18F–FDG PET/CT is widely used to characterize tumor glycolytic activity, which is a valuable marker of tumor biological behavior [16,17]. In the present study, we assessed the usefulness of 18F–FDG PET/CT in differentiating STS and BS from benign lesions. Numerous studies have demonstrated that PET-derived semi-quantitative estimation parameters, such as SUVmax, MTV, and TLG, are valuable diagnostic indicators. Specifically, SUVmax is a marker of glucose metabolism of a single integrin in the tumor. On the other hand, MTV and TLG reflect the global metabolic activity of the tumor. However, the ability of individual parameters to discriminate between malignant and benign tumors in soft tissues and bones is not always adequate. Soft tissue and bone tumors are highly heterogeneous. Importantly, delays in the diagnosis have a negative impact on the final outcome [18]. Thus, the development of a simple and reliable imaging model to characterize biological behavior is critical to overcome the aforementioned limitations. In the present study, we comprehensively evaluated the feature parameters of 18F–FDG PET/CT imaging and constructed an effective model based on SUVmax and HF for differential diagnosis between malignant and benign soft tissue and bone tumors.
SUVmax, MTV, and TLG have been previously demonstrated as strong predictors of sarcoma cell proliferation and disease progression [17,19]. Several studies have claimed that SUVmax and its retention index could both be used to differentiate between benign and malignant soft tissue or bone lesions [20,21]. Nonetheless, SUVmax is not a precise indicator of the global metabolic activity of tumors. Moreover, a positive correlation between the FDG activity and the pathological grade of sarcoma has been established; however, the histological sub-types cannot be always distinguished accurately [22]. Specifically, some benign lesions may exhibit deceptively high FDG uptake, leading to indefinite diagnoses [23]. HF is another parameter obtained from PET images and was reported to reflect the intratumoral heterogeneity of 18F-FDG affinity. A study reported by Alipour R, et al. [24] showed that HF values for malignant parotid tumors were higher than for benign ones. Thus, HF was established as a reliable parameter for distinguishing between benign and malignant parotid tumors. Furthermore, Kim SJ, et al. [25] found that HF could be employed as a predictor for characterization of thyroid nodules. Nevertheless, the above studies only relied on univariate analysis, and multivariate analysis to eliminate the interaction among variables was not conducted. In the current work, the significant feature parameters between the malignant and benign lesion groups were screened according to the results of univariate analysis. The investigated parameters included tumor size, visual characteristics, SUVmax, MTV, TLG, and HF (all P values were <0.05). Additionally, multivariate logistic regression analysis identified SUVmax and HF as the only independent risk factors for malignant tumors. Nakajo M, et al. [26] carried out a univariate analysis on 63 cases of musculoskeletal tumors using the cumulative SUV-volume histogram (CSH) method. The results showed that the AUC of CSH for malignant tumors was higher than that for benign ones, which was in agreement with the outcomes of the present study. However, this approach is analogous to the concept of dose-volume histograms for evaluating radiotherapy regimens, which uses PET/CT functional imaging data; thus, the clinical practicality is extremely limited. Xu R, et al. performed texture analysis for the differential diagnosis of bone and soft tissue lesions. The results obtained in this study revealed that utilizing optimal texture parameters combined with PET and CT imaging showed significantly better performance compared to SUVmax. Accordingly, the importance of combining parameters for differential diagnosis of diseases has been demonstrated. [27].
The vascular distribution and necrosis characteristics of each tumor cell population affect the growth rate [28]. The results obtained in the present study showed that the regression model AUC (AUC: 0.860, 95%CI: 0.771~0.948, P = 0.000) was higher than that of SUVmax (AUC: 0.744, 95%CI: 0.628~0.860, P = 0.000) and HF (AUC: 0.790, 95%CI: 0.684~0.896, P = 0.000). Based on the optimal cut-off values for the model P value (0.47), SUVmax (5.95), and HF (0.46), the diagnostic accuracy of individual parameters and their combination was assessed with respect to sensitivity and specificity. The outcomes demonstrated that the diagnostic performance of the regression prediction model combined with the SUVmax and HF parameters was considerably improved, particularly for specificity (all P values were <0.01).
Generally, the 18F-FDG uptake is not homogeneous within tumors. The biological characteristics of tumors are determined not just by the tumor cells, but also by its microenvironment, including immune cells, endothelial cells, and tumor-related fibroblasts [29]. SUVmax reflects the highest glucose metabolism in tumor cells, while HF indicates the intratumoral heterogeneity of glucose metabolism. The combination of SUVmax and HF incorporates intertumoral structures, comprehensively reflecting the glucose metabolism inside the tumors and enabling more accurate characterization of the biological behavior. Previous research showed that the tumor size and volume are often considered as indicators of tumor malignancy [30,31]. However, the multivariate logistic analysis conducted in this study demonstrated that when SUVmax and HF were simultaneously introduced to the regression model, the tumor size, MTV, and TLG were not statistically significant, indicating the presence of a certain overlaps and interactions between the parameters. The predictive value of tumor size and volume for a single location is limited. In addition, rapid proliferation of lesions indicates the presence of malignant tumors [32].
We subsequently compared the regression model with conventional imaging (DCE-MRI or enhanced CT). The results suggest that the sensitivity was similar for both approaches (20/24 vs.19/24, respectively). In contrast, the specificity for the model P values was significantly higher than for conventional imaging (17/21 vs.12/21, respectively). When the traditional images were reanalyzed, it was determined that two hematomas and a lesion rich in blood supply (Kaposi hemangioendothelioma and pleomorphic hyalinizing angiectatic tumor) were false positive. Notably, those were correctly diagnosed by the regression model. The present study is retrospective; therefore, most of the enrolled patients had suspected malignant lesions and underwent PET/CT imaging, the results of which may be subjective. The obtained results were sufficient to conclude that performing biopsy or surgical resection in patients with suspected malignant disease should be done with caution. Moreover, it was established that assessment using the PET/CT regression model prior to clinical decision might complement radiologic tomography, which is consistent with previous research [33].
Despite encouraging results, six false positive benign lesions (e.g.,giant cell tumors and inflammatory myofibroblastic tumors) as well as six false negative malignant lesions (e.g., myxofibrosarcomas) were determined by the regression model.18F-FDG is an analog of glucose and previous studies claimed that lesions with abundant infiltration of inflammatory cells or ones containing giant cells can display upregulation of hexokinase-2, leading to high FDG affinity [34]. Conversely, malignancies, which were rich in mucous matrix, usually exhibited insufficient glucose transporter expression and showed low FDG uptake [35].In addition, a study by Lee AY, et al. showed that myxofibrosarcomas with a higher proportion of mucus are associated with a better prognosis [36]. Therefore, we speculate that this is the reason why the PET/CT imaging features of these tumors tend to be benign ones. Undeniably, biopsy remains as the gold standard for precise diagnosis.
The current study has certain limitations. Firstly, the sample size was not sufficiently large due to some STT pathological classifications being relatively rare. Secondly, the conducted study is a retrospective one, in which the pathological classification was confirmed by biopsy only in some of the cases. Moreover, the histological sub-type of several cases was not clearly defined. Nevertheless, we believe that the results of the present work provide a valuable reference for further research in this area. We propose a new concept, which effectively integrates the metabolic information obtained from 18F-FDG PET/CT imaging. The described approach can be used as a clinical standardized tool for the management of soft tissue and bone tumors. In particular, the methodology considerably enhances the specificity of imaging to avoid excessive pathological biopsies and unplanned surgical resections. A large sample of prospective cohort studies, involving more characteristic imaging parameters and histopathology factors should be carried out in the future.