Deep learning based tracing difficulty classification on 3D neuron image block (pytorch)

0. Dataset sourece

Sample data are constructed from 3D neuron images of two whole mouse brains (denoted by brain-A and brain-B, respectively), gold standard reconstructions and automatic reconstructions of marked neurons in them. These data were provided by the Southeast University-Allen Institute Joint Center.

1. Data processing

Three types of sample data need to be constructed to achieve the tracing difficulty classification: 1. 3D image blocks, gold blocks, and auto blocks; 2. Neuron distance (ND) of gold blocks and auto blocks, and L-Measure (LM) of auto blocks; 3. Annotation data of 3D image blocks.

2. Model

A model called 3D-SSM is designed to classify the tracing difficultyof 3D image blocks, which is based on ResNet, Fully Connected Neural Net-work (FCNN) and Long Short-Term Memory network (LSTM). 3D-SSMconsists of three modules: Structure Feature Extraction (SFE), Sequence Informa-tion Extraction (SIE) and Model Fusion (MF). SFE utilizes a 3D-ResNet and aFCNN to extract two kinds of features in 3D neuron image blocks and automaticreconstruction blocks. SIE uses two LSTMs to extract sequence information hid-den in features of sequential blocks produced in SFE. MF adopts a concatenationoperation and a FCNN to fuse outputs from SIE.

3. Code

            
├── deal_feature            Extraction and analysis of L-Measure and neuron distance features.
├── automatic_label         Train an FFCNs model and use it to generate automatic labels.
├── classification_model    Models are designed to solve the tracing difficulty classification of 3D image blocks.
├── vaa3d_plugin            Plug-in for getting 3D image blocks, gold blocks, and auto blocks.
├── annotation.md           Annotation relus.
├── README.md              

1. Get 3D image blocks, gold blocks, and auto blocks using vaa3d_plugin file.
2. Get L-Measure and neuron distance features using deal_feature file, and get automatic label using automatic_label file .
3. To train and predict the tracing difficulty of 3D image blocks models using classification_model file.