Multi-Scale Segmentation- and Error-Guided Generative Adversarial Network for Neuron Segmentation

Accurate semantic segmentation is the key step to achieve accurate neuron instance segmentation. However, the complex structure of the brain makes it challenging to automatically segment neurons in the whole brain. A multiscale segmentation- and error-guided generative adversarial network is proposed to address the segmentation of neurons with large intra-class differences among neurons and small inter-class differences among neurons is small, and uneven distribution in different anatomical regions. This method applies the dual attention gate-based multiscale segmentation- and error-guided framework as the generator to predict neuron segmentation results, false negative errors and false positive errors. According to the three prediction results, the neuron semantic segmentation results are iteratively improved. Then, the generated segmentation results and the ground truth are, respectively, multiplied by the original images as the two inputs of the discriminator. The difference between the two branches is calculated by applying multiple convolutional operations to distinguish the generated segmentation results and the ground truth. The experimental results obtained on the macaque brain dataset show that compared with the reference methods such as U-Net, the performance indicators of whole brain neuron segmentation obtained by applying the proposed model have been improved, especially for light-stained neurons. Compared with eight reference models, the Dice value has increased by 20.3%, 1.2%, 15.5%, 2.5%, 8.2%, 159.2%, 0.8%, 13.1%, respectively. The proposed multiscale segmentation- and error-guided generative adversarial network can automatically and accurately segment the whole brain neurons, and can be further used for neuron instance segmentation and neuron morphological analysis.