| Gastric cancer is one of the high-incidence malignant tumors in the world.Early diagnosis and treatment of gastric cancer can effectively reduce mortality and improve the prognosis of patients.At present,the diagnosis of gastric cancer is mainly through gastroscopy.How to assist doctors in rapid detection of lesions during gastroscopy and accurate early diagnosis has important clinical value.In this study,deep learning based on the convolutional neural network model was carried out to develop computer-aided program to detect anatomy,gastric polyps,submucosal tumors(SMTs)and early gastric cancers(EGCs).According to the verification in gastroscopic images and videos,we found that the AUC of the ROC curve detecting gastric polyps,SMTs and EGCs is 0.970,0.892 and 0.811.The sensitivity,specificity and accuracy of EGC-detecting program is 67.9%,95.4%and 98.1%.Computer-aided detecting programs have certain accuracy and clinical value in detecting these lesions.Using phage display peptide library,biopanning was performed to select phages of high affinity to gastric cancer.According to the sequencing of selected phages,two new peptides was screened.The affinity and specificity of peptides were evaluated by cytology and histology.We found that the two fluorescence-label peptides had high affinity to gastric cancer cells and tissues,and good sensitivity in diagnosis.The fluorescence intensity of two peptides was significantly related to invasion depth and tumor size.They are promising to become molecular probes to detect gastric cancers.Part ? Development and validation of a deep-learning algorithm based on convolutional neural networks for detection of anatomy in gastroscopic imagesSubject:Development and evaluation of a detection program to recognize anatomies in gastroscopic images.Methods and materials:An endoscopic image marking platform was constructed.Marking correspond 10 anatomies in 509 cases of 4732 gastroscopy images to obtain an image training set and a test set.Deep learning was performed based on convolutionalneural model,developing a program to detect the anatomies of gastroscopy images.accuracy of the program in detecting anatomies of gastroscopy image is evaluated by the detection of the test set.Results:The AUC of the ROC curve of program detecting the esophagus,dentate line,fundus,corpus,gastric angle,antrum,pylorus,duodenal sphere and descending part of the program is 0.871,0.985,0.969,0.984,0,996,1.000,0.979,0.999,0.990 and 0.977.In the image with 2 marking boxes,the anatomy detectiong rate is 90%.An detection rate of 92.8%in the images with 1 marking box.Conclusion:A program to detect the 10 anatomies in gastroscopy image was developed.program has a high accuracy in detecting anatomies in gastroscopy images.Part ? Development and validation of a deep-learningalgorithm based on convolutional neural networks for detection of gastric protrusion lesions and early gastric cancersSubject:Development and evaluation of aprogram to detect gastric polyps,SMTs and EGCs in gastroscopic images.Methods and materials:Marking corresponding lesions in 307 cases of 455 gastric polyp images,109 cases of 539 SMTs images and 127 cases of 672 EGCs images to obtain image training sets and test sets.Marking 101 cases of 101 images as normal group during validaion.Deep learning was performed based on convolutionalneural network model to develop programs to detect lesions.Test images and videos were used to evaluate the effect of detection of three programs.The accuracy of the detection program in detecting lesion location in images was evaluated by the area ratio of overlapping box and marking box or detection box.Results:The AUC of the ROC curve detecting gastric polyps,SMTs and EGCs is 0.970,0.892 and 0.811.The ratio of overlapping and marking boxes was 100.0%-100.0%,83.2%-86.5%and 81.1%-87.0%.The ratio of overlapping boxes and detection boxes is 41.0%-45.0%?92.1%-92.5%and 68.1%-81.8%.Calculated by frame,the detection sensitivity of the program to gastric polyps,SMTs and EGCs is 53.5%,100.0%and 67.9%,the specificity is 99.2%,99.4%and 95.4%,the accuracy is 85.4%,99.6 and 98.1%.Conclusion:Programs to detect the gastric polyps,SMTs and EGCs was developed.The programs have certain accuracy in images and videos detection,and high accuracy of lesions localization in the image.Part ? Screening affinity phage from phage display library for targeting gastric cancer cellsSubject:Screening affinity phages from phage