Detection of Pulmonary Nodules Using a 2D HASTE MR Sequence: Comparison with MDCT
Abstract
OBJECTIVE. The objective of our study was to determine the diagnostic performance of MRI based on a HASTE sequence for the detection of pulmonary nodules in comparison with MDCT.
MATERIALS AND METHODS. Thirty patients with known pulmonary nodules underwent both MRI and CT. CT of the lung served as the standard of reference and was performed on a 4-MDCT scanner using a routine protocol. MRI was performed with axial and coronal HASTE sequences using a high-performance 1.5-T MR scanner. Image data were analyzed in three steps after completion of all data acquisition. Step 1 was the analysis of all the CT image data. Step 2 was the analysis of all the MR image data while blinded to the results of the CT findings. Step 3 closed with a simultaneous review of all corresponding CT and MRI data, including a one-to-one correlation of the size and location of all the nodules that were detected.
RESULTS. Compared with the sensitivity of CT, the sensitivity values for the HASTE MR sequence were as follows: 73% for lesions less than 3 mm, 86.3% for lesions between 3 and 5 mm, 95.7% for lesions between 6 and 10 mm, and 100% for lesions larger than 10 mm. The overall sensitivity of the HASTE sequence for the detection of all pulmonary lesions was 85.4%.
CONCLUSION. An MRI examination that consists of a HASTE sequence allows one to detect, exclude, or monitor pulmonary lesions that are 5 mm and bigger. Suspicious lesions smaller than 5 mm still need to be validated using CT.
Introduction
Its unsurpassed potential for soft-tissue discrimination, multiplanar imaging, and inherent capabilities for dynamic and functional analyses has advanced MRI to be the technique of choice for the assessment of many organ systems. MRI of the lungs, on the other hand, has not yet reached any clinical relevance. The reasons include limited spatial resolution [1, 2], high susceptibility differences between air spaces and pulmonary interstitium, and the presence of respiratory and cardiac motion artifacts [3].
At the same time, the lungs have proved ideal for exploiting the advantages inherent to CT [4, 5]: The vast density differences between the air-filled alveoli and the pulmonary interstitium allow easy discrimination of the latter and quick identification of dense pulmonary masses even without the administration of IV contrast material. Based on the ability of CT to detect even small amounts of calcium, benign lesions can be readily characterized as such. Hence, there is general agreement that CT represents the standard of reference regarding the detection and characterization of pulmonary lesions.
On the downside, CT is inherently associated with sometimes considerable levels of exposure to ionizing radiation and the potential of nephrotoxicity when contrast agents are applied. Although these issues are largely ignored in the setting of known disease, exposure to ionizing radiation complicates the attempt to exploit the potential benefits associated with CT-based pulmonary screening [5-7]. Thus, the potential to reduce the high incidence of primary and secondary pulmonary malignancies by screening large populations at risk has motivated investigators to search for alternative imaging techniques with lower exposure to radiation [8].
The availability of high-performance gradient systems, in conjunction with phased-array receiver coils and optimized imaging sequences, has introduced new, potentially interesting approaches to MR-based pulmonary imaging [9-12]. These have ranged from pulmonary MR angiography [13] to ventilation-perfusion measurements and virtual bronchoscopy [14-17].
Materials and Methods
Study Design
Between April 2001 and May 2002, 30 patients (19 men, 11 women; age range, 29-87 years; mean age, 53.3 years) with various pulmonary metastasizing malignancies (Table 1) underwent both MDCT and HASTE MRI of the lungs. Inclusion criteria were pulmonary metastases previously confirmed by CT; need to undergo MDCT of the lungs; willingness and ability to provide informed consent as set forth by the local ethics committee, which approved the study; absence of contraindication to MRI such as pacemakers or ocular metal; and availability of MR scanner time within 3 days of the CT examination.
| Primary Malignancy | No. of Patients |
|---|---|
| Breast cancer | 6 |
| Melanoma | 2 |
| Thyroid carcinoma | 3 |
| Gastric cancer | 2 |
| Colorectal carcinoma | 3 |
| Sarcoma | 2 |
| Testicular carcinoma | 4 |
| Hypernephroma | 1 |
| Lymphoma | 1 |
| Central bronchial carcinoma | 3 |
| Peripheral bronchial carcinoma | 3 |
Exclusion criteria were former adverse reactions to IV contrast agents; contraindications to MRI, such as claustrophobia, pacemakers, and metallic implants; and pregnancy.
