Eur Radiol DOI 10.1007/s00330-015-3662-y
ONCOLOGY
Diffusion-weighted MRI for uveal melanoma liver metastasis detection Mathilde Wagner & Pascale Mariani & François Clément Bidard & Manuel Jorge Rodrigues & Fereshteh Farkhondeh & Nathalie Cassoux & Sophie Piperno-Neumann & Slavomir Petras & Vincent Servois
Received: 18 August 2014 / Revised: 11 December 2014 / Accepted: 11 February 2015 # European Society of Radiology 2015
Abstract Objectives We aimed to assess the sensitivity of diffusionweighted (DW) magnetic resonance (MR) imaging for the detection of pathologically confirmed uveal melanoma liver metastases (UMLM). Methods Twenty patients who underwent complete surgical resection of their UMLM (N=83) were included. Pre-surgery liver MR imaging included T2-weighted, T1-weighted, DW and dynamic-gadolinium-enhanced MR sequences. Two radiologists independently reviewed three sets of images (DW / morphologic-dynamic / combined) for each patient using intraoperative and pathological findings as a standard of reference. Results The sensitivities of the morphologic-dynamic and DW images for UMLM detection were 63 % and 59 %, respectively, for reader #1 (R1) and 64 % and 53 %, for reader #2 (R2). Sensitivity of the combined set was higher than sensitivity in the two other sets (R1:69 %, R2:67 %), but was only significantly different than the sensitivity of the DW images (McNemar test). For the three sets and the two readers, the sensitivity for UMLM smaller than 5 mm (37–46 %) was
significantly lower than that for UMLM larger than 5 mm (67–90 %). The sensitivity for UMLM located in the subcapsular area (41–54 %) was significantly lower than that for intraparenchymal UMLM (68–86 %) (Chi-square test). Conclusion Our study shows that the addition of DW imaging to morphologic-dynamic images does not significantly increase MR sensitivities for UMLM detection. Key Points • The MR imaging sensitivity for uveal melanoma liver metastases (UMLM) was 69 %. • Addition of DW imaging to morphologic-dynamic images does not increase sensitivity significantly. • Sensitivity for subcapsular UMLM was significantly lower than sensitivity for intraparenchymal UMLM. • The T2 shortening effect does not appear to influence lesion detection in DWI.
M. Wagner : S. Petras : V. Servois (*) Department of Radiology and Nuclear Medicine, Institut Curie, 26 rue d’Ulm, 75005 Paris, France e-mail: [email protected]
Abbreviations and acronyms UMLM Uveal melanoma liver metastases MR Magnetic resonance DW Diffusion-weighted PACS Picture archiving communication system
P. Mariani : N. Cassoux Department of Surgery, Institut Curie, 26 rue d’Ulm, 75005 Paris, France
Keywords Uveal neoplasm melanoma . Magnetic resonance imaging . Diffusion magnetic resonance imaging . Hepatic metastasis . Detection
F. C. Bidard : M. J. Rodrigues : S. Piperno-Neumann Department of Medical Oncology, Institut Curie, 26 rue d’Ulm, 75005 Paris, France
Introduction
F. Farkhondeh Department of Pathology, Institut Curie, 26 rue d’Ulm, 75005 Paris, France
Uveal melanoma is the most frequent primary malignant intraocular tumour in adults, with an annual incidence of six cases per million. Distant metastatic spread, which occurs in
Eur Radiol
approximately half of patients, is mainly haematogenous. Liver imaging plays a critical role in the follow-up of patients, because the liver is the most frequent and usually exclusive site of metastases [1]. Despite numerous recent trials with chemotherapy, hepatic intra-arterial chemotherapy or targeted therapy, there is no efficient systemic treatment of uveal melanoma liver metastases (UMLM) to date, with an overall survival not exceeding 14 months [2, 3]. It was previously reported that select patients who underwent complete surgical resection of liver metastases had a median overall survival of 27 months; in contrast, patients with incomplete surgical resection had a median overall survival of 14 months [4, 5]. In this context, the sensitivity of the presurgical imaging routine is critical for detecting UMLM and ultimately enabling