Diagnostic Performance of SPECT/CT Versus Diffusion-Weighted Mri in Characterization of Indeterminate Osseous Lesions Detected on Bone Scan

Background: Characterization of solitary osseous lesions on planar bone scintigraphy (PBS) is challenging. In this pilot study, we aimed to explore the diagnostic performance of 99m Tc-MDP bone scintigraphy using SPECT/CT in comparison to magnetic resonance imaging including diffusion-weighted imaging (MR-DWI) in patients with solitary indeterminate osseous lesions detected on PBS. Methods: This pilot study prospectively recruited 46 cancer patients who underwent PBS, with a nding of a solitary osseous lesion that was deemed indeterminate by two nuclear medicine physicians not involved in subsequent reading. A targeted SPECT/CT and MR-DWI were read independently by two nuclear medicine physicians and two radiologists, respectively, on a 5-point probability score. The nal diagnosis of disease status was formulated from subsequent clinical/imaging follow-up within six months. Results: Agreement between SPECT/CT readers was 0.73 (95% condence interval [CI]:0.51-0.85) and 0.99 (95%CI:0.99-1.00) between MR-DWI readers. lesions depicted on PBS. MR-DWI results were more reproducible; however, the overall accuracy of MR-DWI varied according to lesion site, density patterns on CT, and the primary tumor type. Such differences were less obvious with SPECT/CT. osteosarcoma, Ewing’s sarcoma, medullary thyroid cancer, gall bladder carcinoma, renal cell carcinoma, spindle cell


Introduction
Bone is the third most common site of cancer metastases. The involvement of the skeleton mostly changes the tumor staging, the overall prognosis, and could radically alter the management plan [1].
Imaging is the cornerstone for detecting bone involvement in cancer patients. Planar bone scintigraphy (PBS) using diphosphonate compounds is the most commonly used modality to survey the whole skeleton for metastases [2]. Although PBS scanning is a sensitive tool, it frequently lacks speci city, and inconclusive reporting is fairly common, especially when a solitary osseous lesion is depicted [3]. Many benign etiologies can cause this nding; examples include degenerative, arthritic, or post-traumatic changes. The diagnostic algorithm of such lesions frequently involves follow-up bone scanning or targeted cross-sectional anatomical imaging (CT and/or MRI) to resolve the underlying pathology [4].
The introduction of hybrid SPECT/CT machines integrates anatomical mapping and morphological characterization with the functional tracer distribution in one study. Skeletal SPECT/CT has been shown to overcome many of the limitations encountered with PBS leading to superior diagnostic output [5][6][7][8][9][10][11].
However, the uptake of phosphonate bone-seeking tracers re ects only the sites of osteoblastic activity and bone formation [12]; accordingly, it does not adequately accumulate in early bone marrow deposits, which typically do not provoke a secondary osseous bone response. Furthermore, SPECT/CT machines are not widely available, and the procedure could be very expensive.
MRI is known to be more sensitive than CT or radionuclide bone scans for detecting early marrow-based metastases through its superior contrast and better spatial resolution [13,14]. On the other hand, the classic morphological MR sequences may fail to accurately differentiate malignant from benign lesions as the signal characterization may overlap. The emergence of functional imaging, together with the advances in MR hardware and software, lead to improved visualization and quanti cation of the diffusion of water molecules across biological membranes and diffusion-weighted (DW) MRI became more readily available sequence for assessment of bone marrow lesions [1][2][3]. The assumption is that malignant soft tissue lesions with hypercellularity and reduced extra-cellular matrix would display restricted mobility of water molecules and lower mean apparent diffusion coe cient (ADC, calculated in units of mm 2 /sec) [15][16][17][18][19][20]. Bone marrow is peculiar in many aspects, and the malignant marrow lesions show high ADC values than the normal bone marrow [6].
In this prospective study, we evaluated the diagnostic performance of 99m Tc-MDP bone scintigraphy using SPECT/CT in comparison to MRI, including diffusion-weighted imaging (MR-DWI), in patients with solitary indeterminate osseous lesions detected on planar bone scintigraphy (PBS).