display library for targeting gastric cancer.Methods and materials:Taking 3 gastric cancer cell lines AGS,MKN-45 and HGC-27 as target cells,normal gastric epithelial cells GES-1 as adsorptive cells,3 rounds of biopanning from phage display 12 peptide library was performed.Each group randomly selected 30 monoclonal phages for sequencing.The affinity and specificity of phage to gastric cancer cells were evaluated by the analysis of P/N value and specificity calculated from cell ELISA.Results:After 3 rounds of biopanning,phages of affinity to AGS,MKN-45 and HGC-27 cell lines were enriched by 31.19,5.65 and 7.16 times.The sequence analysis suggested that the RNHSS sequences repeated 17,15 and 18 times in AGS,MKN-45 and HGC-27,and IPLVVPF sequences repeated 7 and 2 times in AGS and HGC-27.Cell ELISA indicated that the P/N values of RNHSS phage to AGS,MKN-45 and HGC-27 were 3.25,2.23 and 3.16,the specificity values were 9.51,6.76 and 8.24.The P/N values of IPLVVPF phage to AGS,MKN-45 and HGC-27 were 3.30,2.07 and 3.27,with specific values of 5.97,3.60 and 5.32.Homologous analysis suggested that these two peptides had low similarity rate with existing protein sequences,and no reports were found in literature retrieval.Conclusion:After 3 rounds of biopanning,2 sequences of phages were screened.These two phages had good affinity and specificity to gastric cancer cells.The corresponding peptides sequence were new types of peptides.Part IV Affinity validation of fluorescence-labeled peptides targeting gastric cancer cells and tissuesSubject:To evaluate the efficacy of two fluorescence-labeled peptides targeting gastric cancer cells and affinity of fluorescence-labeled peptides targeting gastric cells and tissues.Methods and materials:Two peptides were synthesized and FITC modified.The fluorescence intensity of AGS,MKN-45 and HGC-27 treated by different concentrations of fluorescent peptides was detected by flow cytometry,and the equilibrium dissociation constant was calculated by fitting the dissociation curve,and the fluorescence intensity of AGS,MKN-45 and HGC-27 with different fluorescence peptides processing time was detected by flow cytometry.The reaction rate constant was calculated by fitting the first-level reaction kinetic curve.The affinity and specificity of the two peptides to gastric cancer cells and tissues were verified by cellular immunofluorescence and tissue immunofluorescence.The sensitivity of two peptides to diagnose gastric cancer was calculated by the fluorescence intensity ratio of gastric carcinoma and the adjacent tissues.The correlation between clinicopathological features with fluorescence intensity of fluorescence-labeled peptides was analyzed.Results:The equilibrium dissociation constants of RNHSS-FITC(RNH*-FITC)to AGS,MKN-45 and HGC-27 cells are 16.48 ?M,35.21 ?M and 26.61 ?M,and the reaction rate constants are 0.052/min,0.057/min and 0.055/min.The equilibrium dissociation constants of IPLVVPF-FTIC(IPL*-FITC)to AGS,MKN-45 and HGC-27 cells are 13.19?M,17.22 ?M and 18.49 ?M,and the reaction rate constants are 0.049/min,0.056/min and 0.041/min.Tissue immunofluorescence indicated that the fluorescence intensity of RNH*-FITC and gastric cancer tissues(5528.41±3551.64)was significantly higher than that of adjacent tissues(1541.19±450.45)and chronic gastritis tissues(1305.36±392.63)(p<0.01).The fluorescence intensity of IPL*-FITC and gastric cancer tissues(3671.20±1313.15)was significantly higher than that of adjacent tissues(1144.04±297.38)and chronic gastritis tissues(1184.31±300.22)(p<0.01).The sensitivity of RNH*-FITC and IPL*-FITC to the diagnosis of gastric cancer was 76.7%and 87.8%when the cutoff value of fluorescence intensity ratio of gastric carcinoma and the adjacent tissues was 2.Fluorescence intensity of RNH*-FITC with gastric cancer tissues and fluorescence intensity ratio of carcinoma and the adjacent tissues was related to the size of tumors(p<0.05).Fluorescence intensity of IPL*-FITC with gastric cancer tissues and fluorescence intensity ratio of carcinoma and the adjacent tissues was related to the invasion of tumors(p<0.05).Conclusion:The affinity and specificity of RNH*-FITC and IPL*-FITC to gastric cancer cells are good.The sensitivity of diagnosing gastric cancer was high.Fluorescence intensity was significantly related to tumor size and depth of invasion,respectively.They are expected to become molecular probes for detecting gastric cancers. |
PDF Full Text Request