Study On Prediction Of Gene Mutation And Disease-free Survival Of Lung Squamous Cell Carcinoma Based On Digital Pathological Images

During the past few years,lung cancer has become a global fatal disease with highest incidence rate and mortality rate.As one of the common pathological types of lung cancer,lung squamous cell carcinoma has unique clinical pathology and molecular characteristics.Besides,compared with other pathological types of lung cancer,patients with lung squamous cell carcinoma have shorter survival time and higher mortality rate.To reduce the mortality rate of patients,many studies have proposed to utilize gene detection techniques to identify unique gene mutations of cancer cells and design specific drugs for targeted genes.However,gene detection usually requires a long time and this may cause the patients to miss the optimal opportunity for treatment.To help patients obtain gene mutation information rapidly and assist doctors in developing optimal treatments,it becomes an emerging trend to predict gene mutations of lung squamous cell carcinoma based on digital pathological images.Apart from gene mutation,digital pathological images can also be utilized to predict disease-free survival of lung squamous cell carcinoma.In addition,several studies have pointed out that genomics data plays an important role in the development of lung squamous cell carcinoma.Therefore,it will be benefit to predict disease-free survival of lung squamous cell carcinoma by integrating digital pathological images and genomics data,which can also provide a scientific basis for doctors to make clinical diagnosis and help improve the life quality of patients.The main contributions of this paper are as follows:(1)In order to predict gene mutation of lung squamous cell carcinoma with digital pathological images,CellProfiler is first utilized to extract cell,cell nuclei and image features from each digital pathological image,and then multiple traditional algorithms are applied to predict the gene mutation of lung squamous cell carcinoma.The results show that those algorithms have obtained good performance.(2)To assess the effectiveness of digital pathological images and genomics data in predicting disease-free survival of lung squamous cell carcinoma,multiple traditional algorithms are utilized and the results show that the combination of digital pathological images and genomics data can improve the prediction performance.In order to better fuse different types of data,a novel lung squamous cell carcinoma disease-free survival prediction method called LSCDFS-MKL is proposed,in which multiple kernels for each data type are first built,then the kernel that efficiently reflects the characteristics of the lung squamous cell carcinoma dataset is selected,and the kernels are weighted to obtain the optimal kernel.Compared with other existing methods,our proposed method achieves a significant better performance.Finally,an independent validation dataset is utilized to assess the generalization ability of LSCDFS-MKL.Performance analysis shows that LSCDFS-MKL obtains a high accuracy.