Multiparametric magnetic resonance imaging for characterizing renal tumors: A validation study of the algorithm presented by Cornelis et al.

INTRODUCTION

As a result of the increased and widespread use of cross-sectional imaging as computed tomography (CT) and magnetic resonance imaging (MRI), the number of incidental renal masses has increased.^[1,2] These incidental renal masses vary from benign cysts to renal cell carcinomas (RCCs), having variable aggressiveness and metastatic progression potential.^[3] RCC occurs in different forms, of which clear cell RCC (ccRCC) is the most common, followed by papillary and chromophobe RCCs (chRCC).^[4] The subtypes of RCC exhibit different biologic behavior in terms of growth rate and treatment response, as well as different prognoses.^[5,6] It is therefore crucial to determine the specific subtype to determine the optimal therapeutic approach.

Current diagnostic guidelines for differentiating subtypes include a histopathological examination. Recent studies have suggested the use of multiparametric MRI combined with the administration of a contrast agent as a diagnostic method capable of differentiating between benign and malignant renal tumours.^[7-11] If proved accurate, multiparametric MRI of renal tumors could provide new in vivo information regarding tumor type, decreasing the need for percutaneous histopathological biopsy and thereby decreasing the risk for resulting complications.^[12-17]

In 2017, Cornelis and Grenier presented a practical algorithm for classifying renal tumors, differentiating between RCC and benign subtypes using multiparametric MRI.^[18] The algorithm consists of five individual MRI sequences designed to classify ccRCC, papillary RCC (pRCC), chRCC, oncocytoma (OC), and angiomyolipoma (AML). These sequences are T2 weighted, dual chemical shift MRI, diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) T1 in both the wash-in and wash-out phases. According to Cornelis et al., the sequences must be read in this specific order to differentiate one subtype from another, as well as for a standardized approach. The algorithm is based solely on a theoretical review of published papers and has not previously been investigated in a clinical and research setting. The aim of this study was therefore twofold: To collect validity evidence by testing MRI sequences according to Cornelis et al.’s algorithm regarding the classification of renal tumor subtypes, and to compare those results with the post-operative histopathological results. The hypothesis was that the MRI protocol could correctly classify a majority of the subtypes.

MATERIAL AND METHODS

Patients

Consecutive patients in a fast-track program for the investigation of renal tumors were prospectively recruited from the hospital’s urology outpatient clinic. Patients were eligible if all of the following criteria were met:

• - Presentation of suspected renal tumor either found by computed CT or ultrasound, and

• - Age ≥18 years, and

• - Normal kidney function defined as estimated glomerulus filtration ratio ≥60 mL/min/1,73 m², and

• - Informed consent understood and provided.

Exclusion criteria were: Pregnancy, graft kidney, contrast allergy, patients with non-MRI-compatible components, or the inability to meet a patient’s fast-track guarantee for diagnosing renal tumors within a stated time frame.

Included patients were offered an additional MRI examination with the experimental sequences on the same day of the biopsy but before the invasive procedure. [Figure 1] presents the flowchart of patients included in the study. Histopathological diagnosis, obtained either from a subsequent percutaneous biopsy or surgical radical or partial nephrectomy, served as the gold standard.

Close

Figure 1:

Patient flowchart.

Export to PPT

MRI

All MRI examinations were performed using a clinical 3.0 T system (Ingenia, Phillips Healthcare, Best, Netherlands) with a 32-channel large anterior surface body coil combined with a posterior moveable coil. The respiration triggering device was placed under the anterior coil at the diaphragm level. All patients underwent a survey in two plans with the respiration trigger and two plans with breath-hold, respiration triggered T2-weighted (T2w) turbo spin-echo sequence in transversal and coronal plan, a fat-suppressed diffusion-weighted imaging (DWI) sequence in the transverse plane, and T1-weighted (T1w) DIXON sequence performed both before and 25 s after administration of a gadolinium-containing agent for corticomedullary phase as well as at 120 s after administration for the nephrographic phase. The MRI protocol is presented in [Table 1]. The contrast agent was injected rapidly through an antecubital vein and immediately followed by a flush of 30 mL saline solution (0.9% NaCl). Gadovist (Bayer, Leverkusen, Germany) was used as a contrast agent and administered at 0.1 mL/kg, and a maximum of 7.5 mL was injected. For all MRI sequences, sensitivity encoding was used. [Figure 2] presents the sequences used in the assessments according to Cornelis’ proposed algorithm.

Table 1: MRI sequence parameters for RCC protocol.

MRI sequences
MRI protocol	BTFE	DWI	T1-Weighted DIXON	T2 MVXD	T2-Weighted
Plane	Coronal	Transverse	Transverse	Transverse	Coronal
Fat saturation	No	Yes	No	No	No
TR (ms)	3.5	1940	3.7	2725	4960
TE (ms)	1.75	83	1.31	135	80
Angulation (°)	45	90	10	90	90
Thickness (mm)	8	4	3	4	3
Matrix (mm×mm2)	351×229	400×200	405×199.5	400×199	499×118.5
Scanning time (s)	28	297	18	306	49
Delay (s)			0, 25, 120
b-values (s/mm2)		0, 150, 1000

RCC: Renal cell carcinoma, MRI: Magnetic resonance imaging

Close

Figure 2:

Magnetic resonance imaging sequences assessed in the algorithm including T2-weighted, Dual chemical shift, Diffusion-weighted imaging (DWI), Dynamic contrast-enhanced (DCE) T1-weighted in in-and out-phase.

