Efficient cell classification of mitochondrial images by using deep learning
Efficient cell classification of mitochondrial images by using deep learning
复制标题
DOI：
10.1007/s12596-018-0508-4
复制
发表时间：
2019-03-01
影响因子：
1.8
通讯作者：
Luo, Bin
中科院分区：
文献类型：
--
作者：
Iqbal, Muhammad Shahid;El-Ashram, Saeed;Luo, Bin
受影响细胞、进化生物学和精准医学面临的关键挑战包括药物的影响以及了解药物处理细胞的粘度和强度。然而，这是非常困难的，因为巨大的细胞受到药物的影响。我们开发了一个基于深度学习的框架DNCIC，可以准确预测正常线粒体和罕见或未观察到的药物影响细胞。为了优化，我们使用卷积神经网络，并使用通过共聚焦显微镜收集的线粒体图像数据集进行训练。在正常细胞和病变细胞图像上验证了所获得的算法。我们已经训练了CNN，可以对（正常和受影响的细胞）双光子激发荧光探针图像进行分类。该模型对图像和视频的分类准确率达到98%。本研究结果为药物影响细胞的诊断奠定了基础。
Key challenges for affected cells, evolutionary biology and precision medicine include the effect of drug and understanding viscosity and intensity of drug-treated cells. However, this is extremely difficult because the enormous cells are affected by the drug. We developed a deep learning-based framework DNCIC that can accurately predict normal mitochondria and drug-affected cells that are rare or not observed. For optimization, we used a convolutional neural network and trained using a dataset of mitochondrial images, which were collected through the confocal microscope. The obtained algorithm was validated on the normal and affected cell images. We have trained CNN that can classify (normal and affected cells) two-photon excited fluorescence probe images. The proposed model has classified images and videos with 98% accuracy. Our results provided a foundation for drug-affected cell diagnosis.