Diffusion-weighted MRI for differential diagnosis of pulmonary tuberculosis-like lesions:A pilot study

Background To investigate the value of 3T magnetic resonance imaging (MRI) diffusion-weighted and intro-voxel incoherent motion imaging (DWI-IVIM) in differential diagnosis of tuberculosis-like lesions. Methods 78 patients with 99 lesions comprised of 36 tuberculosis (TB), 38 pulmonary malignancy (PM), 25 non-tuberculosis benign lesions (NTB) underwent 3T-MRI examination, including star-volumetric interpolated breath-hold examination (Star-VIBE), fast-BLADE turbo spin echo (fBLADE TSE), DWI-IVIM (b = 0,30,50,80,100,200,400,600,800,1000 s/mm2). Signal ratio (SR) of lesions to rhomboid muscle, apparent diffusion coecients (ADCs) and IVIM D, D* and f values were compared using Mann-Whitney U test. The sensitivity, specicity and accuracy were obtained via the receptor operative curve (ROC). Delong test was used for comparisons of ROCs. and immunocompromised patients with pulmonary tuberculosis.


Introduction
Pulmonary TB is ranked as the one of the top ten causes of death in the worldwide, which is generally susceptible in all age groups from infancy to the elderly, especially for those with malnutrition, immunocompromised, and smoking history. There were about 10 million new TB patients in the worldwide in 2018, with women and children ( 15 years old) accounting for 32% and 11% respectively.
Among the elderly and immunode ciency patients with pulmonary TB, the incidence of co-existent lung cancer and opportunistic infections evidently grow [1][2][3][4][5][6] . The de nite diagnosis came to priority in the management of tuberculosis.
Chest computer tomography (CT) and x-ray are historically used for evaluation of pulmonary lesions, which are usually harmful due to cumulative radiation exposure, especially for children, the pregnant and young women. Previous studies have reported that cumulative radiation exceed 100mSv caused an increased cancer mortality risk [7][8] . It is also quite di cult for radiologists to distinguish TB from malignancy or other infectious diseases by means of morphology and subjective observation. MRI has advantages such as non-radiation, functional measurements and so on, which is suitable for recurrent examinations and qualitative diagnosis of lesions. DWI and IVIM, based on inherent diffusion motion of water, are clinically used for differential diagnosis of benign and malignant lesions, which also had great diagnostic value for suspicious pulmonary nodules. Shen and his colleagues reported a pooled sensitivity of 0.84 (95%CI: 0.82-0.86) and pooled speci city of 0.84 (95%CI: 0.81-0.87) in identifying pulmonary malignancy using DWI [9] . IVIM D of malignancy was signi cantly lower than benign pulmonary lesions [10,11] , while the diagnostic value of IVIM D* and f value were controversial [12][13][14] . Generally, MRI is available to differentiate pulmonary malignancy and benign lesions, however, it is still unknown about discrimination among pulmonary tuberculosis-like lesions, which was hardly ever reported previously.
In this study, DWI ADCs and IVIM parameters (D, D* and f values) were obtained based on monoexponential and bi-exponential respectively. The diagnostic performances of MRI DWI and IVIM were compared between pulmonary tuberculosis and malignancy, tuberculosis and non-tuberculosis benign lesions. The preliminary study was expected to be helpful for differential diagnosis of granulomatous disease and clinical management of tuberculosis. tuberculosis, 1 non-tuberculosis mycobacterium infection, 2 granulomatous vasculitis, 14 fungal infections, 7 organizing pneumonia, 1 lung abscess. The work-ow was listed as Figure 1. According to the pathological and/or laboratory test evidences, all lesions were divided into three groups including PM (n=38), TB (n=36) and NTB (n=25). The PM consisted of 34 nodules/masses, 1 consolidation and 3 cavities. The TB included 29 nodules/masses, 1 consolidation and 6 cavities. And the NTB had 22 nodules/masses, 1 consolidation and 2 cavities. The average size of lesions was (2.6±1.6)cm, rang from 0.7cm to 7.8cm.