complete surgical resection. Over the past few years, magnetic resonance (MR) imaging has become the gold standard for early liver metastasis detection; its superiority over ultrasound and computed tomography imaging has been demonstrated, mainly in colorectal cancer patients [6–8]. A recent study also demonstrated the efficiency of liver MR for UMLM screening [9]. The performance of MR imaging for liver metastasis detection has recently been increased by diffusion-weighted (DW) imaging. The added value of DW imaging for detection of colorectal liver metastases has been extensively reported [7, 10–13], and its value in patients with neuroendocrine tumours was recently demonstrated [14]. DW imaging appears to be particularly valuable for small lesions, i.e., lesions smaller than 1 cm [10, 12]. In contrast to colorectal cancer metastases, UMLM have some particularities: they frequently contain melanin, which is a paramagnetic agent; they are commonly small in diameter (i.e., ≤ 5 mm); and they are preferentially located in the subcapsular area [4]. In a previous study, the surgical discovery of such miliary liver metastases was frequently a limiting factor to complete surgical resection [4]. The performance of MR imaging as part of the presurgery imaging routine for UMLM detection has rarely been studied, although the reported sensitivity is 67 % [15]. To our knowledge, no published study evaluates the added value of DW imaging for UMLM preoperative staging. The aim of our study was to assess the sensitivity and the added value of DW imaging for UMLM detection, and to compare its sensitivity to that of morphologic-dynamic imaging using intraoperative findings and pathological analyses as a standard of reference in patients who underwent complete surgical resection of UMLM.
Material and methods Study population This monocentric and retrospective study was performed in a reference centre for uveal melanoma care. The institutional
review committee approved the study and waived written informed consent. From November 2009 to March 2013, all patients who underwent liver resection for UMLM in our centre were recorded in a prospective database. Patients with a preoperative MR examination including a DW imaging sequence less than 3 months before the surgery and surgical and pathological complete resection (BR0^, resection with no grossly visible tumour and no microscopic residual tumour—microscopically negative margins) were included. Patients with surgical macroscopically incomplete resection (BR2^) were excluded, due to the large number of UMLM not allowing a lesion-by-lesion analysis. The other exclusion criterion was related to the quality of the MR imaging (poor signal-to-noise ratio or motion artefacts). Figure 1 details the flow chart of the study population. Ultimately, 20 patients, eight women [median age=63 years old (47–66)] and 12 men [median age=62 years old (43–69)], were included in the study. The median delay between MR imaging and surgery was 34.5 days (9–79). MR imaging MR imaging was performed on a 1.5 T clinical system (Siemens Symphony, Erlangen, Germany). The liver MR imaging protocol and details of the MR acquisition parameters are shown in Table 1. All sequences covered the whole liver with a total examination time of approximately 30 minutes. MR analysis Two readers with 6 and 20 years of experience in abdominal MR imaging, respectively, independently reviewed three sets of images in three different sessions during a 1-month interval on a PACS workstation (Carestream Health, Version 11.3.2.0220, Mountain View, CA, USA). The first set included only the DW images; the second set, called the Bmorphologic-dynamic set^, included T1-weighted images in-phase and out-of-phase, T2-weighted images and gadolinium chelate-enhanced MR images; the third set, called the Bcombined set^, included all the previous MR images. For each set, the readers were blinded to the results of the other sets and the surgical and pathological findings. For each patient and each set, the number, size and location (according to the Couinaud numbering system) of the potential metastases were noted. In the DW images, a lesion was considered a liver metastasis if it had high signal intensity that was persistent on the highest ‘b’ value. In the morphologic-dynamic images, a lesion was considered a liver metastasis if it exhibited either 1) moderate hyperintensity relative to the liver parenchyma in T2-weighted images, and hypointensity relative to the liver parenchyma in T1-weighted images, together with diffuse or non-globular peripheral enhancement after gadolinium chelate injection, or 2) moderate hyperintensity or hypointensity relative
Eur Radiol Fig. 1 Flow chart of the study
to the liver parenchyma in T2-weighted images and hyperintensity relative to the liver parenchyma in T1-weighted images, regardless of the enhancement. During the last reading session, for all UMLM confirmed by surgical and pathological findings and detected by MR imaging, the two readers reviewed all the images in consensus and noted the signal intensity of each metastasis relative to the liver parenchyma for each sequence (hypointense, hyperintense or not seen in T1, T2 and DW imaging; arterial, portal, delayed or three-phase enhancement). Subtraction images were used systematically. In some cases, subtraction images Table 1
were not contributory due to movement artefacts, and the enhancement was considered to be not available. Liver cyst, hemangioma and focal nodular hyperplasia were diagnosed using the usual criteria [11, 16]. Standard of reference The combination of surgical and pathological findings were retrospectively analysed on a lesion-by-lesion basis and defined as our standard of reference. Surgical findings were reported by a surgeon with 15 years of experience who was
MR acquisition parameters (GE: gradient echo, FSE: fast spin echo, EPI: echoplanar, SPAIR: spectral attenuated inversion recovery, CA: contrast agent)
Sequence
Dual GE T1-weighted FSE T2-weighted imaging imaging
EPI DW imaging
3D GE Contrast-enhanced imaging
Fat saturated Respiratory compensation Acquisition time (min) Repetition time (ms) Echo time (ms) Flip angle (degree) Parallel imaging factor Number of repetitions Field of view (mm) Matrix Section thickness (mm) Intersection gap (mm) Voxel size Others
No Breath hold 0.44 101 2.38–4.76 70 — 1 320×300 (93.8 %) 256×230 (90 %) 6 1.8 1.4×1.3×6
Yes (spair, spectral) Triggered 4–5 ≈1,400 69 90 2 2 350×262 (75 %) 192×173 (90 %) 6 1.8 2×1.8×6 b values=0, 50, 300, 600 s/mm2 Echo planar imaging factor=130
Yes (quick fat sat, chemical shift selective) Breath hold 0.24 5.75 2.7 10 2 1 350×262 (75 %) 384×307 (80 %) 3.5 0.7 1.1×0.9×3.5
Yes (spair, spectral) Triggered 4–5 ≈1,400 90 138 — 2 320×295 (92 %) 512×307 (60 %) 6 1.2 1×0.6×6
Repeated 4 times: Pre-contrast Arterial phase: 20 s Portal phase: 70 s Delayed phase: 180 s Non-hepatospecific CA
Eur Radiol
specialized in UMLM care. Analysis included visual inspection, manual palpation, and intraoperative ultrasound of the liver. Pathological findings included a gross examination of any resected liver tissue sliced every 5 mm, followed by microscopic analysis. Cytological reports on fine needle aspiration biopsies were collected for any liver lesion treated by intraoperative radiofrequency ablation. When available, data concerning UMLM pigmentation were recorded for each lesion. The spatial location of all liver metastases was available in surgical and pathological reports using the Couinaud numbering system. The intraparenchymal and subcapsular location of each UMLM was recorded. A UMLM was defined as subcapsular if it was in contact with the hepatic capsule on the pathological examination and was visible during the surgical inspection. MR images, surgical findings and pathological analysis findings were compared on a lesion-by-lesion basis in the last session, according to the guidelines reported by Obuchowski et al. for the detection of multiple lesions per subject [17]. Finally, all patients had a MR imaging follow-up, every 4 months over a period of at least 1.5 years.