Participants
The Institutional Review Board approved this prospective study. During the period from November 2014 to February 2016, consecutive patients with a known primary tumor, referred to perform planar bone scintigraphy (for staging or re-staging/follow-up), were considered eligible for this study if their PBS showed a solitary osseous lesion, which was deemed "indeterminate" based on PBS by consensus reading of two physicians not involved in this research study. Accordingly, the patient underwent a targeted SPECT/CT acquisition on the same day and MR-DWI within one month from the date of bone scanning.
We excluded patients with anterior rib lesions because the respiratory movement, which is more appreciated anteriorly, could degrade MR-DWI acquisitions [21]. Patients who cannot perform or have contraindications for MRI examination, as those having claustrophobia, cardiac pacemaker, or metallic prosthesis not compatible with MRI, were also excluded.
Planar bone scintigraphy and SPECT/CT imaging PBS was performed using a hybrid SPETC/CT machine (Symbia T, Siemens Medical Solutions, Erlangen, Germany), 2-4 hours after IV injection of 20-25 mCi (740-925 MBq) 99m-technetium methylene diphosphonate ( 99m Tc-MDP). The dose was modi ed according to Webster's rule for pediatric patients [22]. The patients were asked to lie supine with arms down. Anterior and posterior projection images of the whole skeleton were acquired using a parallel-hole low-energy high-resolution collimator at a speed of 12 cm/min. The images were acquired into a matrix size of 256 × 1024 with an energy window centered on the 99m Tc 140 keV photopeak (±15%).
One-bed SPECT was acquired for the region of concern as detected on PBS. SPECT settings were 128 x 128 matrix over 360° circular arc using 64 stops and 25 seconds per stop. Immediately after SPECT acquisition, low-dose CT images were obtained using a tube current of 80 mA and a voltage of 130 kV.
The average radiation dose from low dose CT was 4 mGy±2.5. The total SPECT/CT acquisition time was approximately 20 min. CT images were reconstructed using a high-resolution reconstruction algorithm (B80 kernel) with 1-mm slice thickness.
Attenuation-and scatter-corrected SPECT images were reconstructed using the manufacturer's iterative protocol (Flash 3D) with 4 iterations, 4 subsets, and 8 mm Gaussian lter. Subsequently, tomographic slices were generated and displayed as transaxial, coronal, and sagittal slices. SPECT emission images were co-registered and fused with CT images and visually inspected for any misregistration.
MRI imaging MR studies were conducted using a 1.5 Tesla MR machine (Magnetom Avanto, Siemens Medical Solutions, Erlangen, Germany) using an external coil array with subjects examined in the supine position. The different acquired planes and their parameters are summarized in Table 1.  Image interpretation and follow-up Images from SPECT/CT were reviewed by two nuclear medicine physicians (R1; 5-year experience and R2;12-year experience). Areas with increased or focal uptake that does not conform with the normal distribution, without clear CT evidence of benign underlying etiology were considered malignant. Bone destruction on CT was also indicative of malignancy. All MRI images, including DWI, were reviewed by two radiologists (20-year experience and 19-year experience). Malignant lesions were de ned on MRI as lesions that had low signal intensity on T1WI, iso-or high-signal intensity on T2WI, restricted diffusion, and high signal intensity on STIR sequence. Readers were aware of the patient's age and primary tumor type but were blinded to other clinical/imaging data, including the images from PBS. Accordingly, each lesion was evaluated according to a 5-point scale: score 1 = de nitely benign, score 2 = probably benign, score 3 = equivocal, score 4 = probably malignant and score 5 = de nitely malignant.
The true status of the lesion(s) was formulated as positive or negative based on subsequent clinical/imaging follow-up within six months. On imaging studies, size increase or changes in the character (osteolytic to sclerotic) were considered positive for malignancy. Stationary lesions without therapy were considered benign. In the presence of interval treatment, lesions that showed a response/stable course were considered malignant.
During analysis, lesions with scores 1 or 2 were considered benign, while those scored 3-5 were considered malignant. Equivocal ndings (score 3) were considered to the malignant side. The reason for that 'pessimistic' classi cation is that a report with equivocal ndings would commonly necessitate further intervention (e.g., more imaging and closer follow-up).

Statistical analysis
Kappa test was used for assessing the degree of agreement between the readers. True-positive (TP), truenegative (TN), false-positive (FP), and false-negative (FN) readings were identi ed based on subsequent follow-up. Diagnostic performance parameters were calculated in the form of sensitivity, speci city, accuracy, positive predictive value (PPV), and negative predictive value (NPV). To limit the number of comparisons presented here, we compared only the results from the more experienced reader for each modality. The non-parametric McNemar's test was used to evaluate the statistical signi cance of the differences in sensitivity and speci city. The area under the curve (AUC) from receiver operating characteristic (ROC) analysis was used to demonstrate the overall accuracy of each modality using the 5points score. One radiologist measured the mean ADC value, and ROC analysis with 1000-sample bootstrapping of the Youden index was used to elicit the predictive accuracy of different mean ADC cutoff values in differentiating benign from malignant lesions.
Data were summarized as percentages, mean, and standard deviation. Con dence intervals were reported where applicable. In all analyses, P-value < 0.05 (two-tailed) was considered statistically signi cant. The analysis was performed using SPSS v.21 (IBM Corp, Armonk, New York) and MedCalc (MedCalc Software, Mariakerke, Belgium).

