基于希尔伯特曲线-卷积神经网络的肿瘤类型预测

Tumor Type Prediction Based on Hilbert Curve and Convolutional Neural Network

针对肿瘤的早期检测,提出基于希尔伯特曲线-卷积神经网络(H-CNN)的肿瘤类型预测模型。该模型首先使用变分自编码器对32种肿瘤类型病人的RNA表达量和DNA甲基化数据进行融合,然后通过使用希尔伯特曲线将融合数据可视化后送到CNN进行训练。基于以上过程,可以实施关于新样本的肿瘤类型预测。实验结果表明,基于融合数据的H-CNN模型在肿瘤分类问题上具有优秀的性能,并且对肿瘤病人的早期诊断和治疗具有重要的指导意义。

A tumor type prediction model based on Hilbert curve and convolutional neural network(H-CNN)was proposed for the early detection of tumors.In this model,the data on the RNA expression and the DNA methylation from 32 tumor types was merged by the variational auto-encoder at first.Then,the merged data was visualized by using the Hilbert curve,and convolutional neural network was employed to train the visualized data.Based on the above process,the prediction of tumor types on new samples could be implemented.The experimental results showed that the excellent performance for tumor classification was gained in the H-CNN model based on the merged data,which played an important role in early diagnosis and treatment of tumor.