Construction Of Computer-aided Medical Diagnosis System Based On Blood Test Data

In recent years,with the continuous establishment and improvement of medical information databases,artificial intelligence and medical health have become the important direction for the transformation of medical health industry,and computer-aided medical diagnosis has been gradually attracted major attention.In this study,machine learning algorithms are used to explore clinical testing data and to establish a robust computer-aided medical diagnosis system based on clinical needs to identify malignant diseases.The system can help doctors quickly identify malignant diseases and take timely measures.The successful construction of the system could imply the deep association between the diseases and routine blood indices,but also help to explore the correlation between other types of diseases and multi-component blood indices.In the first chapter,we briefly introduce the clinical status,methods and significance of routine clinical laboratory and roughly summarize the important role of medical big data in the implementation of precision medicine.At the same time,we also describe the shortcomings of current tissue biopsy and liquid biopsy in clinical application,as well as the research progress and advantages of CAD.Ultimately,the main machine learning algorithm(random forest)is introduced in detail.In the second chapter,the model constructed by the complex combination of 19 routine blood indexes that selected by random forest algorithm can accurately identify lung cancer patients from tuberculosis patients and healthy people.A total of 277 patients with 49 routine blood indices were collected in the study,including 183 lung cancer patients and 94 non lung cancer patients.After 10-fold cross-validation,the sensitivity,specificity and accuracy of the model were 96.3%,94.97% and 95.7%respectively,which revealed the potential correlation between the routine indices and lung cancer to some extent.The recognition model called RBLC shows stable prediction performance in the test set with sensitivity,specificity and accuracy of85.71%,90% and 88.24%,respectively.The combination of these routine bloodindices would be developed an effective tool to help clinicians quickly identify lung cancer samples from patients with tuberculosis.In the third chapter,in order to further verify the intrinsic relationship between machine learning and routine blood test data,we tried to identify patients with high incidence of gastric cancer from various gastric diseases and different cancers.In this study,more diversified data were included.A total of 2951 samples containing 58 routine blood indices were collected,including 2629 in the cross-validation set and322 in external validation set.The random forest algorithm finally selected 17top-ranked blood indices as input values for the gastric cancer early warning system called GCdiscrimination.After 10-fold cross-validation on the cross-validation set,the sensitivity,specificity,accuracy and AUC of the model were 0.9067,0.9216,0.9138 and 0.9720,respectively.The early warning system could not only provide a new strategy for fast and real-time identification of gastric cancer samples,but also reveal the profound correlation between gastric cancer and those routine indices,which helps to further understand the relationship between these parameters and gastric cancer and lay the foundation for clinical value.In the fourth chapter,in order to extend the above methods for other types of disease screening,we tried to identify infectious patients with active tuberculosis from latent tuberculosis infections.A total of 478 samples were enrolled in this study.The random forest algorithm selected 36 routine blood indices and two specific T-SPOT.TB results from 58 blood parameters.A recognition method for detecting active tuberculosis infection would be exploited through the complex combination of these indices.The method reveals good classification performance with the AUC of0.9256 and 0.8731 for the cross-validation set and external validation set,respectively.The work not only conceived an innovative strategy to identify active tuberculosis infection from the combination of routine blood indices and T-SPOT.TB results for the first time with the advantages of timely,efficient and economical,but also provided valuable information for comprehensive understanding of tuberculosis,which is helpful to further explore the far-reaching relationship between tuberculosis infection and routine blood examination.