Statistical analysis Results are presented as the mean± standard deviation [range] or median [range] for quantitative data and as Table 2
the number of cases (percentage of cases) for categorical variables. Lesion characteristics (maximal diameter, T1 and T2weighted imaging signal intensity) between achromic and pigmented UMLM were compared using Pearson’s Chisquared test for binary and nonrelated variables. Sensitivity was defined as the number of liver metastases correctly depicted in the MR images divided by the number of liver metastases diagnosed with the standard of reference. The corresponding 95 % confidence intervals were also computed. Sensitivity differences in detection between the three sets of images regarding liver metastasis diameter (≤ / > 5 mm) and location (intraparenchymal or subcapsular / right or left liver) were analysed using Pearson’s Chisquared test for binary and non-related variables. The sensitivities of the three sets were compared using McNemar’s test for binary and related variables. The average number of false positives per patient (total number of false positive lesions / total number of patients) for the three sets were compared using the Wilcoxon matched pairs test. Finally, interobserver agreement for liver metastasis detection was assessed with kappa coefficients (0.00–0.20 indicated slight agreement; 0.21–0.40, fair agreement; 0.41– 0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.00, excellent agreement). Tests were always twosided, with a level of significance set at p
ONCOLOGY
Diffusion-weighted MRI for uveal melanoma liver metastasis detection Mathilde Wagner & Pascale Mariani & François Clément Bidard & Manuel Jorge Rodrigues & Fereshteh Farkhondeh & Nathalie Cassoux & Sophie Piperno-Neumann & Slavomir Petras & Vincent Servois
Received: 18 August 2014 / Revised: 11 December 2014 / Accepted: 11 February 2015 # European Society of Radiology 2015
Abstract Objectives We aimed to assess the sensitivity of diffusionweighted (DW) magnetic resonance (MR) imaging for the detection of pathologically confirmed uveal melanoma liver metastases (UMLM). Methods Twenty patients who underwent complete surgical resection of their UMLM (N=83) were included. Pre-surgery liver MR imaging included T2-weighted, T1-weighted, DW and dynamic-gadolinium-enhanced MR sequences. Two radiologists independently reviewed three sets of images (DW / morphologic-dynamic / combined) for each patient using intraoperative and pathological findings as a standard of reference. Results The sensitivities of the morphologic-dynamic and DW images for UMLM detection were 63 % and 59 %, respectively, for reader #1 (R1) and 64 % and 53 %, for reader #2 (R2). Sensitivity of the combined set was higher than sensitivity in the two other sets (R1:69 %, R2:67 %), but was only significantly different than the sensitivity of the DW images (McNemar test). For the three sets and the two readers, the sensitivity for UMLM smaller than 5 mm (37–46 %) was
significantly lower than that for UMLM larger than 5 mm (67–90 %). The sensitivity for UMLM located in the subcapsular area (41–54 %) was significantly lower than that for intraparenchymal UMLM (68–86 %) (Chi-square test). Conclusion Our study shows that the addition of DW imaging to morphologic-dynamic images does not significantly increase MR sensitivities for UMLM detection. Key Points • The MR imaging sensitivity for uveal melanoma liver metastases (UMLM) was 69 %. • Addition of DW imaging to morphologic-dynamic images does not increase sensitivity significantly. • Sensitivity for subcapsular UMLM was significantly lower than sensitivity for intraparenchymal UMLM. • The T2 shortening effect does not appear to influence lesion detection in DWI.
M. Wagner : S. Petras : V. Servois (*) Department of Radiology and Nuclear Medicine, Institut Curie, 26 rue d’Ulm, 75005 Paris, France e-mail: [email protected]
Abbreviations and acronyms UMLM Uveal melanoma liver metastases MR Magnetic resonance DW Diffusion-weighted PACS Picture archiving communication system
P. Mariani : N. Cassoux Department of Surgery, Institut Curie, 26 rue d’Ulm, 75005 Paris, France
Keywords Uveal neoplasm melanoma . Magnetic resonance imaging . Diffusion magnetic resonance imaging . Hepatic metastasis . Detection
F. C. Bidard : M. J. Rodrigues : S. Piperno-Neumann Department of Medical Oncology, Institut Curie, 26 rue d’Ulm, 75005 Paris, France
Introduction
F. Farkhondeh Department of Pathology, Institut Curie, 26 rue d’Ulm, 75005 Paris, France
Uveal melanoma is the most frequent primary malignant intraocular tumour in adults, with an annual incidence of six cases per million. Distant metastatic spread, which occurs in
Eur Radiol