Patients
A total of 48 eligible patients were initially enrolled in this study. One patient was excluded due to lack of follow-up data. The other was excluded because the detected lesion on PBS was proved to be extraosseous on SEPCT/C; accordingly, did not perform MR-DWI. A valid cohort of 46 patients was available for analysis (32 females, 14 males), with a median age of 52 years (range: 10-77 years). The median interval between SPECT/CT and MRI studies was one week. Forty-four patients (92%) had their MRI study performed within one month from SPECT/CT.
All the follow-up methods/procedures were performed at the discretion of the referring physician. The nal diagnosis of bone lesions was obtained from subsequent imaging follow-up after a mean duration of 5.3 ± 2.8 months in 20 patients (18 with repeated bone scanning and SPECT/CT, 1 with MRI, and 1 with X-ray,). Bone biopsy was performed in one patient only. The clinical course of the patient, as evaluated by the treating physician, after a mean follow-up duration of 5.0 ± 1.6 months, was considered the reference standard in the remaining 25 patients. Of these 25 patients, both bone SPECT/CT and MR-DWI were concordantly reported negative in 14 and positive in 5 patients. Thirty-one patients had no interval therapy between their initial scanning and follow-up, 12 had systemic therapy (hormonal or chemotherapy), and 3 had radiotherapy.
According to the nal follow-up, the reference status of the 46 patients was classi ed as positive for osseous metastasis in 18 patients (39%), while the remaining 28 patients (61%) were considered diseasefree.

Agreements within modalities
Both radiologists perfectly agreed on describing the pattern of lesions in T2WI and STIR, and also on the presence or absence of soft tissue component (k = 1.0). In one patient only, there was no agreement on describing T1WI (k = 0.97; 95%CI: 0.9-1.0) or the presence of restriction on DWI (k = 0.95; 95%CI: 0.87-1.0). There was an identical positive/negative categorization between MRI readers (k = 1.0).

Diagnostic performance of SPECT/CT and MR-DWI
Diagnostic performance indices are given in Table 2. Overall, SPECT/CT performed similar to MR-DWI with AUC of 0.77 (95%CI: 0.62-88) and 0.71 (95%CI: 0.55-0.83); respectively (Fig. 1). The difference between the two curves was not signi cant (p = 0.6). When the diffusion pattern is considered separately, restricted diffusion resulted in slightly lower sensitivity with comparable speci city when compared to the overall impression from all MRI sequences, including DWI. Lesions without restricted diffusion were mostly benign (20/26); however, only 8/19 lesions that showed restricted diffusion were malignant (p = 0.03). In one patient, no de nite lesion was seen on MRI, and a separate evaluation of each of the MRI sequences, including DWI, was not possible. However, the study overall was considered negative.  The performance of the two modalities was affected by the primary tumor site, lesion site, and CT pattern ( Table 3). The overall accuracy was slightly lower for patients with breast cancer, vertebral metastasis, or non-sclerotic lesions. The numbers in parenthesis are the 95% con dence interval MR-DWI = magnetic resonance with diffusion weighted imaging

Primary tumor site
Breast cancer was seen in 54% of this cohort (n = 25). Other tumor sites included: urinary bladder cancer (n = 6), lymphoma (n = 2), and lung cancer (n = 2). Other primary tumors encountered were osteosarcoma, Ewing's sarcoma, medullary thyroid cancer, gall bladder carcinoma, renal cell carcinoma, spindle cell tumor, liver, prostate, endometrial, pancreatic and tongue cancers (n = 1 each). Since breast cancer represented more than half of our cohort, a sub-analysis of this group might be relevant. Osseous metastasis was seen in 11/25 patients with breast cancer and 6/21 patients with non-breast cancer. Overall accuracy (Table 3) and all other diagnostic indices were slightly better for both modalities in the non-breast cancer group.
Vertebral lesions were encountered slightly more frequently among breast cancer group; however, the overall distribution of appendicular and axial lesions did not vary signi cantly according to primary tumor type or lesion nature on CT (Table 4).
MR-DWI was more accurate for lesions located in the appendicular skeleton compared to axial lesions ( Table 3). The performance was comparable when the spine was considered against other axial and appendicular sites. On the other hand, SPECT/CT did not show signi cant variability in the overall accuracy according to the lesion site ( Table 3).