Export to PPT

Before receiving histopathology results and diagnosis, two radiologists (OG and JTA), with 9 and 14 years MRI experience in the interpretation of renal tumors, assessed and described in consensus the images in the prescribed order. For each tumor, the radiologists evaluated tumor appearance, one sequence at a time, according to Cornelis et al.’s algorithm [Table 2].^[18] If an MRI diagnosis was established after 1–4 sequences, all remaining sequences were still evaluated. For example, high signal intensity in the first sequence (T2w) suggests cc-RCC or OC, [Table 2]. If the tumor shows a signal drop on the out-phase on the second sequence (dual chemical shift MRI), cc-RCC or AML is suggested. The combination of the first two sequences excludes all remaining tumor types, and cc-RCC is proposed as the diagnosis. In another scenario, a tumor might present intermediate/middle intensity during the first sequence, suggesting Ch-RCC. In these instances, images from all remaining sequences were still evaluated.

Table 2: The practical MRI algorithm proposed by Cornelis et al.

Order	Sequences	Imaging features	Results
1	T2w	Signal intensity⍉	High cc-RCC/OC	Mid Ch-RCC	Low AML/p-RCC
2	Dual chemical shift MRI	Signal drop on out-phase	Yes cc-RCC/AML	-	No OC/p-RCC
3	DWI	ADC⍑	High OC/cc-RCC	Mid Ch-RCC	Low AML/p-RCC
4	DCE-T1w	Wash-in	Fast and intense cc-RCC/AML	Mid and delayed Ch-RCC/OC	Slow p-RCC
5	DCE-T1w	Wash-out	Yes cc-RCC/AML	Mid Ch-RCC/OC	No P-RCC

⍉Compared to renal cortex, ^⍑Apparent diffusion coefficient, T2W: T2 weighted images, ccRCC: Clear cell renal cell carcinoma, OC: Oncoytoma, ChRCC: Chromophobe renal cell carcinoma, AML: Angiomyolipoma, pRCC: Papillary renal cell carcinoma, DWI: Diffusion-weighted imaging, ADC: Apparent diffusion coefficient, DCE: Dynamic contrast enhanced, T1w: T1-weighted, MRI: Magnetic resonance imaging

Tumor size was measured in two dimensions on the transverse T2w image: medial-lateral and anterior-posterior, where the largest diameter was noted.

RESULTS

DISCUSSION

In this study, the diversity of both benign and malign renal tumors’ MRI patterns in different sequences was prospectively explored regarding the practical algorithm proposed by Cornelis and Grenier.^[18] Remarkably, we were unable to reproduce the results from the original paper and were unable to correctly differentiate between the subtypes of renal tumors using the multiparametric MRI sequence algorithm. To the best of our knowledge, this is the first prospective study attempting to validate an MRI protocol for differentiating between renal tumor subtypes. Cornelis et al. recommend a structured step-by-step process of five sequences to differentiate between renal tumor subtypes using the resulting images. The sequences include: (1) Signal intensity compared to the renal cortex in the T2w sequence, (2) signal drop in out-phase of the dual chemical shift MRI sequence, (3) ADC calculation from images acquired with the DWI sequence, (4) wash-in analysis of a DCE T1w sequence, and (5) wash-out analysis of a DCE T1w sequence.

In vivo characterization of tumors is of paramount importance, and image-guided characterization would decrease the use of percutaneous biopsies and the resulting complications. In our prospective study, the multiparametric MRI sequences correctly classified only eight tumors; therefore, the results do not provide sufficient evidence to recommend relying on MRI as a diagnostic imaging technique as opposed to invasive procedures and histopathological examinations. A major challenge to the algorithm, and subsequently a challenge to the conduction of our study, is the indefinite description of the categoric classification in each tumor subtype’s profiles. The algorithm does not propose any specific or quantitative measures, for example, defining ADC values as high, middle, or low; or characterizing a DCE-T1w wash-in phase as fast and intense or mid and delayed. As a result, the assessments can be affected by a subjective judgment or institutional cultural differences, among other factors.