MR scanning was performed within zero and 3 days (mean, 1.33 days) of the CT examination.
Imaging Techniques
CT was performed on a 4-MDCT scanner (Volume-Zoom, Siemens Medical Solutions) using the following parameters: 140 kVp; 100 mAs; slice width, 5 mm; collimation, 2.5 mm; table feed, 15 mm/sec; rotation speed, 0.5 sec; and effective slice thickness and reconstruction, 5 mm. The in-plane matrix size was 512 × 512. Before imaging, an 18- to 20-gauge IV catheter was placed into an antecubital vein. An automated injector system (CT9000, Liebel-Flarsheim) was used to administer 70 mL of an IV contrast agent (Ultravist 300 [iopromide], Schering) at a rate of 3 mL/sec. After a delay of 30 sec, the image set covering the entire lung was collected over 9-12 sec. This protocol represents the in-house standard for routinely performed thoracic CT.
MRI was performed on a 1.5-T MR scanner (Magnetom Sonata, Siemens Medical Solutions) that was equipped with a high-performance gradient system characterized by an amplitude of 40 mT/m and a slew rate of 200 mT/m per millisecond. A standard phased-array torso surface coil was used for signal reception. The imaging parameters for the axial and coronal HASTE sequence were as follows: TR/TE, 2 R-R intervals/23 msec; flip angle, 160°; effective slice thickness, 5 mm without interslice gaps; matrix size, 158 × 256; and spatial resolution, 2.4 × 1.3 mm2. No contrast agent was applied. Each image set covered the entire lung and was collected during breath-holding. For the reduction of cardiac motion artifacts, ECG-triggering was performed using an active fiberoptic ECG system. The phase-encoding direction was anteroposterior. Thirty-six slices covering the entire chest were collected in two interleaved concatenations of 17-19 sec each, depending on the patient's heart rate.
Image Analysis
The analysis of the imaging data was performed in a three-step manner after completion of data acquisition for the entire patient population. It was based on reviewing hard and soft copies, which were available on a workstation. Although the observers knew pulmonary metastases were present, they remained blinded to the type of primary malignancy and the extent of disease.
In step 1 of the analysis, the standard of reference was defined by two experienced radiologists in consensus by reviewing and interpreting the MDCT scans. All round or ovoid noncalcified lesions within the lung parenchyma were counted as pulmonary nodules. The number, location, and size of the detected lesions were recorded. The nodule diameter was defined as the largest diameter on MDCT scans as displayed with lung window settings (window width, 2,000 H; window level, -500 H). Particular attention was paid to the assignment of nodules into four size categories: less than 3 mm, 3-5 mm, 6-10 mm, and more than 10 mm. To increase the number of observations for the statistical analysis, the observers assigned each of the lesions to one of five pulmonary lobes (upper, middle, or lower right lobe; upper left lobe, including lingula; and lower left lobe) and to its position within the lung parenchyma (peripheral, < 1 cm to pleura and pericardium; central, > 1 cm to pleura and pericardium). Classification of the lobes as affected or not affected by pulmonary lesions was based on a conspicuity scale that ranged from 1 to 4: 1, definitely not affected; 2, likely not affected; 3, likely affected; and 4, definitely affected. To determine the sensitivity and specificity for pulmonary HASTE MRI regarding disease in the pulmonary lobes, we defined grades 1 and 2 as not affected and grades 3 and 4 as affected. To avoid miscounting, the observers marked each detected lesion on hard-copy film.
In step 2 of the analysis, all MR image data, both in the axial and coronal planes, were analyzed in consensus by two experienced radiologists who were unaware of the results of the MDCT examinations. The reviews were performed between 11 and 24 days (mean, 16 days) after the corresponding CT reviews. Despite the variable signal intensities of the solid lesions that were detected, the size, number, and location of all were recorded.
