Automated integrated system for stained neuron detection: An end-to-end framework with a high negative predictive...

Area of Science:

• Neuroscience
• Computational Biology
• Histology

Background:

• Brain architecture mapping is crucial for understanding neural computations influencing behavior.
• Traditional histology methods for neuron identification are manual, time-consuming, and labor-intensive.
• Automated systems are needed for high-throughput analysis of brain tissue.

Purpose of the Study:

• To develop an integrated hardware and software system for automated image acquisition, processing, and neuron detection in brain slices.
• To enable rapid, high-throughput analysis for comprehensive brain mapping.

Main Methods:

• An automated system was developed to detect retrogradely labeled neurons in monkey brain slices.
• Image pre-processing involved adaptive histogram equalization.
• Neuron candidates were segmented using marker-controlled watershed transformation (MCWT) with maximally stable extremal regions (MSERs).
• Deep transfer learning with pre-trained convolutional neural networks (CNNs) was used for neuron classification.

Main Results:

• The automated system achieved a 0.918 F1-score and an 86.6% negative prediction value.
• Precision, recall, and F-scores exceeded 90%, surpassing conventional methods.
• The MCWT algorithm and CNN-based classification demonstrated high performance compared to existing tools and classifiers.

Conclusions:

• A fully automated system for rapid image acquisition and neuron identification from stained brain slices was successfully demonstrated.
• The system's adaptability allows for the identification of stained features in various biological tissues.
• This technology significantly advances high-throughput tissue analysis for neuroscience research.