Several studies have recently suggested different approaches for using multiparametric MRI to differentiate between renal masses, either as a dichotomic outcome (benign versus malignant) or by distinguishing between benign or malignant subtypes.^[21-24] The content of these approaches varies from 3 to 11 MRI sequences and assessments, and includes both categoric and dichotomic assessments. Kay et al. also begin with an assessment of signal intensity compared to the renal cortex in the T2-weighted non-fat suppressed images, but unlike Cornelis et al.’s algorithm, the isointense or medium signal intensity does not narrow down the possibilities of the subtypes to one specific type.^[24] According to Cornelis et al.’s algorithm, an isointense or medium signal intensity using T2w denotes a subtype of Ch-RCC, and in our study, a total of 11 tumors were categorized as Ch-RCC after this first sequence assessment. Kay et al. only excluded AML in the isointense assessment of the T2w sequence, and ccRCC, which accounts for a majority of the malignant tumor subtypes, can have all three categoric appearances (high, isointense, and low), although most commonly showing high or isointense T2w patterns. This finding supports previous publications describing ccRCC’s presentation using MRI but contradicts Cornelis’ algorithm.^[25,26]

A limitation of not only Kay et al., but most of the studies proposing MRI as a diagnostic tool for differentiating tumor subtypes, is the retrospective design and the results being presented as sensitivity, specificity, positive predictive values, and negative predictive values.^[24,27-30] Most of the studies initiate the conduct of their study with the histopathological results and then divide the MRI assessment and features based on the histopathological findings, instead of providing prospective assessments. This design could affect and change the prevalence of the different subtypes which will affect the positive and negative predictive values. In additiom, the sensitivities and specificities are calculated and presented for each tumor subtype, which is relevant for research purposes but is of limited interest in a clinical setting. Our prospective validation and cohort were not designed to explore sensitivities or specificities of incorrect diagnoses with MRI as a diagnostic test and histopathological examination as the gold standard.

Another issue that needs to be addressed when considering implementing image-guided characterization of tumor subtypes is interrater reliability. In our cohort, the assessments were done in agreement between two radiologists, and in the study, we did not explore inter- or intra-reader reliability. Several studies examine the interrater reliability or agreement (between observers), or intraobserver agreement (within observers) for ADC values or assessment of hypo-, iso-, or hypointense tumor profiles.^[31-33] The literature suggests that such agreement varies from moderate to substantial, with the highest kappa values for ccRCC and pRCC.^[24] When exploring a new diagnostic method with the potential to decrease the number of invasive procedures, the inter- and intraobserver agreement needs to be sufficient to avoid errors, and diagnostic accuracy must be explored not only for subspecialists with research and/or clinical interest in MRI subtyping but for the common radiologist working with MRI and renal tumors.

No clear guidelines or recommendations exist for the use of MRI in the diagnostic work-up, staging, or follow-up.^[34-36] The ESMO guideline for clinical practice for diagnosis, treatment, and follow-up of RCC states “MRI may provide additional information in investigating local advancement and venous involvement by tumor thrombus.”^[34] In clinical practice, MRI plays an important role in the treatment of patients where contrast-enhanced CT is contraindicated; however, exploiting the highest potential of MRI also requires intravenous contrast.

Despite the small sample size (n = 30), the frequency of the RCC subtypes reported elsewhere (ccRCC: 80–90%, pRCC: 10–15%, and chRCC: 4–5%^[37]) was similar to the findings in this study (ccRCC: 80.0%, pRCC: 13.3%, and chRCC: 3.3%). Literature reports a 1.5:1 predominance for men over women to be diagnosed with a renal tumor, with peak incidence occurring between 60 and 70 years of age.^[37] This study had a male: female ratio of 2.4:1 and a mean age of 66 years. Despite this difference, this study’s prospective collected sample was considered a representative sample of the true population. Three patients were excluded from the study population due to inconclusive histopathology from biopsy. The European Association of Urology guidelines for RCC from 2019 presented a similar range for non-diagnostic biopsies of 2.5–22%,^[36] which again demonstrated that the study sample was representative of the true population.

We present the first prospective validation study exploring the Cornelis et al. algorithm for the use of multiparametric MRI to differentiate between renal tumors. Despite being unable to reproduce the results, the findings are relevant for future work and progress on using MRI for image classification of renal tumors. This study does have several limitations. Radiographers reported the time from contrast injection to manually initiating the time of T1w DIXON as between 25 s and 120 s. Ideally, this interval could have been standardized if manual monitoring had been avoided. Secondly, the small sample size of only 44 renal tumors could also limit the results of this study. However, the distribution of subtypes was consistent with the epidemiologically reported frequencies. Thirdly, according to the literature, there was a potential risk of misdiagnosing when relying solely on core biopsy, where only 86–98% are diagnostically accurate.^[38,39] Excisional biopsy material was available in most cases (n = 29) in this study, but in cases where the core biopsy diagnosed a benign renal tumor (n = 6) or for the tumors receiving cryoablation (n = 15), the diagnosis depended solely on the core biopsy and therefore there existed a potential risk of misdiagnosing. This was especially true concerning OC s, where biopsies are not always able to determine exact diagnosis, but only point in the direction of oncocytic neoplasm. However, as all histopathology reports were re-evaluated by the same pathologist, the risk of misdiagnosing was considered limited, and histopathology could still be considered the gold standard.

CONCLUSION

We aimed to prospectively validate a multiparametric MRI algorithm for classifying renal tumors but could not reproduce the results from the original study. The MRI algorithm shows few promising results to categorize renal tumors, and histopathological examinations are still needed for clinical decisions and follow-up regimes of renal masses.