In step 3, the corresponding CT and MR data sets were reviewed again simultaneously for one-to-one comparison of the size and location of the detected and marked nodules. Sensitivity and specificity values for nodule detection of the HASTE sequence were determined using MDCT as the standard of reference. Receiver operating characteristic (ROC) curve analysis was performed for MRI by correlating the presence or absence of lung lesions with the degree of observers' diagnostic certainty. Estimates of the area under the ROC curve were made using the method described by Metz et al. [23].
Results
All the MDCT and HASTE MR sequence examinations were judged to be diagnostic. Both imaging examinations were well tolerated; no adverse reactions were observed with either. The mean room time in the MR scanner was 17 min [13-23], compared with 10 min [8-12] in the CT scanner.
Lesion Detection
MDCT revealed 1,102 lung lesions in 30 patients that were located in 104 of 150 examined lobes. The diameter of 383 lesions (peripheral, 129; central, 254) measured less than 3 mm; the diameter of 300 lesions (peripheral, 113; central, 187), 3-5 mm; the diameter of 232 lesions (peripheral, 95; central, 137), 6-10 mm; and that of 187 lesions (peripheral, 76; central, 111), larger than 10 mm. The mean number of lesions per patient was 37, ranging from four to 174.
The HASTE MR sequence revealed a total of 1,031 pulmonary lesions that were distributed among all 30 patients. Of these lesions, 336 (peripheral, 118; central, 218) measured less than 3 mm; 281 lesions (peripheral, 104; central, 177), 3-5 mm; 225 lesions (peripheral, 91; central, 134), 6-10 mm; and 189 lesions (peripheral, 76; central, 113), larger than 10 mm.
Based on the one-to-one correlation between MRI and CT, the findings in both imaging techniques correlated well (Figs. 1A and 1B). Because of their high native signal intensity, central bronchial masses (n = 3) could be easily distinguished from neighboring mediastinal structures (Figs. 2A, 2B, and 2C). The appearance of pulmonary metastases was not influenced by the underlying primary malignancy (Table 1).
The rate of false-positive and false-negative findings increased with descending lesion size. In the group of lesions smaller than 3 mm, MRI missed 101 lesions (peripheral, 57; central, 44) and revealed 55 false-positive lesions (peripheral, 46; central, 9). In the group of 3- to 5-mm lesions, MRI missed 47 lesions (peripheral, 27; central, 20) and revealed 28 false-positive lesions (peripheral, 18; central, 10). In the group of 6- to 10-mm lesions, MRI missed 10 lesions (peripheral, 7; central, 3) and revealed three false-positive lesions (peripheral, 3; central, 0). In the group of lesions bigger than 10 mm, MRI did not miss any lesion but revealed one false-positive finding, which was located in the lung periphery. On CT, this lesion appeared as a conglomerate of three lesions, each of which was 5 mm or smaller. Generally, the falsely detected or missed lesions were located in peripheral areas close to the pericardium or were surrounded by orthostatic fluid at the pleura (Figs. 3A and 3B).
Accordingly, the sensitivity values for the HASTE MR sequence were 73% for lesions smaller than 3 mm, 86.3% for lesions between 3 and 5 mm, 95.7% for lesions between 6 and 10 mm, and 100% for lesions bigger than 10 mm. The overall sensitivity for the detection of all pulmonary lesions was 85.4%.
The number, size, and location of the detected lesions are summarized in Table 2.
