Intravoxel Incoherent Motion for Differentiating Spinal Metastasis: A Pilot Study

Background: To investigate the value of an intravoxel incoherent motion (IVIM) MRI for discriminating spinal metastasis from tuberculous spondylitis. Methods: This study included 50 patients with spinal metastasis (32 lung cancer, 7 breast cancer, 11 renal cancer), and 20 tuberculous spondylitis. All patients underwent IVIM MRI at 3.0T before treatment. The IVIM parameters including single-index model ( Apparent diffusion coecient (ADC)-stand), double exponential model (ADC slow (cid:0) ADC fast and f) and stretched-exponential model parameters (distributed diffusion coecient (DDC) and α) were acquired. Two radiologists separately measured these parameters for each lesion through drawing region of interest. Receiver operating characteristic (ROC) and the area under the ROC curve analysis was used to evaluate the diagnostic performance. Each parameter was substituted into the Logistic regression model to determine the meaningful parameters, and the combined diagnostic performance was evaluated. Results: The ADC fast and f showed signicant differences between spinal metastasis and tuberculous spondylitis. (for all, p < 0.05). The Logistic regression model results showed that ADC fast and f were independent factors affecting the conclusion (P<0.05). The AUC values of ADC fast and f were 0.823 (95%CI:0.719 to 0.927) and 0.876 (95%CI: 0.782 to 0.969), respectively. ADC fast combined with f showed the highest AUC value of 0.925 (95%CI: 0.858 to 0.992). Additional signicant differences were found in ADC stand , ADC slow, DDC and α among different metastasis type. Conclusions: IVIM MR imaging may be helpful for differentiating spinal metastasis from tuberculous spondylitis and may be used to detect the origin tumor for those patients who could not identify primary tumors, and provide help for clinical treatment.


Background
Osteoporosis, trauma, infection, or tumors are typically causes of clinically common diseases of the spine. Its diagnosis mainly depends on imaging ndings and clinical features. But benign and malignant lesions in the spine may have similar imaging ndings, especially in the early stages of the disease [1] . In addition, with the increasing incidence of cancer in modern society, the diagnosis of spinal bone lesions becomes more and more di cult. Spine is a site of metastasis in 10-15% of cancers, which is the third most common site for cancer cells to metastasize [2] and spinal metastases are the most common tumors of the spine [3] . The most common cancer that causes metastases to the spine is lung cancer, breast cancer and renal cancer. The major differential diagnoses for spinal metastases include spinal tuberculosis. Spinal metastases and tuberculosis with typical and speci c imaging ndings are relatively easy to diagnose correctly. However, most of the clinical presentation between them are usually not characteristic, leading to misdiagnosis. Besides, the image features of different diseases have some overlap. Therefore, an effective differential diagnosis method to identify different lesions in the spine is of great signi cance in clinical practice [4] . In addition, for patients with spinal metastases, it is important to treat the primary lesion as well as the metastatic tumor. The treatment of metastatic tumors from different sources is also different, so it is possible to infer the type of primary lesion based on the imaging ndings of spinal metastatic tumors, which may provide assistance for clinical treatment.
Because the clinical course, prognosis and treatment strategy of different types of spinal metastatic and spinal tuberculosis differ greatly, the differential diagnosis of them is di cult and should be accurate diagnosis to improve patients' quality of live.
Intravoxel incoherent motion (IVIM) is a noninvasive MRI method to visualize microscopic motions of water, refers to translational motion that presents a velocity direction and/or amplitude distribution within a given voxel and over a measured time [5] . Water molecular diffusion and perfusion-related diffusion can be achieved by allowing monitoring the extraction of molecular diffusion coe cients (ADC slow ), the perfusion-related diffusion (ADC fast ), and perfusion fraction (f). Recently, it has been used to estimate tissue water diffusivity and micro-capillary perfusion. Perfusion MRI based on IVIM, which does not require contrast agents, has been widely used in studies of oncology. IVIM has been reported for glioma grading, tumor diagnosis, central nervous system and abdomen since its development over the past years [6][7][8][9][10] .
It was reported that IVIM may help to differentiate malignant spinal tumors from acute vertebral compression fractures and tuberculous spondylitis [4] and IVIM-DWI could make for the early diagnosis of ductal carcinoma in situ of the breast, and reduce the misdiagnosis and over-treatment of benign lesions [11] . IVIM and conventional radiologic features improve preoperative assessment of microvascular invasion in patients with hepatocellular carcinoma [12] . IVIM-DWI-derived parameters, especially the pseudo diffusion coe cient, were related to tumor grade and stage in rectal cancer patients, and the difference between subjects with extramural vascular invasion and those without extramural vascular invasion was statistically signi cant. IVIM-DWI derived parameters would help to predict tumor aggressiveness and prognosis [13] . IVIM MRI quantitatively measures local microvascular muscle perfusion to detect muscle activation patterns through walking and running [14] . Zhu G et al. reported that IVIM with these b values (0, 50, 200, 1000 ) collects diffusion and perfusion information from a single short MRI sequence, which may have important implications for the imaging of acute ischemic stroke patients [15] . Accordingly, our aim was to investigate the IVIM parameters in spinal diseases and to discriminate spinal metastasis from tuberculous spondylitis and to investigate the feasibility of tumor type differential diagnosis of different metastases, including lung cancer, breast cancer and renal cancer.

