MA10.05 Potential of CT and PET-Based Radiomics for the Diagnosis of Lung Adenocarcinomas Indicated for Limited Resection

Abstract

We previously developed radiomics technology using computed tomography (CT) to identify cases of preinvasive lung adenocarcinoma that may be indicated for limited resection. This study aimed to investigate whether the maximum standardized uptake value (SUVmax) on positron emission tomography (PET) has an add-on effect for this technology. This retrospective observational study was conducted at two medical centers. The resected pulmonary lesions were reviewed between April 2014 and March 2020. Patients with clinical stage IA lung adenocarcinomas were included. Thin-section CT images (less than 2.5 mm) were used to assess the clinical stage. For the image analysis algorithm of plain CT-based radiomics, we previously selected seven parameters: tumor volume (cm3), volume rate of the solid component (%solid; %), mean CT value (Hounsfield units), kurtosis, skewness, variance, and entropy. In this study, a new prediction model based on multiple logistic regression comprising eight parameters (CT＋PET-based radiomics) was generated, including SUVmax, after standardization to eliminate inter-facility bias. Next, stepwise variable selection was applied to the prediction model to evaluate whether SUVmax was a more important factor for identifying pathological preinvasive adenocarcinomas. A bootstrap method for randomized selection was used to assess the accuracy of the results. A total of 262 lung adenocarcinomas were categorized into the invasive group (n = 166 [63%]) or a preinvasive group (n = 96 [37%]), the latter of which comprised adenocarcinomas in situ (n = 32 [33%]) and minimally invasive adenocarcinomas (n = 64 [67%]). Before the surgery, the mean total tumor size of the invasive group (17.0 ± 6.5 mm) was significantly greater than that of the preinvasive group (13.9 ± 5.6 mm; p < 0.001). The pathological T factors were Tis (n = 32), T1mi (n = 64), T1a (n = 83), T1b (n = 66), and T1c (n = 17). Although spread through air spaces (STAS) was observed in 24 specimens (9.2%), five STAS-positive cases (5.2%) were noted in the preinvasive group. Upstaging was confirmed in one patient with lymph node-positive metastasis (N1) after pathological diagnosis. However, there were no cases of upstaging among the preinvasive tumors. Before standardization, the median SUVmax was 0.6 (interquartile range [IQR], 0.00–1.40) in the preinvasive group and 1.7 [IQR, 0.99–2.85]) in the invasive group (p < 0.001). After a stepwise method involving the eight factors, %solid (odds ratio [OR], 0.30; 95% confidence interval [CI], 0.20–0.46; p < 0.001) and entropy (OR, 0.46; 95% CI, 0.32–0.67; p < 0.001) were selected, but not SUVmax. The area under the curve was 0.83 (95% CI, 0.78–0.88). In the bootstrapping analysis, SUVmax was never selected as a key factor compared with the original parameters of CT-based radiomics. Even when the five STAS-positive cases in the preinvasive group were reclassified into the invasive group, SUVmax did not remain a key factor in the discrimination. In the identification of cases of preinvasive adenocarcinomas for which limited resection may be indicated, SUVmax did not increase the diagnostic efficacy of the original CT-based radiomics.