| Background and objectives: Pleural and peritoneal effusions are common in clinical practice, and the causes are mainly tuberculosis, cancer and major organ dysfunction, accounting for 80% to 90%. Despite diagnostic methods are developing, it is still difficult to diagnose some pleural and peritoneal effusions. By means of thoracoscopy or laparoscopy, thoracic or peritoneal biopsies and other invasive methods, it is found out that cancer and tuberculosis account for 80% of the causes of the cause-unknown pleural and peritoneal effusions. For this reason, researchers have been engaged in developing effective diagnostic methods, especially for pleural and/or peritoneal effusions caused by malignant tumours and tuberculosis. In recent years, the studies have focused on single markers for the laboratory diagnosis of pleural and/or peritoneal effusions etiologically and no significant breakthrough has been achieved. Computer-aided diagnosis (CAD) is a combination of computer science, mathematics and medicine, with the advantages of integrating multiple factors to create mathematical models for diagnosis, which is a promising approach for clinical diagnosis. In this study, we focused on creating mathematical models on the basis of cellular morphology data, the clinical data, results of routine blood and biochemical tests and pleural and/or peritoneal effusion tests, in order to find a new approach to etiologically diagnose pleural and/or peritoneal effusions.Methods: (1) Patients and diagnosis: 222 patients with pleural and/or peritoneal effusions caused by tuberculosis or malignant tumors hospitalized in the First Affiliated Hospital of Nanchang University from 2001 to 2007 were enrolled in the study, with average years of 53.1, 120 males and 102 females, and consisting of 46 cases of tuberculosis peritonitis, 26 cases of tuberculosis pleurisy, 50 cases of lung cancer, 23 cases of gastric cancer, 11 cases of liver cancer, 43 cases of other malignant tumors. Tuberculosis peritonitis or pleurisy is diagnosed by medical imaging, cytological tests and results of specific chemical treatment, pleural and/or peritoneal effusions caused by tumors diagnosed by medical imaging, cytology and histopathology. (2) Cytological morphological data collection: the smears of pleural and/or peritoneal effusion cells were prepared and stained with H-E. More than 4 typical pictures for each individual specimen were taken and the division and measurement of single cells were performed with HMIA-2000 high resolution pathological imaging analysis system. The means of all parameters were calculated with a total of more than 200 cells. (3) Clinical data collection: The medical documents of the cases were reviewed and the diagnosis were checked and confirmed. The following data were collected: age, gender, temperature, results of blood cell analysis, the results of blood biochemistry profile, routine, biological and cytological test results of pleural and/or peritoneal effusion, the results of serum tumor marker tests, imaging examination results, pathological findings. The data were inputted in the Excel sheet. (4) Establishment and evaluation of mathematical models: Based on morphology data, clinical data and the both, respectively, the diagnostic mathematical models for differentiating pleural and/or peritoneal effusions caused by tuberculosis and cancer and for these caused by different kinds of cancers. The models were testified and evaluated by the data back substitution. Multivariate discrimination analysis or Logistic regression analysis were finished by SPSS 13.0 software.Results: (1) The diagnostic models for differentiating tubercular and tumorous pleural and/or peritoneal effusions and their evaluations: The models for tubercular and tumorous pleural and/or peritoneal effusions were established on the basis of Y-axis projection and others 13 cell morphology parameters, and the diagnostic accuracies were 95.7% in the tubercular effusions and 92.7% in tumorous effusions, total 93.6%. The another models for tubercular and tumorous pleural and/or peritoneal effusions were established on the basis of 12 clinical parameters including gender, hemoglobin and others, and the accuracies were 82.2% in the tubercular effusions and 87.3% in tumorous effusions, total 85.2%. The models for tubercular and tumorous pleural and/or peritoneal effusions were also established on the basis of 17 morphological and clinical parameters including temperature, area volume and others, and the accuracies were 100.0% in the tubercular effusions and 96.3% in tumorous effusions, total 97.3 %. (2) The diagnostic models for differentiating tumorous pleural and/or peritoneal effusions and their evaluations: The models for differentiating tumorous pleural and/or peritoneal effusions caused by ovarian cancer, lung cancer, stomach cancer, liver cancer and other cancers were established on the basis of area volume and other 14 cellular morphological parameters, and the diagnostic accuracies are 75.0%, 85.7%, 88.9%, 85.7% and 80.0% in the correspondent effusion diagnosis, respectively, total 82.1%. The another models for differentiating above 5 kinds of tumorous pleural and/or peritoneal effusions were established on the basis of gender and other 17 clinical parameters, and the accuracies are 93.3%, 66.7%, 76.9%, 88.9% and 82.4% in 5 kind of different tumorous effusion diagnosis, respectively, total 81.8%. The models for differentiating above 5 kinds of tumorous pleural and/or peritoneal effusions were also established on the basis of gender, area volume and other 18 clinical or cell morphology parameters, and the accuracies are 92.9%, 100.0%, 100.0%, 100.0%, 100.0%, respectively, total 98.0%. Conclusions: (1) The mathematical models created on the data of exfoliated cellular morphology are valuable to differentiate tubercular and tumorous pleural and/or peritoneal effusions and different tumorous ones, with accuracy of 93.6% in discrimination of tuberculosis and cancer effusions and 82.1% in discrimination of different cancer effusions. (2) The mathematical models created on the data of the simple clinical and laboratory data are valuable to differentiate tubercular and tumorous pleural and/or peritoneal effusions and different tumorous ones, with accuracy of 85.2% in discrimination of tuberculosis and cancer effusions and 81.8% in discrimination of different cancer effusions. (3) The mathematical models created on both of the simple clinical and laboratory and morphological data are valuable to differentiate tubercular and tumorous pleural and/or peritoneal effusions and different tumorous ones, with accuracy of 97.2% in discrimination of tuberculosis and cancer effusions and 98.0% in discrimination of different cancer effusions... |
PDF Full Text Request