Patients
This retrospective study was approved by the standards of the local ethics committee, with waiver of informed consent granted. This study included 50 patients with spinal metastasis (32 lung cancer, 7 breast cancer, 11 renal cancer), and 20 tuberculous spondylitis from Peking Third Hospital (Beijing, China) with histopathological diagnosis. All patients underwent IVIM MRI at 3.0T before treatment and puncture.

MRI acquisitions
Scanning was performed using a 3.0T MRI system (GE Healthcare 750) and 8-channel spinal coils.

Regions of interest
The region of interest (ROI) was set in the center of the lesion with an area of 30-100 mm 2[16] and the solid components of the tumor were included as much as possible avoiding the lesion edge, necrosis, bone cortex, and blood vessels. The analysis was performed independently by two radiologists (initials: EZ, XX) with more than 5 years of experience on the GE AW4.5 workstation, and the average of the ROI measurements for each parameter was taken as the nal measurement result.
Post-processing and MR Imaging analysis All data were transferred to an imaging workstation for analysis. The IVIM parameters including singleindex model ( Apparent diffusion coe cient (ADC)-stand), double exponential model ( ADC slow ,ADC fast and perfusion fraction (f)) and stretched-exponential model parameters (distributed diffusion coe cient (DDC) and intravoxel water diffusion heterogeneity (α)) were acquired. Two radiologists separately measured these parameters for each lesion through drawing region of interest. Receiver operating characteristic (ROC) and the area under the ROC curve (AUC) analysis was used to evaluate the diagnostic performance.
The ADC value was obtained by using high and low b values monoexponentially tted to the following where Sb is the signal intensity observed in the absence of a diffusion gradient. The formula for the biexponential model was calculated as follows: , where S0 is the mean signal intensity, ADC slow is the molecular diffusion coe cient, ADC fast is the perfusion-related coe cient, and f is the perfusion fraction. The formula for the stretched-exponential model is as follows: Sb = S0 × exp (-b × DDCα), where DDC is a distributed diffusion coe cient, and α re ects the diffusion heterogeneity of water molecules in voxels.

Statistical analysis
All statistical analyses were performed using SPSS version 22.0 (SPSS, Inc., Chicago, IL, USA). Continuous variables are expressed as the mean ± standard deviation. The intraclass correlation coe cient (ICC) was used to analyze the agreement between the two radiologists and was interpreted as

Logistic regression model
The variables with statistical signi cance in univariate analysis and those that were considered to have in uence on the outcome were included in the multivariate Logistic regression model, and the results showed that ADC fast and f were independent variables affecting the conclusion (p < 0.05) ( Table 4).

ROC-analysis
In the ROC-analysis for differentiation of spinal metastasis and tuberculous spondylitis, AUC values of ADC fast and f were 0.876 (95% CI, 0.782 to 0.9969) and 0.823 (95% CI, 0.719 to 0.927), respectively. The sensitivity and speci city of ADC fast value to differentiate spinal metastasis and tuberculous spondylitis were 80.0% and 85.0%. The sensitivity and speci city of f value were 86.0% and 65.0%. ADC fast combined with f showed much higher AUC than ADC fast and f, which the AUC values were 0.925 (95% CI, 0.858 to 0.992). By using the ADC fast combined with f as a discriminative index for the discrimination of spinal metastasis from tuberculous spondylitis, the sensitivity and speci city were 94.0% and 80.0% (Table 5,  Fig3 ).