approximately half of patients, is mainly haematogenous. Liver imaging plays a critical role in the follow-up of patients, because the liver is the most frequent and usually exclusive site of metastases [1]. Despite numerous recent trials with chemotherapy, hepatic intra-arterial chemotherapy or targeted therapy, there is no efficient systemic treatment of uveal melanoma liver metastases (UMLM) to date, with an overall survival not exceeding 14 months [2, 3]. It was previously reported that select patients who underwent complete surgical resection of liver metastases had a median overall survival of 27 months; in contrast, patients with incomplete surgical resection had a median overall survival of 14 months [4, 5]. In this context, the sensitivity of the presurgical imaging routine is critical for detecting UMLM and ultimately enabling complete surgical resection. Over the past few years, magnetic resonance (MR) imaging has become the gold standard for early liver metastasis detection; its superiority over ultrasound and computed tomography imaging has been demonstrated, mainly in colorectal cancer patients [6–8]. A recent study also demonstrated the efficiency of liver MR for UMLM screening [9]. The performance of MR imaging for liver metastasis detection has recently been increased by diffusion-weighted (DW) imaging. The added value of DW imaging for detection of colorectal liver metastases has been extensively reported [7, 10–13], and its value in patients with neuroendocrine tumours was recently demonstrated [14]. DW imaging appears to be particularly valuable for small lesions, i.e., lesions smaller than 1 cm [10, 12]. In contrast to colorectal cancer metastases, UMLM have some particularities: they frequently contain melanin, which is a paramagnetic agent; they are commonly small in diameter (i.e., ≤ 5 mm); and they are preferentially located in the subcapsular area [4]. In a previous study, the surgical discovery of such miliary liver metastases was frequently a limiting factor to complete surgical resection [4]. The performance of MR imaging as part of the presurgery imaging routine for UMLM detection has rarely been studied, although the reported sensitivity is 67 % [15]. To our knowledge, no published study evaluates the added value of DW imaging for UMLM preoperative staging. The aim of our study was to assess the sensitivity and the added value of DW imaging for UMLM detection, and to compare its sensitivity to that of morphologic-dynamic imaging using intraoperative findings and pathological analyses as a standard of reference in patients who underwent complete surgical resection of UMLM.
Material and methods Study population This monocentric and retrospective study was performed in a reference centre for uveal melanoma care. The institutional
review committee approved the study and waived written informed consent. From November 2009 to March 2013, all patients who underwent liver resection for UMLM in our centre were recorded in a prospective database. Patients with a preoperative MR examination including a DW imaging sequence less than 3 months before the surgery and surgical and pathological complete resection (BR0^, resection with no grossly visible tumour and no microscopic residual tumour—microscopically negative margins) were included. Patients with surgical macroscopically incomplete resection (BR2^) were excluded, due to the large number of UMLM not allowing a lesion-by-lesion analysis. The other exclusion criterion was related to the quality of the MR imaging (poor signal-to-noise ratio or motion artefacts). Figure 1 details the flow chart of the study population. Ultimately, 20 patients, eight women [median age=63 years old (47–66)] and 12 men [median age=62 years old (43–69)], were included in the study. The median delay between MR imaging and surgery was 34.5 days (9–79). MR imaging MR imaging was performed on a 1.5 T clinical system (Siemens Symphony, Erlangen, Germany). The liver MR imaging protocol and details of the MR acquisition parameters are shown in Table 1. All sequences covered the whole liver with a total examination time of approximately 30 minutes. MR analysis Two readers with 6 and 20 years of experience in abdominal MR imaging, respectively, independently reviewed three sets of images in three different sessions during a 1-month interval on a PACS workstation (Carestream Health, Version 11.3.2.0220, Mountain View, CA, USA). The first set included only the DW images; the second set, called the Bmorphologic-dynamic set^, included T1-weighted images in-phase and out-of-phase, T2-weighted images and gadolinium chelate-enhanced MR images; the third set, called the Bcombined set^, included all the previous MR images. For each set, the readers were blinded to the results of the other sets and the surgical and pathological findings. For each patient and each set, the number, size and location (according to the Couinaud numbering system) of the potential metastases were noted. In the DW images, a lesion was considered a liver metastasis if it had high signal intensity that was persistent on the highest ‘b’ value. In the morphologic-dynamic images, a lesion was considered a liver metastasis if it exhibited either 1) moderate hyperintensity relative to the liver parenchyma in T2-weighted images, and hypointensity relative to the liver parenchyma in T1-weighted images, together with diffuse or non-globular peripheral enhancement after gadolinium chelate injection, or 2) moderate hyperintensity or hypointensity relative