Nature of the lesions on CT
The spectrum of CT ndings seen in the lesions that were proved benign included vertebral wedging, hemangioma, fractures, degenerative changes, spondylodiscitis, arthritis, bone cyst, and bone island. No CT ndings were encountered in 10 benign lesions and one malignant lesion, which was proved to be due to malignant marrow in ltration with no obvious cortical component.
Lesions were sclerotic (n = 22), lytic (n = 6) or mixed (n = 7). Six of the 11 false results on MRI were sclerotic on CT; 3 of these 6 lesions were located in the spine; 2 of them were secondary to vertebral wedging (1 FN, 1 FP), and one falsely interpreted as Schmorl's nodule.

Analysis of different MRI sequences and mean ADC
The pattern of qualitative MRI features associated with malignancy was mostly hypointense signal on T1WI (52%), hyperintensity on STIR (65%), and restricted diffusion (65%).
ADC value was not predictor of lesion status with AUC of 0.58 (95%CI = 0.42-0.72; p = 0.4). The analysis identi ed a mean ADC value of ≤ 0.8x10 − 3 mm 2 /s as the most discriminative cut-off value. This point resulted in good speci city of 74%, albeit with modest sensitivity of 50%.

Discussion
Multimodality evaluation of equivocal osseous lesions can potentially overcome the limitations of each imaging modality and favorably impact lesion characterization. However, access to these imaging services may not be available at all geographic locations. Therefore, reaching a practical institutional algorithm to diagnose these lesions could save resources and improve outcomes.
In this study, we evaluated the diagnostic performance from SPECT/CT and MRI with diffusion-weight imaging (DWI) for characterizing the nature of solitary osseous lesions that were classi ed as indeterminate based on conventional PBS.
Two raters independently evaluated each SPECT/CT study. In contrast to the study of Sharma et al. [23], the overall performance of SPECT/CT reported in our study was lower and also the level of agreement.
Their study focused on lung cancer patients only with a high prevalence of metastatic osseous disease (70%) compared to 37% in our study. Disease prevalence could impact diagnostic performance indices [24]. Also, treating equivocal lesion as malignant (pessimistic analysis) could shift the values of sensitivity and speci city [25,26].
Furthermore, the difference in years of experience in reading nuclear medicine studies in general and SPECT/CT, in particular, could lead to variable results between centers. On the other hand, the two radiologists demonstrated perfect agreement on scoring MR-DWI studies.
In this study, no signi cant difference could be seen between the two modalities for any of the diagnostic performance indices. To our knowledge, no prior studies directly compared MR-DWI and SPECT/CT in the proposed clinical setting where PBS detects solitary equivocal lesions. Previous reports, performed in a different clinical context, compared MR-DWI to PBS and documented comparable results between the two modalities [27]. One large study [28] compared whole-body scans using DWI and SPECT in addition to other modalities (conventional MRI and PET/CT) in 115 patients with lung cancer. The reported speci city and accuracy of DWI alone were lower than those of SPECT for 1025 sites (93.7% vs. 95.0 and 93.9% vs. 95%, respectively). However, the addition of conventional WB MRI resulted in signi cantly higher speci city (96.1%) and accuracy (96.1%) over SPECT alone. In the same work, the addition of DWI improved overall accuracy when compared to either conventional MRI or DWI alone [28]. These results, though not directly comparable to our study, may suggest that isolated DWI may not be superior to SPECT. In a meta-analysis [18] which included data from 32 studies with 1507 patients, the pooled sensitivity, speci city, and AUC of MR-DWI were 0.95, 0.92 and 0.98 on a per-patient basis, and 0.91, 0.94, and 0.97 on a per lesion basis. There was no statistically signi cant difference found in the sensitivity and speci city of using DWI only and DWI combined with other morphological or functional imaging sequences for either analysis.
We did not perform an independent evaluation of DWI alone for discriminating benign versus malignant features. It has been recommended that DWI should be interpreted with other imaging sequences because benign lesions may result in false positives [29], and certain malignant lesions, such as sclerotic bone metastases, can lead to false-negative results [30,31]. However, if restricted diffusion alone were used as a criterion for malignancy in our study, the sensitivity and speci city would be 63% (10/16) and 72% (20/28), respectively.