Discussion And Conclusions
IVIM method is a diffusion-weighted MRI sequence used to estimate perfusion parameters, which has several advantages over commonly used methods. It is a non-invasive alternative to perfusion measurement, eliminating the need for intravenous injection of exogenous contrast agent through a single image sequence, reducing examination time. In addition, its signal is highly spatially speci c because it comes primarily from a place where measurements are taken independently of the arterial blood ow path before reaching there. Finally, it provides additional information compared to ASL, and the combination of the two approaches can be used for the assessment of neurological diseases [17] .
It recently reported that the IVIM parameters could help to more precisely assess the early diagnosis and differentiation of diseases, as well as quantitatively monitor the effectiveness of treatment for tumors and other diseases. Ding Y etl. [18] compare the diagnostic values of IVIM, conventional DWI, and diffusion kurtosis imaging (DKI) in differentiating between benign and malignant renal tumors. They found the D value is the best parameter for differentiating cell renal cell carcinoma (ccRCC) from benign renal tumors. The f value is the best parameter for differentiating non-ccRCC from benign renal tumors. IVIM parameters are the best, while DWI and DKI parameters have similar performance in differentiating malignant and benign renal tumors. Another article found that rectal cancers with different KRAS mutation statuses had distinctive diffusion/perfusion characteristics. D values were lower in the KRAS mutant group. A higher D* value was demonstrated in the KRAS mutant group. IVIM MRI may potentially help preoperative KRAS mutant status prediction [19] . Zhu L etl. [20] evaluated the performance of tumor size and IVIM-derived parameters in predicting long-term prognosis, found that IVIM MR imaging has great potential in predicting long-term prognosis in patients with advanced cervical cancers treated with concurrent chemo-radiotherapy.
In early or atypical spinal tuberculosis, there is no typical obvious bone erosion or abscess, and the imaging ndings are complex and sometimes similar to tumors, leading to misdiagnosis. While this is not common, it is still necessary to nd ways to avoid it. Within the spinal column, metastasis is more commonly found in the thoracic region, followed by the lumbar region, while the cervical region is the least likely place professionals nd metastasis. The aim of this study is to evaluate the performance diagnostic values of IVIM MRI for differentiating spinal metastasis. To the best of our knowledge, our investigation is the rst to illustrate the IVIM relative parameters for differentiation of spinal tuberculosis and spinal metastasis.
Our results shown that some IVIM parameters could help discriminate spinal metastasis from tuberculous spondylitis and could investigate the feasibility of tumor type differential diagnosis of different metastases, including lung cancer, breast cancer and renal cancer. Conventional DWI is based on the micro-movement of water molecules, which re ects the speed of water diffusion in the tissue. ADC values can quantitatively evaluate tissue diffusion and show microscopic changes at the cellular level caused by pathophysiological changes. In our study, no signi cant difference in ADC values was found between spinal metastases and tuberculous spondylitis, suggesting an ADC overlap in distinguishing them. Therefore, the results show that the role of traditional ADC DWI in differentiating benign and malignant lesions of the spine is limited, which is consistent with previous studies [4 , 21-22] .
It is known that the ADC slow is mainly affected by water molecule diffusion of the lesion tissue and ADC fast is mainly affected by capillary microcirculation perfusion. Our results showed the ADC fast value of spinal metastases was signi cantly higher than that of tuberculous spondylitis, which suggests that perfusion greatly increases. However, no signi cant differences were found between the ADC slow values between them. But the ADC slow value of spinal metastasis with lung cancer was signi cantly lower than that of spinal metastasis with breast cancer as well as with renal cancer, suggesting that true diffusion is more restricted in lung cancer than breast cancer and renal cancer. ADC fast and f values except for ADC slow value showed better diagnostic performance than ADC value for differentiating spinal metastasis and tuberculous spondylitis. The f value is mainly affected by the blood volume of microcirculation perfusion, re ecting the proportion of microcirculation perfusion in tissue diffusion. Our results showed that f value of spinal metastasis was lower than that of tuberculous spondylitis group. Similar ndings have been found in nasopharyngeal, pancreatic, and cervical lymph node metastases in previous studies [23][24] . The reasons might be the high cell density of spinal metastasis, the microvessels in the intercellular stroma were compressed, and the new vessels in malignant lesions were compressed, deformed and branched disorderly, leading to the decrease of the proportion of microperfusion components. In our study, ADC fast combined with f showed much higher AUC than ADC fast and f. these nding suggested that ADC fast combined with f was more valuable for the differential diagnosis of spinal metastasis and tuberculous spondylitis.
The DDC and α value re ects the heterogeneity of diffusion in the tissue. The value of α is set between 0 and 1. The closer the α value is to 1, the higher the homogeneity of the diffusion component is; the closer the α value is to 0, the higher the heterogeneity of the diffusion component is and the more complex the diffusion component is. In this study, the DDC and α value of spinal metastasis was lower than that of tuberculous spondylitis. however, there was no signi cant. indicating that compared with tuberculous spondylitis, malignant tissue is more complex and heterogeneous, which leads to the decrease of DDC and α value. Additional signi cant differences were found in the ADC stand , ADC slow , DDC and α among different metastasis type. The reasons for the above results may be related to the pathophysiological changes of the disease and the theory behind different imaging techniques.
Although with novel ndings, our study also has some limitations, which we would further improve in the future work. Firstly, the small sample size of spinal metastasis and tuberculous spondylitis. Secondly, All values were measured by manual outlining ROI, and the ROI was placed on the solid components of the tumor to calculate the average value. Although it was representative to some extent, it was not conducive to the evaluation of tumor heterogeneity. Thirdly, IVIM technology itself is not stable. So far there is no speci c standard about multiple b value of IVIM sequences, and the calculations still need further exploration [25] .
In conclusion, IVIM MR imaging may be helpful for differentiating spinal metastasis from tuberculous spondylitis and provide help for clinical treatment.