Diagnostic Performance of MRI Diffusion Weighted Heterogeneity Index to Discriminate Brain Metastases from Normal Brain Tissue in Hybrid PET/MR

To evaluate diagnostic performance of apparent diffusion coecient (ADC) heterogeneity index to differentiate brain metastasis (BM) from normal appearing brain parenchyma (NABP) and to nd out the correlation between 2-[ 18 F]-uoro-2-deoxy- D -glucose ( 18 F-FDG) standardized uptake value (SUV) and ADC heterogeneity index derived from hybrid PET/MRI. Methods Whole-body PET/MRI was performed to evaluate proven 40 BM of 18 oncology patients (9 females, 9 males; mean age 61±16 years), sourced from different primary cancer. Brain sequences, which were dixon and diffusion weighted imaging (DWI) protocols with simultaneous PET were used to calculate coecient of variance of the ADC (ADC CV ) and SUVmax. All images were assessed by three radiologists and the same size of VOI was placed on BM and NABP. Inter-rater reliability was tested by inter-class correlation (ICC). The correlation of ADC CV and SUVmax and the differences in ADC values and SUVmax between BM and NABP were investigated. The excellent consistency was found between raters at ADCmean (0.972) and ADC CV (0.995). There was a strong correlation between ADC CV and SUVmax (r=0.763) and a slight inverse correlation between ADCmean and SUVmax (r=-0.122). A statistically signicant difference between BM and NABP was determined for ADC CV (p<0.001) and SUVmax (p<0.001). An area under the curve (AUC) of 0.960, 0.998 and 0.574 were obtained with ROC analysis of SUVmax, ADC CV and ADCmean, respectively. CV may be considered as a potential biomarker that quantitatively discriminates BM from NABP with excellent interrater reliability.


Introduction
Brain metastasis (BM) is associated with poor survival outcomes and poses distinct clinical challenges.
Lung cancer, renal cell carcinoma, breast cancer, melanoma and colorectal cancers are the most common causes of BM [1]. Due to great variation in imaging appearances, these metastases present a common diagnostic challenge which can affect patient management.
Computed tomography (CT) and magnetic resonance imaging (MRI) are the key imaging modalities used in the diagnosis of BM. In some cases, advanced imaging techniques including proton magnetic resonance spectroscopy (MRS), contrast enhanced magnetic resonance perfusion (MRP), diffusion weighted imaging (DWI), and diffusion tensor imaging (DTI) may help for the diagnosis [2]. Although these imaging techniques are essential in the diagnosis, using quantitative data may lead to improved detection of BM.
DWI is a fast, non-contrast MR technique that indicates the random microscopic motion of free water molecules. It is widely appreciated as a qualitative tool in the examination of the central nervous system (CNS). Apparent diffusion coe cient (ADC) is a measure, calculated using DWI and re ects the magnitude of diffusion quantitatively.
Tumors are heterogeneous because of the spatial variation in the cellularity, angiogenesis, extracellular matrix and necrosis [3,4]. Higher intratumoral heterogeneity is related with poor prognosis due to its aggressive behavior [5][6][7]. Thus, measuring of tissue heterogeneity may be helpful in the detection of tumors and selection poor prognostic patients for more intensive therapy. There are various methods using complex textural analysis in the detection of tissue heterogeneity [8]. Of all these, the coe cient of variation (CV) is easily calculated and shows relative variability. In line with this, ADC CV as a reliable heterogeneity index was used in different studies [9][10][11][12].
Positron Emission Tomography (PET)/MRI is a new imaging technology that allows for PET and MRI scans to be acquired simultaneously. Although MRI is the standard neuroimaging technique for detection of tumors and the surrounding anatomical structures in the brain, PET aids to complement MRI in lesion grading, tumor extent delineation and evaluation of the treatment response. Allowing both structural and functional evaluation of tumors in one single scan makes PET/MRI more popular in oncology imaging.
The primary target of our study was to determine the diagnostic performance of ADC CV , as a heterogeneity index, to differentiate BM from normal appearing brain parenchyma (NABP), as compared to conventional MRI metrics used in daily routine. A secondary target of this research was to investigate correlation of ADC CV and standard uptake value (SUVmax) measured on PET-MRI. To the best of our knowledge, this is the rst study that evaluates the diagnostic performance of ADC CV in brain metastases and its correlation with SUVmax on PET/MRI hybrid system.

Material And Methods
Study Design 347 consecutive adult patients with known malignancies who underwent PET/MRI between January 2017 and September 2019 were evaluated. Forty-ve patients who had BM were enrolled in this retrospective single center study. The patients who has multiple lesions (if there is no enough NABP), a massive brain edema, a history of radiotherapy and no 2-[ 18 F]-uoro-2-deoxy-D -glucose ( 18 F-FDG) uptake were excluded from the study. Decision of BM was given if lesions growth at least two imaging methods in the follow-up imaging (3-6 months) or proven with biopsy (single lesion). Thus, 40 lesions of 18 patients were included and analyzed for this study (Fig. 1). All primary malignancies were proven histopathological by biopsy or surgery.

