Evaluation of Deep Learning Topcoders Method for Neuron Individualization in Histological Macaque Brain Section

Area of Science:

• Digital Pathology
• Computational Neuroscience
• Artificial Intelligence

Background:

• Cell individualization is crucial for digital pathology image analysis.
• Deep Learning (DL) excels at instance segmentation but requires extensive manual annotation.
• Existing DL applications predominantly focus on oncological data.

Purpose of the Study:

• To develop a pipeline for synthesizing pixel-level labels from point annotations.
• To apply an ensemble DL algorithm for cell individualization in neurological data.
• To address the limitations of manual annotation and data bias in DL for pathology.

Main Methods:

• A novel pipeline was established to generate pixel-level labels using only point annotations.
• An ensemble Deep Learning algorithm was employed for cell individualization.
• The methodology was specifically tested on neurological imaging datasets.

Main Results:

• The proposed method successfully achieved cell individualization on neurological data.
• Accurate segmentation of neuronal cells was demonstrated at both object and pixel levels.
• An average detection accuracy of 0.93 was achieved, indicating high performance.

Conclusions:

• The developed pipeline effectively synthesizes pixel-level labels, reducing annotation labor.
• The ensemble DL approach demonstrates strong performance for neuronal cell segmentation in digital pathology.
• This work expands the application of DL in digital pathology to neurological data.