Purpose:
Pulmonary inflammatory pseudotumor (PIP) is an inflammatory proliferative tumor-like lesion that frequently exhibits hypermetabolism on 18F-fluorodeoxyglucose (FDG) positron emission tomography/computed tomography imaging (PET/CT) and is readily misdiagnosed as a malignant tumor. The purpose of this study was to identify PIP by combining PET/computed tomography metabolic and blood test characteristics with machine learning.
Patients and Methods:
We recruited 27 patients with PIP and 28 patients with lung cancer (LC). The PET metabolic and blood test parameters were collected, and the differences between the groups were evaluated. In addition, we combined the support vector machine (SVM) classifier with the indicators that differed between the groups to classify PIP and LC.
Results:
For PET metabolic parameters, our findings showed that, as compared with the LC group, maximal standardized uptake value (P< 0.001, t = −4.780), Mean standardized uptake value SUVmean, P< 0.001, t = −4.946), and SD40% (P< 0.001, t = −4.893) were considerably reduced in the PIP group, whereas CV40% (P= 0.004, t = 3.012) was significantly greater. For blood test parameters, the total white blood cell count (P< 0.001, t= 6.457) and absolute neutrophil count (P< 0.001, t= 6.992) were substantially higher in the PIP group than in the LC group. Furthermore, the performance of SVM trained solely on PET metabolic parameters (mean area under the curve [AUC] = 0.84) was comparable to that of SVM trained solely on blood test parameters (mean AUC = 0.86). Surprisingly, utilizing the combined parameters increased SVM performance significantly (mean AUC = 0.98).
Conclusion:
PET metabolic and blood test parameters differed significantly between the PIP and LC groups, and the SVM paradigm using these significantly different features has the potential to be used to classify PIP and LC, which has important clinical implications.