Image acquisition
Patients fasted at least 6 hours before starting examination and injection of 18 F-FDG was given if blood glucose levels were < 140 mg/dL (7.77 mmol/L). All scans were performed with the patient in the supine position on the 3 Tesla Biograph mMR scanner (Siemens Healthcare, Erlangen, Germany) using a 16channel head and neck surface coil and three 12-channel body coils and the total scanning time was 60±3 minutes. The whole-body images, which cover the entire body from head to heel, were obtained in ve to six bed positions according to body-mass index (BMI) of patient. PET attenuation correction was performed using four compartment model attenuation map calculated from a Dixon-based volumetric interpolated breath-hold examination (VIBE) sequence. The MRI protocol included sequences as below: T1-weighted slice-selective Turbo Flash (TR/TE, 1600 msec / 2.5 msec) in the axial plane, free breath diffusion-weighted imaging using the echo planar imaging technique (EPI) (TR/TE, 12000 msec / 78 msec, b=0 s/mm 2 and 800 s/mm 2 ) in the axial plane and T2-weighted single-shot echo train (HASTE) (TR/TE, 1500 msec / 87 msec) in the coronal plane. Contrast enhanced protocol including the breath-hold 3D VIBE sequence (TR/TE, 4.56 msec / 2.03 msec) in the arterial, portal venous and equilibrium phases covering whole-body in the axial plane was performed with using a weight-adapted gadolinium-based contrast agent and all sections were then combined.

Image analysis
All image datasets were transferred to the dedicated Syngo.via PET/MRI workstation (Siemens Healthcare) and images were assessed separately by three radiologists (İ.H.S, B.K.S and N.İ.G) with at least 6 years of experience who were blinded to the patients' information. A volume of interest (VOI) was placed manually on axial PET images and all three planes were controlled for ensuring to not over ow the limits of the lesions. The VOI was coregistered and placed on ADC images overlapping with PET images. Manual correction was used to ne tune when the images were not overlapped. For each determined lesion, a similar size of VOI was used on NABP (Fig. 2). Care was taken to keep away from edema, blood vessels and cerebrospinal uid and for preventing bias, white matter, which did not include sulcus, was used to evaluate NABP. SUVmax (SUV of the hottest voxel within a de ned VOI), which is easy to use and operator independent, was calculated automatically and measured on PET images. The mean (ADCmean) and standard deviation (SD) of ADC (ADC SD ) were calculated automatically by software for each measurement. ADC CV was created by dividing the SD by the ADCmean.
Statistical analysis IBM Statistical Package for the Social Sciences (SPSS ver. 21 for windows, Chicago, IL, USA) software was used for all statistical analysis. Intra-class correlation coe cient (ICC) was used for determining inter-rater reliability in ADCmean, ADC SD and ADC CV . The ICC value ² 0.50, 0.51-0.75, 0.76-0.90 and > 0.90 were evaluated as indicating poor, moderate, good and excellent reliability, respectively. The tness of numeric data set to normal distribution was determined by the Shapiro-Wilk test. Due to non-normal distribution, correlation between SUVmax and ADC CV was tested by Spearman's rank correlation.
Wilcoxon signed rank test was carried out to measure differences between BM and NABP for all variables. Receiver operating characteristics (ROC) analysis based on histopathological results was performed to determine cut-off value, which differentiate BM from NABP, by the Youden index. The area under the curve (AUC), sensitivity and speci city were calculated for each variable. A p-value 0.05 was accepted as statistically signi cant.

Interrater reliability
There was an excellent consistency between raters at ADCmean, ADC SD , ADC CV

Correlation with SUVmax
For all values, the mean of three raters was calculated and presented as ADCmean, ADC CV and SUVmax.
According to Spearman's correlation coe cient, there was a strong correlation (r=0.763, p<0.000) between ADC CV and SUVmax when all measurements included (BM+NABP). A slight inverse correlation was found between ADCmean and SUVmax (r=-0.122).

Differences between BM and NABP
A statistically signi cant difference between BM and NABP with p <0.001 value was found for ADC CV and SUVmax. There was no statistically signi cant difference for ADCmean (p=0.253). The mean±SD values of ADCmean, ADC CV and SUVmax of all lesions were presented in the table (table 2).