MRI protocol
All patients underwent MRI examination using 3T whole-body MR scanner (MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany) with an 18-element body wrap coil. All parameters was listed as follows: axial T1-

CT scan
All patients underwent examination on CT 320-detector system (Aquilion Vision, Canon Medical Systems, Japan) in the case of breath-holding after inspiration. The following imaging parameters were applied: detector width, 80 x 0.5 mm; pitch, 0.813; tube voltage, 120 kV; automatic tube current with SD 10 (Sure Exp 3D set, maximum: 440 mA, minimum: 60 mA). All CT images with 1 mm contiguous section thickness were reconstructed by means of adaptive iterative dose-reduction with three-dimensional processing (AIDR 3D, standard) and a high frequency reconstruction algorithm (FC56) for the lung window setting, and with a standard reconstruction algorithm (FC17) for the mediastinum window setting. The lung window width and level were adjusted appropriately by the reference standards of 1,600 and 600 Houns eld unit (Hu). The imaging volume was from the pulmonary apex to the costophrenic angle.

Data postprocessing
Firstly, using serial thin-section CT images as references, the morphology of lesions was documented in MRI images. The signal ratio in T1WI (T1SR) and T2WI (T2SR) images was calculated as the ratio of the lesions' signal intensity (SI) to the rhomboid muscular signal. All ROIs were the same pixel. The formula listed as: SR= SI lesions / SI muscular . The hyper-intensity in this study was described as the SR more than 1, otherwise, it was hypo-intensity.
Then, the raw data was imported to the body diffusion toolbox (Siemens, Germany) for acquiring ADC values and IVIM D, D* and f values. The measurements were performed by 2 radiologists respectively with 3-and 6-years experiences of chest radiographs, who were blind to patients' clinical data. And two post-processing experts employed by the Siemens company helped supervise the data-analysis. Using T2WI images as references, ADC (0,600) (b=0,600 s/mm2), ADC (0,800) (b=0,800 s/mm2) and ADC (0,1000) (b=0,1000 s/mm2) were obtained in a basis of DWI model. IVIM D, D* and f value were calculated with IVIM model (b=0,30,50,80,100,200,400,600,800,1000 s/mm2). The region of interests (ROIs) were manually drawn on the three slices with less motion artifacts and large-scale and then were averaged.
ROIs area were more than 50% of lesions at least. And all ROIs should avoid air and necrosis as much as possible.

Statistical analysis
MedCalc 15.2.2 was used for data analysis. The inter-class correlation coe cient (ICC) was used for assessing inter-observer agreement. A Mann-Whitney U test was used for comparisons of ADC-values, IVIM parameters and signal ratio in T1WI/T2WI images between TB and PM, TB and NTB. P-value 0.05 was de ned as statistical differences. Multi-parameters logistic regression was used for analyzing the combined diagnostic e cacy of IVIM and ADC. The diagnostic e cacy was obtained based on receptor operative characteristics (ROC). A maximum Youden's index was used for determining the cut-off value, sensitivity, speci city and accuracy. A Delong-test was used for comparisons of area under the curve (AUC).

IVIM results
The D* in TB and PM were (19.

