Establishment of multi-modal image database of drug-resistant pulmonary tuberculosis and research on evaluation system of artificial intelligence products
Objective To establish a multi-modal image database of drug-resistant pulmonary tuberculosis and to evaluate it with artificial intelligence product evaluation system.Methods The data of 750 patients with drug-resistant pulmonary tuberculosis were collected,and a multi-modal image database of drug-resistant pulmonary tuberculosis was constructed.Established AI-aided di-agnosis model based on CT image big data,study automatic segmentation model and differential diagnosis model of drug-resistant pulmo-nary tuberculosis focus,and design activation function.AI clinical auxiliary diagnosis decision-making system for pulmonary tuberculo-sis was researched and developed.Results The results of segmentation of drug-resistant pulmonary tuberculosis lesions by AI algo-rithm model were close to the gold standard,and the segmentation accuracy was high.Among the 750 patients,726 cases were diagnosed by the artificial intelligence product evaluation system(96.80％).ROC curve analysis showed that the sensitivity,specificity,accuracy and AUC of the artificial intelligence product evaluation system for the diagnosis of drug-resistant tuberculosis were 95.73％,85.70％,96.80％and 0.912(95％CI is 0.865～0.956).Conclusion The multi-modal image database of drug-resistant pulmonary tuber-culosis in China is constructed for the first time,and the AI-aided diagnosis system developed has high value in accurate diagnosis and differential of pulmonary tuberculosis.
万方数据
维普
