.

Purpose of the Study:

• To develop and validate a novel, end-to-end trainable algorithm for automated cell detection and localization.
• To integrate deep convolutional neural networks (CNNs) with compressed sensing (CS) or sparse coding (SC) for enhanced performance.

Main Methods:

• A new algorithm combining CNNs and CS/SC for end-to-end training was designed.
• Cell detection was framed as a point object detection task, leveraging CS/SC for compact representation.
• A novel backpropagation rule was derived for training sparse code recovery layers.

Main Results:

• End-to-end training of the integrated pipeline demonstrated superior accuracy compared to separate training.
• The algorithm successfully detected and localized cells in microscopy images.
• Validation on five benchmark datasets yielded excellent results.

Conclusions:

• The proposed CNN-CS/SC algorithm offers a robust and accurate solution for automated cell detection and localization.
• End-to-end training is a key factor in improving the performance of such integrated pipelines.
• This approach has significant potential for advancing biomedical research and clinical applications.