CT-based radiomics to construct a differential diagnosis model for immune-related interstitial pneumonia

<bold>Background:</bold> Immune-related interstitial pneumonia(CIP) is a serious adverse reaction after immune checkpoint inhibitor treatment, impacting patient prognosis and posing life-threatening risks if untreated promptly. However, the diagnosis of CIP heavily relies on clinician expertise, even requires lung tissue biopsy, and a number of patients are misdiagnosed. There is no accurate and efficient non-invasive artificial intelligence model for distinguishing CIP. <bold>Objective:</bold> To build a differential diagnosis model , to distinguish between CIP and various pneumonias. It will improve the accuracy and efficiency of non-invasive diagnosis of CIP. <bold>Methods:</bold> The experimental group collected lung CT images at the onset of CIP,while and the control group collected lung CT images after using ICIs but did not develop CIP. The SVM differential diagnosis model was constructed through machine learning method. And build a higher accuracy regression model combined with clinical characteristics. <bold>Result:</bold> 72 patients were included in experimental group and 69 patients in control group.A total of 258 significant imaging features were identified. By regressing these features, constructed an SVM model and drawn the ROC curve. The area under the curve was 0.77(Figure 1). Then combined with clinical characteristics, we build an enhanced regression model. In the validation set, the regression model had a better diagnostic rate than the SVM model(0.69<0.98)(Figure 2). <bold>Conclusion:</bold> Our study constructed a reliable regression model for differential diagnosis of CIP. This non-invasive model improves the accuracy and efficiency of CIP diagnosis and reflects the benefits of artificial intelligence in disease diagnosis.