Eur Radiol Fig. 1 Flow chart of the study
to the liver parenchyma in T2-weighted images and hyperintensity relative to the liver parenchyma in T1-weighted images, regardless of the enhancement. During the last reading session, for all UMLM confirmed by surgical and pathological findings and detected by MR imaging, the two readers reviewed all the images in consensus and noted the signal intensity of each metastasis relative to the liver parenchyma for each sequence (hypointense, hyperintense or not seen in T1, T2 and DW imaging; arterial, portal, delayed or three-phase enhancement). Subtraction images were used systematically. In some cases, subtraction images Table 1
were not contributory due to movement artefacts, and the enhancement was considered to be not available. Liver cyst, hemangioma and focal nodular hyperplasia were diagnosed using the usual criteria [11, 16]. Standard of reference The combination of surgical and pathological findings were retrospectively analysed on a lesion-by-lesion basis and defined as our standard of reference. Surgical findings were reported by a surgeon with 15 years of experience who was
MR acquisition parameters (GE: gradient echo, FSE: fast spin echo, EPI: echoplanar, SPAIR: spectral attenuated inversion recovery, CA: contrast agent)
Sequence
Dual GE T1-weighted FSE T2-weighted imaging imaging
EPI DW imaging
3D GE Contrast-enhanced imaging
Fat saturated Respiratory compensation Acquisition time (min) Repetition time (ms) Echo time (ms) Flip angle (degree) Parallel imaging factor Number of repetitions Field of view (mm) Matrix Section thickness (mm) Intersection gap (mm) Voxel size Others
No Breath hold 0.44 101 2.38–4.76 70 — 1 320×300 (93.8 %) 256×230 (90 %) 6 1.8 1.4×1.3×6
Yes (spair, spectral) Triggered 4–5 ≈1,400 69 90 2 2 350×262 (75 %) 192×173 (90 %) 6 1.8 2×1.8×6 b values=0, 50, 300, 600 s/mm2 Echo planar imaging factor=130
Yes (quick fat sat, chemical shift selective) Breath hold 0.24 5.75 2.7 10 2 1 350×262 (75 %) 384×307 (80 %) 3.5 0.7 1.1×0.9×3.5
Yes (spair, spectral) Triggered 4–5 ≈1,400 90 138 — 2 320×295 (92 %) 512×307 (60 %) 6 1.2 1×0.6×6
Repeated 4 times: Pre-contrast Arterial phase: 20 s Portal phase: 70 s Delayed phase: 180 s Non-hepatospecific CA
Eur Radiol
specialized in UMLM care. Analysis included visual inspection, manual palpation, and intraoperative ultrasound of the liver. Pathological findings included a gross examination of any resected liver tissue sliced every 5 mm, followed by microscopic analysis. Cytological reports on fine needle aspiration biopsies were collected for any liver lesion treated by intraoperative radiofrequency ablation. When available, data concerning UMLM pigmentation were recorded for each lesion. The spatial location of all liver metastases was available in surgical and pathological reports using the Couinaud numbering system. The intraparenchymal and subcapsular location of each UMLM was recorded. A UMLM was defined as subcapsular if it was in contact with the hepatic capsule on the pathological examination and was visible during the surgical inspection. MR images, surgical findings and pathological analysis findings were compared on a lesion-by-lesion basis in the last session, according to the guidelines reported by Obuchowski et al. for the detection of multiple lesions per subject [17]. Finally, all patients had a MR imaging follow-up, every 4 months over a period of at least 1.5 years.
Statistical analysis Results are presented as the mean± standard deviation [range] or median [range] for quantitative data and as Table 2
the number of cases (percentage of cases) for categorical variables. Lesion characteristics (maximal diameter, T1 and T2weighted imaging signal intensity) between achromic and pigmented UMLM were compared using Pearson’s Chisquared test for binary and nonrelated variables. Sensitivity was defined as the number of liver metastases correctly depicted in the MR images divided by the number of liver metastases diagnosed with the standard of reference. The corresponding 95 % confidence intervals were also computed. Sensitivity differences in detection between the three sets of images regarding liver metastasis diameter (≤ / > 5 mm) and location (intraparenchymal or subcapsular / right or left liver) were analysed using Pearson’s Chisquared test for binary and non-related variables. The sensitivities of the three sets were compared using McNemar’s test for binary and related variables. The average number of false positives per patient (total number of false positive lesions / total number of patients) for the three sets were compared using the Wilcoxon matched pairs test. Finally, interobserver agreement for liver metastasis detection was assessed with kappa coefficients (0.00–0.20 indicated slight agreement; 0.21–0.40, fair agreement; 0.41– 0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.00, excellent agreement). Tests were always twosided, with a level of significance set at p
