Nuclei_Segmentation_Experiments_Demo_By_DIMAN

We present a novel and efficient computing framework for segmenting the overlapping nuclei by combining Marker-controlled Watershed with our proposed convolutional neural network (DIMAN). We implemented our method based on the open-source machine learning framework TensorFlow and reinforcement learning library TensorLayer.This repository contains all code used in our experiments, incuding the data preparation, model construction, model training and result evaluation. For comparison with our method, we also utilized TensorFlow and TensorLayer to reimplement four known semantic segmentation convolutional neural networks: FCN8s, U-Net, HED and SharpMask. Beside this, we also compare our method with four published state-of-art methods.

本研究所提出一种新颖且高效的核分割计算框架，该框架通过结合标记控制的分水岭算法与所提出的卷积神经网络（DIMAN）实现重叠核的分割。本研究方法基于开源机器学习框架TensorFlow及强化学习库TensorLayer进行实现。本仓库包含实验中所使用的全部代码，包括数据准备、模型构建、模型训练及结果评估。为与本研究方法进行对比，我们还利用TensorFlow和TensorLayer重新实现了四种已知的语义分割卷积神经网络：FCN8s、U-Net、HED和SharpMask。此外，我们还对比了四种已发表的业界领先方法。