Discussion
In this study, we investigated the role of ADC CV derived from PET/MRI as a heterogeneity index in discriminating BM from NABP. The main nding of this study was that, ADC CV is more effective to discriminate BM from NABP compared to conventional ADC parameters. Besides, ADC CV was highly correlated with SUVmax from simultaneous derived PET/MRI system. Integrated PET / MRI scanners with the recent developments, new opportunities have emerged for quantitative molecular imaging. PET / MRI provides multimodal analysis of concurrently acquired functional parameters that can contribute to better characterization of tumor biology and also help identify markers to predict response to therapy [13,14].
Due to the high 18 F-FDG uptake of the cerebral cortex and the low spatial resolution of PET imaging, FDG PET/CT imaging has limitations, especially in the detection of small metastases in the brain. Sensitivity of FDG PET/CT in brain imaging is low. In retrospective comparative studies, it is stated that FDG PET/CT imaging at the time of diagnosis can capture up to 61% of metastatic brain lesions that can be detected by MRI [15]. Therefore, PET / MR imaging may be preferred in brain metastasis scanning because of the high soft tissue contrast of MR imaging.
SUVmax measured by PET and ADC measured by MRI allow assessment of water diffusion and glucose metabolism in tumor cells. The results of the present study show that ADC CV exhibits an improved correlation with SUVmax. Moreover, it provides better quantitative separation between BM and NABP, as compared to common MRI metrics.
In this study, ADCmean showed a signi cant negative correlation with SUVmax; however, ADC CV showed higher correlation with SUVmax than ADCmean parameter. There are previously reported oncologic studies of the inverse correlation found between ADC and SUV. Several of these studies reported signi cant strong inverse correlation between ADCmean and SUVpeak in rectal cancer [16], a signi cant inverse correlation between ADCmean and SUVmean in gastrointestinal stromal tumor [17], and recently an inverse correlation between ADC and PET SUV in liver tumors [18].
Tumor heterogeneity consists of marked differences in cell mix, size, and arrangement. Heterogeneity also plays a role in micro-environmental factors (including oxygenation, pH, interstitial pressure, blood ow), metabolism, and gene expression. This deep heterogeneity is extremely important for prognosis, treatment planning, and drug distribution, which ultimately affects patient outcomes. There are a number of ways to investigate tumor heterogeneity, some of which include functional and molecular imaging, which can be applied to clinical data [19].
The characterization of tissues can be improved using histogram-based assessments of the distribution of ADC values. Histogram approaches have multiple advantages, including volume-of-interest (VOI) assessments, thus avoiding the subjectivity that is inherent with ROI placements. Importantly, histograms can provide additional metrics that re ect the texture of lesions, thereby allowing heterogeneity of ADC distribution within tissue to be assessed [20]. Tissue heterogeneity analysis is rapidly evolving by various methods. Despite most of the tools currently offered are often complex and computationally costly, it is an easy to calculate texture parameter of ADC CV . Several studies have used CV as an index of heterogeneity in recent years.
In a study in liver metastasis, the results of this study show that ADC CV can signi cantly distinguish between liver metastasis and normal-appearing liver [9]. Similar to our study, there was a good correlation between ADC CV and SUV peak in this study. Signi cant differences in CV diffusion index was found in another study about hepatocellular carcinoma in fresh liver explants [21].
PET / CT and DWI share applications in clinical oncology. While both SUV and ADC correlate with cellularity, SUV is also associated with several other pathological markers such as mitotic count, presence or absence of necrosis [22]. For this reason, PET / MRI oncological evaluation is also valuable when these two parameters (SUV and ADC) are obtained together in the same examination. In a study by Nakajo et al. [23], 44 patients with breast cancer underwent preoperative PET/CT and DWI within an average of 17 days between examinations, and both SUVmax and ADC were signi cantly associated (p < 0.05) with histologic grade (independently), nodal status, and vascular invasion. This nding suggests that SUVmax and ADC correlate with several of the pathologic prognostic factors and that both values may have the same potential for being predictive of the prognosis of breast cancer.
In oncology, imaging has a very important place in evaluating response to treatment. For this reason, many studies are aimed at understanding the structure and heterogeneity of the tumor. Therefore, it is essential to develop quantitative imaging methods and objective biomarkers to improve the diagnosis of brain metastasis. As a volume-independent index of heterogeneity, ADC CV can be considered as a potential biomarker that quantitatively differentiates BM from NABP. Tissue heterogeneity has been proposed as a basis for a biomarker for tumors [3,4,24].
This hybrid PET / MRI study shows a signi cant negative correlation between metabolic activity on 18 F-FDG PET and water diffusion over DWI in brain metastasis, possibly because both parameters are directly related to tumor cellularity. The correlation found between SUVs and ADCmean, ADC CV values supports the idea that high cellularity due to tumor proliferation results in greater metabolic activity and restricts water diffusion.
This study has several limitations. First, this was a retrospective study and performed on a relatively small study population. Another limitation was the di culty in determining the limits of the lesions due to the limited resolution of PET. The accuracy of the results obtained from our study should be supported by using different software in larger patient groups and with multi-center studies. The last limitation of our study was that brain metastases originated from different sources.

Conclusions
In conclusion, using PET/MRI instead of PET/CT decreases radiation dose, however radiation exposure caused by short term follow-up imaging of oncology patients continue to be an issue. Although determining brain metastases compared to normal brain parenchyma are not the main challenge in oncologic patients, ADC CV may be helpful to clinicians for avoiding further radiation exposure of patients and for managing patients when using contrast media is contraindicated. Moreover, it would be easy to implement ADC CV in a clinical setting. Future studies that will blindly and independently identify BM in NABP using PET 18 F-FDG SUV and DWI ADC CV will present potential to investigate ADC CV as a biomarker for BM.

Declarations
Funding: Tables   Table 1 Primary source and histopathology of metastases and distribution of the study population.