Neuronal Population Reconstruction From Ultra-Scale Optical Microscopy Images via Progressive Learning

Area of Science:

• Neuroscience
• Computational Biology
• Image Analysis

Background:

• Reconstructing neuronal populations from ultra-scale optical microscopy (OM) images is crucial for understanding brain circuits.
• Challenges include noise, low contrast, high memory, and computational costs, often requiring expensive manual annotations for deep neural network (DNN) training.

Purpose of the Study:

• To develop a novel framework for dense neuronal population reconstruction from ultra-scale images.
• To eliminate the need for costly manual annotations in training DNNs for neuronal reconstruction.

Main Methods:

• Proposed a progressive learning scheme for neuronal population reconstruction (PLNPR) combining traditional neuron tracing and deep segmentation networks.
• Introduced an automatic framework for adaptive block-wise tracing and smooth fusion of neurites in terabyte-sized images.
• Developed the VISoR-40 dataset comprising 40 large-scale OM image blocks from mouse cortical regions.

Main Results:

• Demonstrated the effectiveness and superiority of the PLNPR method on the VISoR-40 and public BigNeuron datasets for both dense and single neuron reconstruction.
• Successfully applied the method to reconstruct dense neuronal populations from an ultra-scale mouse brain slice.
• Adaptive block propagation and fusion strategies significantly improved neurite completeness in dense reconstruction.

Conclusions:

• The proposed framework enables efficient and accurate dense neuronal population reconstruction from ultra-scale OM images without manual annotations.
• PLNPR offers a cost-effective and scalable solution for big data neuroimaging analysis.
• The method advances the investigation of neuronal circuits and brain mechanisms through improved large-scale image reconstruction.