ABSTRACT
Objective:To establish a differential diagnosis model for IgA nephropathy and non-IgA nephropathy based on machine learning algorithms.Methods:Retrospective study adopted,from 2019 to 2020,260 patients were referred to the Department of Nephrology at the First Affiliated Hospital of Kunming Medical University, the First People′s Hospital in Yunnan province, and Yan′an Hospital of Kunming city. All patients were diagnosed by renal pathology, 130 cases of primary IgA nephropathy, the 130 cases of non-IgA nephropathy. Collection of materials, including gender and age, 28 clinical data, and routine laboratory test results,the sex ratio of IgA nephropathy group and non-IgA nephropathy group were 59∶71 and 64∶66 respectively, the ages were 37.20 (21.89, 53.78) and 43.30 (27.77, 59.18) years, respectively. 260 patients were divided into a training set (70%, 182 cases) and a test set (30%, 78 cases). Using the decision tree, random forests, support vector machine, extreme gradient boosting to establish a differential diagnosis model for IgA nephropathy and non-IgA nephropathy. Based on the true positive rate, true negative rate, false-positive rate, false-negative rate, accuracy, subjects features work area under the curve(AUC), the precision ratio, recall ratio, and F1 score, comprehensively evaluate the performance of each model, finally, the best performance of the model was chosen. Using SPSS 25.0 to analyze the data, P<0.05 was considered to be statistically significant. Results:The accuracy of the decision tree, support vector machine, random forests and extreme gradient boosting establish differential diagnosis model was 67.95%, 70.51%, 80.77% and 83.33%, respectively; AUC values was 0.74, 0.76, 0.80 and 0.83; Judgment for primary IgA nephropathy F1 score was 0.73, 0.72, 0.80 and 0.83, respectively. The efficiency of the extreme gradient boosting model based on the above evaluation indicators is the highest, its diagnosis of IgA nephropathy of the sensitivity and specificity respectively 89% and 79%. The variable importance from high to low was blood albumin, IgA/C3, serum creatinine, age, urine protein, urine albumin, high-density lipoprotein cholesterol, urea.Conclusion:The differential diagnosis model for IgA nephropathy was established successfully and non-IgA nephropathy and the efficiency performance of the extreme gradient boosting algorithm was the best.
ABSTRACT
Objective To explore infrared spectrum characteristics of different voltages induced electrical injuries on swine skin by using Fourier transform infrared-microspectroscopy (FTIR-MSP) combined with machine learning algorithms, thus to provide a reference to the identification of electrical skin injuries caused by different voltages.Methods Electrical skin injury model was established on swines.The skin was exposed to 110 V, 220 V and 380 V electric shock for 30 s and then samples were took, with normal skin tissues around the injuries as the control.Combined with the results of continuous section HE staining, the FTIR-MSP spectral data of the corresponding skin tissues were acquired.With the combination of machine learning algorithms such as principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA), different spectral bands were selected (full band 4 000-1 000 cm-1and sub-bands 4 000-3 600 cm-1, 3 600-2 800 cm-1, 2 800-1 800 cm-1, and 1 800-1 000 cm-1), and various pretreatment methods were used such as orthogonal signal correction (OSC), standard normal variables (SNV), multivariate scatter correction (MSC), normalization, and smoothing.Thus, the model was optimized, and the classification effects were compared.Results Compared with simple spectrum analysis, PCA seemed to be better at distinguishing electrical shock groups from the control, but was not able to distinguish different voltages induced groups.PLS-DA based on the 3 600-2 800 cm-1band was used to identify the different voltages induced skin injuries.The OSC could further optimize the robustness of the 3 600-2 800 cm-1band model.Conclusion It is feasible to identify electrical skin injuries caused by different voltages by using FTIR-MSP technique along with machine learning algorithms.