Primary renal sarcomas: imaging features and discrimination from non-sarcoma renal tumors

Renal cancer accounted for 2.2% of all malignant diseases worldwide in 2018 [1]. The most common renal cancer subtypes, such as clear cell or papillary renal cell cancer, originate from renal epithelial cells [2]. One tumoral mechanism linked to an increased likelihood of local invasion and distant metastases is the so-called epithelial-mesenchymal transition (EMT): this process includes disruption of the renal epithelial polarity and barrier integrity, resulting in a mesenchymal phenotype termed sarcomatoid renal cancer [3].

In contrast, sarcomas are rare tumors of mesenchymal origin that can manifest throughout the body [4]. In two older studies, renal sarcomas are exceptionally rare with a prevalence of less than 1% of all renal malignancies [5, 6]. Although there are no recent epidemiological studies on renal sarcomas, unpublished data from the United States National Cancer Database and Surveillance, Epidemiology, and End Results (SEER) database conform the rarity of renal sarcomas contributing < 1% of renal malignancies (unpublished data, 2021).

The prognosis of patients presenting with renal sarcomas is worse compared to non-sarcoma renal tumors, with 5-year overall survival rates of 14.5% reported in one study by Wang and colleagues [6]. This dismal prognosis highlights the clinical need for a timely and accurate diagnosis of renal sarcomas to provide individualized treatment strategies for affected patients.

For renal mass assessment in general, cross-sectional imaging studies, such as computed tomography (CT) and magnetic resonance imaging (MRI), are the cornerstone for initial diagnosis and treatment planning. For renal sarcoma in particular, CT and MRI imaging features have been evaluated in three review articles and numerous case reports [7‐9]. Several imaging features unique to renal sarcomas have been described in these studies, such as capsular growth and central calcifications [7‐9]. However, the literature does not assess the frequency of these specific features and lacks standardized imaging assessment.

Diagnosing renal sarcomas based on cross-sectional imaging alone remains a radiological challenge given the wide imaging spectrum and overall rarity of the disease, as well as oftentimes non-standardized imaging techniques [7‐9]. Still, imaging-based diagnosis of renal sarcoma might optimize clinical management by identifying patients with dismal prognosis and high probability of metastatic disease. Further, international guidelines advocate for risk-adapted perioperative treatment of retroperitoneal sarcomas, including neoadjuvant chemotherapy and radiation, which could play a role in renal sarcoma treatment as well [10]. On the other hand, the role and diagnostic accuracy of renal biopsy in cases with large necrotic tumors, such as renal sarcomas, remains unclear [11].

Therefore, this multicenter study aims to systematically assess imaging features of renal sarcomas, and to present a method for accurate discrimination of renal sarcomas from non-sarcoma renal tumors.

This retrospective study was approved by the ethics committee of Magdeburg University (Nr 148/20). Imaging assessment and statistical analyses were conducted between September and December 2020.

Renal sarcomas are rare mesenchymal tumors with a reported incidence < 1% of all renal malignancies [5, 6]. Although renal sarcomas are associated with a dismal prognosis, so far there were no efforts to systematically describe their imaging features and to clinically discriminate renal sarcomas from non-sarcoma renal tumors.

Including data from 11 European centers, this study reports on a total of 34 patients with renal sarcomas, with leiomyosarcoma and Ewing sarcoma being the most common histological subtypes (23.5% and 14.7%, respectively). Given the multicenter design and long accrual period, the resulting database was heterogeneous regarding imaging techniques.

Compared to non-sarcoma renal tumors, renal sarcomas generally manifested as more complex renal masses in younger patients and were more complex renal masses with larger diameter and irregular shape as well as ill-defined margins, compared to non-sarcoma renal tumors. Renal sarcomas also demonstrated a more aggressive local growth pattern with more frequent invasion of the renal vein or inferior vena cava, tumor necrosis, direct invasion of adjacent organs, and contact to renal artery or vein.

Using imaging features and demographics, a random forest algorithm showed good discrimination of renal sarcomas versus non-sarcoma renal tumors with an AUC = 93.8%. Using a robust 10-fold cross-validation approach, the diagnostic performance was based on out-of-bag CT studies that were not available to the random forest model during training phase, which could therefore be considered independent CT samples during the respective cross-validation iteration. The random forest identified tumor diameter, patient age, and RENAL score as the most important imaging features for renal sarcoma identification, highlighting the clinical applicability of the proposed algorithm. In separate analysis using imaging features only, the statistical model still yielded a high diagnostic performance with AUC = 91.9%, thus corroborating the diagnostic importance of imaging features for renal sarcoma identification.

The patient cohorts assessed in this study compares well to the published literature. For example, Wang et al reported a median patient age of 42 years and a predominant leiomyosarcoma histology (39%) in the majority of renal sarcoma cases [6].