In the current work, 6/11 false MRI results were sclerotic on CT part of the SPECT/CT, with half of them located in the spine (3/6). Common causes of false results on MRI include hematopoietic islands (e.g., secondary to use of hematopoietic growth factors), benign compression fractures (e.g., osteoporotic), lesions con ned to the cortex only [32], and sclerotic lesions especially after therapy [33,34]; as a response to treatment may result in edema, the return of marrow fat, brosis, and sclerosis [35]. It worth noting that nearly 25% of our study participants were receiving some sort of systemic therapy at the time of this study.
However, our best discriminatory ADC cut-off value falls within the previously reported range [15,16,36,37]. Furthermore, the quantitative assessment of ADC solely via its mean value, as performed in the current report, might not yield a proper discriminating criterion between benign and malignant lesions, and obtaining multiple quantitative metrics is encouraged.
The discrepancy of ADC values and different cut-off values is likely attributed to the choice of b-value, methods of fat suppression, and the bone marrow cellular density and matrix composition of the bony lesions. For example, osteolytic lesions are better detected on DWI due to a higher content of cellularity and water within the lesions [1,2,6,7]. In our study, the employed b-factors were 50, 400,800 s/mm 2 , and the generated ADC values were quite different as we did not employ b0 as in the previous work by Messiou et al. [2]. Also, the age of our study population had a wide range (10-70 years), which caused variations in bone marrow constituents and distribution (e.g., red vs. yellow bone marrow). The yellow bone marrow is characterized by low cell density, an abundance of fat cells, reduced water protons, and hydrophobic nature of fat in addition to lower bone perfusion compared to the red bone marrow [6].
However, there are overlapping morphological signal characteristics between yellow bone marrow and neoplastic tissue [4]. Tumor cells in ltrating bone marrow displace fat cells, thus increasing cellular and water proton density, and augmenting blood perfusion; accordingly, malignant yellow bone marrow lesions show high signal intensity on DWI, but higher ADC values than normal yellow bone marrow. In contrast, soft tissue malignant lesions show high signal intensity on DWI and lower ADC values than the corresponding normal tissue [6].
Our current results concurred with the conclusions from many prior works, which emphasized the multiple constraints for the application of DWI in the evaluation of oncological patients, including, the high cellularity lesions such as sclerotic lesions, red/yellow bone marrow matrix, malignant lesions with less evident on the normal background and treatment with bone marrow stimulating factors as Granulocyte colony-stimulating factor (G-CSF) [2,6,7]. The currently-reported unsatisfactory performance of DWI supports the conclusion of Grankvist et al. [7] who concluded that DWI cannot replace T1 and STIR sequences but should be added to improve diagnostic e cacy Our study has multiple limitations, including a small sample size, which reduced the power of our conclusions and made sub-group analysis more challenging. Because there were no prior reports directly comparing the two modalities in this speci c clinical setting, the results from our exploratory study could be used to power and inform future studies. Another disadvantage is the inclusion of patients was based on bone scan ndings; we understand that bone scan may lack adequate sensitivity for detecting small early marrow lesions and those with pure lytic nature; however, we aimed to reproduce the commonly utilized clinical scenario, in which bone scintigraphy is the rst performed screening modality for detecting osseous metastasis, then cross-sectional imaging is added for equivocal or indeterminate ndings.
Another limitation is that clinical MRI protocols may vary by the region scanned. Also, we used the pattern from all MRI sequences, including DWI, to reach a diagnostic score, hence, isolating the impact of adding DWI was not possible with the current study design. Furthermore, MRI has more advantages in evaluating soft tissues even in the same procedures, which was not evaluated in this study.
A common disadvantage of many imaging studies that discuss the problem of indeterminate osseous lesions is the absence of histopathological validation in the majority of patients. Bone biopsy for the sole purpose of research could be both technically challenging and ethically unjusti able. Further larger prospective studies are needed to address the current limitations and con rm our ndings.
In conclusion, we performed a prospective comparison between MR-DWI and SPECT/CT for characterizing solitary indeterminate osseous lesions seen on standard planar bone scans. MR-DWI results were more reprodicuble among readers. However, no signi cant difference in diagnostic performance was depicted between the two modalities. The overall accuracy of MRI varied according to the lesion site, the pattern on CT, and the primary tumor type. Declarations