Discussion
In this study, we noted that 3T-MRI is comparable to thin-section CT for morphology of pulmonary lesions. Our previous studies also veri ed that free-breathing T1WI StarVIBE and T2WI fBLADE TSE were alternative for chest imaging, both with high signal noise ratio (SNR) and contrast noise ratio(CNR) [15] .
Additionally, in terms of specially pathological caseous necrosis in TB [16][17] , T2WI MRI is optimal for identifying such tuberculoma, which typically showed central lower intensity with peripherally ring-like higher intensity. Reasonably, T2SR of TB was evidently lower than lung cancer in this study. It is consistent with previous results [18][19][20] , which reported that more cases showed hyper-intensity in T1WI images and hypo-intensity in T2WI images in tuberculoma with statistical signi cance, when compared to lung cancer. However, in this study, no evident difference of T1SR existed in TB and PM. We assumed that optional T1WI Dixon with in-out phase images would be more promising for differential diagnosis between tuberculoma and lung caner. Moreover, Comparisons of T2SR and T1SR between TB and NTB manifested no statistical difference, which has hardly been reported before.
The signal intensity of the tuberculoma were usually inhomogeneous because of in ammatory hyperplasia, caseous necrosis and liquefaction. Batra, et al. [26] reported that the T2WI signal intensity in tuberculoma was basically consistent with DWI signal intensity. Our results also manifested that 93.1% of tuberculoma (n = 29) had similar signal intensity in T2WI and DWI images, of which 51.7% (n = 15) was both hyper-intensity in T2WI and DWI images, 10.3% (n = 3) was both hypo-intensity, 31.0% (n = 9) was both central hypo-intensity with peripherally ring-like hyper-intensity, and 6.9% (n = 2) was central hypointensity with peripherally ring-like hyper-intensity in T2WI images while hyper-intensity in DWI images.
Functional MRI have been applied for differential pulmonary benign and malignant lesions. DWI ADC of lung cancer is signi cantly lower than benign lesions [10,11,21] , with sensitivity of 80%-88%, and speci city of 89%-93% [22,23] . In this study, ADCs of PM were signi cantly lower than tuberculosis, as previously reported. The ADC (0,800) and ADC (0,1000) obtained the same highest AUC of 0.85 [0.76-0.94], and the accuracy of 82.4%. While, the the total ADC (0,800) had more greater Youden index and less scan time, when compared with ADC (0,1000) , which was optional for identifying TB and PM. All false negative cases in this group were lung adenocarcinoma (n = 6), mainly due to lower cell density and increased intracellular mucus [24,25] . The ADCs [(1.30 ± 0.27) x 10 − 3 mm2/s] in adenocarcinoma was evidently higher than other malignancy [(0.83 ± 0.22) x 10 − 3 mm2/s] in this study. IVIM parameters, especially for D* and f value, were controversial in the differential diagnosis of malignant and benign lesions [12,13] . Our study demonstrated that IVIM D was valuable in identifying TB and PM, with sensitivity of 78.9%, speci city of 83.3% and accuracy of 81.1%. The performance of IVIM D not better than ADC (0,1000) , which might attributed to motion artifacts interfering model tting in a basis of multiple b values and long time. Also, a combination of IVIM D and ADC (0,1000) obviously improved the AUC from 0.85 to 0.91 and the sensitivity from 84.2-94.7%. Thus, a combination of IVIM D and ADC (0,1000) was recommend for differential diagnosis of TB and PM. However, IVIM D* and f in this study were of no use in distinguishing TB from lung caner.
Compared to TB, NTB such as fungal pneumonia, non-tuberculosis mycobacterial (NTM) infection and so on, have co-existent clinical and radiographic manifestations, which are di cult for discrimination, especially for elderly people with underlying diseases and HIV patients. MRI comparisons between TB and NTB have been hardly reported. In this study, there was signi cant difference in ADCs, when compared TB and NTB, of which ADC (0,800) was considerable for differential diagnosis. The sensitivity, speci city and accuracy were 69.4%, 84.0% and 75,4%. IVIM results also showed that IVIM D* in NTB was signi cantly higher than TB, with sensitivity of 77.8% and speci city of 60.0%. Caseous necrosis with reduced micro-circulation might be responsible for IVIM D* difference in TB and NTB. Yet, combination of IVIM D* and ADC acquired no improvement in diagnostic e cacy. And IVIM D and f can not be used for identifying TB and NTB.
Given of standard recommendation of the Fleischner Society [27] , previous studies mainly focused on the differential diagnosis of pulmonary nodules over 8mm. However, imaging features such as nodules, consolidation and so on, all exist in tuberculosis, malignancy or other infectious disease, especially for those with co-existent tuberculosis-lung cancer and mixed infectious diseases. In this context, our inclusion enrolled a span of patients with pulmonary nodules, consolidation and cavity. However, there also were still some limitations in this study. Firstly, in view of this was a small sample study, additional cases were needed for reliability and repeatability. And the benign tumors were absent in present study. Secondly, more detailed grouping were expected, such as fungal infections group, benign tumor group and so on, especially for other infectious diseases. Lastly, our study explored the T2WI, DWI and IVIM features in tuberculosis and other infectious diseases. These results might cannot be extrapolated to areas in which granulomatous diseases are not endemic.
In conclusion, DWI and IVIM were helpful for differential diagnosis between TB and PM, TB and NTB. Additionally, T2WI signal intensity was also valuable in identifying TB and PM. The clinical use of pulmonary MRI is extremely suitable for the elderly and immunocompromised patients with pulmonary tuberculosis, who have high-risk of co-existent lung caner, opportunistic infections and tuberculosis. Availability of data and materials

Abbreviations
The datasets generated and/or analyzed during the current study are not publicly available due privacy but are available from the corresponding author on reasonable request.

Competing interests
Non-nancial competing interests must be declared in this section.    and T2WI fBALDE TSE images(C). The tuberculoma was slightly hyper-intensity in T1WI StarVIBE(B) and central hypo-intensity with peripherally ring-like hyper-intensity in T2WI images, which was corresponding to the central caseous necrosis( ) and peripheral in ammatory granulomatous hyperplasia(black ▲) in the pathology(D).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Tables.docx