Further, the random sample of non-sarcoma renal tumors in our study yielded a distribution of histological subtypes that is in line with earlier studies, reporting clear cell RCC as the most frequent subtype in 75–80% of malignant cases, followed by papillary RCC in 10–15% [2]. In this context, it has to be highlighted that benign AMLs and oncocytomas were purposely included in our study to better reflect diagnostic challenges in assessment of renal masses of uncertain differentiation, such as discrimination of fat-rich AMLs from fat-containing renal liposarcomas. Similarly, sarcomatoid RCCs were purposely sampled to compare imaging features of RCCs which underwent epithelial-mesenchymal transition to those of primary mesenchymal renal sarcomas. Further, the broad inclusion of various non-sarcoma histological subtypes maximized the generalizability and clinical applicability of the random forest algorithm to discriminate between renal sarcomas and non-sarcoma renal tumors.

Renal sarcoma imaging features have so far been described in several case reports and 3 larger reviews [7‐9]. The literature suggests that the imaging features of renal sarcomas vary with the numerous renal sarcoma subtypes. Comparing the most common renal sarcoma subtypes in our study, the only specific imaging features was a higher proportion of macroscopic fat for liposarcomas (75%) versus Ewing sarcoma and leiomyosarcoma (each 0%, p = 0.02). Still, missing statistical significance in imaging feature subgroup analyses might be attributable to the small subgroup sizes resulting from the overall rarity of renal sarcomas.

In general, the literature supports our finding that renal sarcomas present with a larger tumor diameter compared to non-sarcoma renal tumors. For example, Lalwani et al noted an average renal sarcoma size between 55 and 230 mm at time of diagnosis, while Karaosmanoglu et al specified a mean size of 80 mm for renal synovial sarcoma [8, 9]. The locally aggressive growth of renal sarcomas observed in our study is in line with the literature, i.e., reporting inferior vena cava tumor thrombus in patients with renal liposarcomas, fibrosarcomas, leiomyosarcomas, malignant fibrous histiocytoma, and synovial sarcomas [7, 14, 15]. Renal sarcoma patients frequently presented with tumor necrosis in our cohort (73.5%), which has been described to be a common feature of renal sarcomas [16‐18]. Still, the high proportion of continuous renal sarcoma invasion of adjacent organs, such as spleen or liver, detected in 38.2% of patients in this study, has not been reported so far.

Among 34 included renal sarcoma cases, only one case showed imaging features that were previously described as specific for renal sarcomas, demonstrating a capsular growth pattern in a patient with renal liposarcoma [14]. In contrast, imaging features such as tumoral calcification and fat that some authors reported to be imaging hallmarks of renal sarcomas were observed in both renal sarcoma and non-sarcoma renal tumors in this study, thus limiting their discriminatory value [7‐9].

With an AUC = 93.8%, the random forest trained in this study demonstrated a good diagnostic performance, highlighting the potential clinical benefit of this method for accurate renal sarcoma classification. Especially given the low incidence of renal sarcomas and the resulting unfamiliarity of radiologists with associated imaging features, radiological identification of renal sarcomas might be improved by utilization of the proposed random forest algorithm.

In light of the dismal prognosis of renal sarcomas and their propensity to metastasize, an accurate and timely image-based diagnosis might improve patient management. Specifically, the ESMO guideline recommends a risk-adapted perioperative treatment of patients with retroperitoneal sarcoma, including neoadjuvant chemotherapy and radiation, which might play a role in renal sarcomas as well [10]. On the other hand, the role and diagnostic accuracy of renal biopsy in cases with large necrotic tumors, such as renal sarcomas, remains unclear [11]. There are several limitations to this study. First, given the overall rarity of renal sarcomas, the sample size of specific histological subgroups was small, with resulting comparably low statistical power to detect differences in imaging feature subgroup analyses. Second, not all patients received imaging studies according to standardized protocols or three-phasic studies, as recommended for CT assessment of renal masses. This might reduce their interpretability regarding imaging features such as vascularization or tumor necrosis. Third, no external cohort was available to validate the proposed random forest algorithm, which might have resulted in statistical overfitting. Still, a conservative 10-fold external cross-validation was used to minimize potential overfitting, and the accrual of an adequately sized external cohort seems impractical given the rarity of renal sarcomas.

Further, the varying accrual periods of the renal sarcoma and non-sarcoma subgroups might have introduced bias through varying standards in image acquisition, although imaging features analyzed in this study were assessable in all patients from both cohorts.

Finally, there is a lack of studies on renal sarcomas from the last 5 years, and thus, newest epidemiological developments as well as advancements in renal tumor diagnosis and treatment might not be reflected by the cited references.

Renal sarcomas are rare mesenchymal renal tumors that tend to manifest as large, complex renal masses in young patients. A random forest algorithm trained on standardized imaging features and patient demographics shows good diagnostic accuracy for discrimination of renal sarcomas from non-sarcoma renal tumors using a cross-validation approach, though limited by the low number of patients. Tumor diameter and RENAL score were the most relevant imaging features to identify renal sarcomas.

This algorithm might aid in timely and accurate diagnosis of renal sarcomas in clinical practice in an effort to optimize and individualize treatment of patients with renal cancer.