| No. of Lesions Detected on MDCT | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Right Lung | Left Lung | |||||||||||||||
| < 3 mm | 3–5 mm | 6–10 mm | > 10 mm | < 3 mm | 3–5 mm | 6–10 mm | > 10 mm | |||||||||
| Lobe | Center | Periphery | Center | Periphery | Center | Periphery | Center | Periphery | Center | Periphery | Center | Periphery | Center | Periphery | Center | Periphery |
| Uppera | 36 | 22 | 52 | 21 | 33 | 18 | 25 | 15 | 43 | 18 | 30 | 18 | 22 | 17 | 14 | 10 |
| Middle | 22 | 10 | 28 | 17 | 26 | 15 | 9 | 8 | ||||||||
| Lower | 74 | 41 | 42 | 30 | 25 | 24 | 36 | 22 | 79 | 38 | 35 | 27 | 31 | 21 | 27 | 21 |
| Total | 132 | 73 | 122 | 68 | 84 | 57 | 70 | 45 | 122 | 56 | 65 | 45 | 53 | 38 | 41 | 31 |
| Total for size category | 205 | 190 | 141 | 115 | 178 | 110 | 91 | 72 | ||||||||
| No. of Lesions Detected on MR HASTE | ||||||||||||||||
| Uppera | 31 | 22 | 49 | 19 | 32 | 17 | 25 | 15 | 38 | 20 | 29 | 17 | 22 | 17 | 14 | 10 |
| Middle | 16 | 8 | 27 | 15 | 26 | 14 | 10 | 8 | ||||||||
| Lower | 63 | 31 | 38 | 28 | 24 | 23 | 37 | 22 | 70 | 37 | 34 | 25 | 30 | 20 | 27 | 21 |
| Total | 110 | 61 | 114 | 62 | 82 | 54 | 72 | 45 | 108 | 57 | 63 | 42 | 52 | 37 | 41 | 31 |
| Total for size category | 171 | 176 | 136 | 117 | 165 | 105 | 89 | 72 | ||||||||
Note–Periphery, < 1 cm to pleura and pericardium; center, > 1 cm to pleura and pericardium.
a
In the left lung, including lingula
Affected Lobes
The overall conspicuity index gave rise to a value of 0.996 for the area under the ROC curve (Fig. 4). The lobes were classified as definitely not affected (n = 41), likely not affected (n = 4), likely affected (n = 9), and definitely affected (n = 96). When the definitely and likely categories are considered together, MRI identified 105 affected lobes, which in comparison with the CT standard of reference included three false-positive and two false-negative lobes (Table 3).
| MDCT Results | MR HASTE Results | |
|---|---|---|
| Positive: 105 | Negative: 45 | |
| Positive: 104 | True-positive: 102 | False-negative: 2 |
| Negative: 46 | False-positive: 3 | True-negative: 43 |
The resulting sensitivity of the HASTE MR sequence regarding the detection of affected lobes thus amounted to 98.1%, whereas the specificity was 93.5%. The corresponding positive and negative predictive values were 97.1% and 95.6%, respectively.
Discussion
This study documents the ability of an unenhanced 2D HASTE MR sequence to detect pulmonary lesions with high accuracy as long as the diameter of the lesion exceeds 5 mm. These data are encouraging and point to an alternative imaging strategy when considering follow-up studies of proven pulmonary cancer and probably for pulmonary screening of populations at risk for pulmonary cancer.
Although the spatial resolution of the HASTE MR sequence is lower than that of MDCT, both imaging techniques correlated well regarding the determination of size, number, and location of the pulmonary lesions. Compared with a routine CT protocol, the HASTE MR sequence protocol depicted pulmonary lesions with a diameter of 3 mm or larger with a sensitivity of 92.1%. For lesions bigger than 5 mm, the sensitivity rose to 97.6%. When based on affected lobes, the sensitivity rose to 98.1%. Thus, the results of this study confirm the data of previous studies that documented MRI as being only slightly less sensitive than CT in the detection of pulmonary lesions [5, 9, 10, 12].
Of course, these data must be viewed in light of considerable study limitations. Only patients with known pulmonary metastases were examined, which introduced a vast interpretation bias. We believe to have compensated for this bias by performing a one-to-one comparison of all identified lesions. Furthermore, performance of the lobar analysis introduced the possibility of true-negatives. In fact, approximately 30% of the lobes were not affected by pulmonary lesions.
A number of factors favor HASTE as the sequence of choice for MRI of the lungs. High T2 relaxivity translating into high signal intrinsic to most types of neoplastic tissue ensures high lesion conspicuity in the setting of the surrounding air-filled, and thus low-signal, pulmonary parenchyma. Furthermore, pulmonary arteries and veins are depicted as flow voids without any apparent signal. This represents an advantage over CT, on which small pulmonary masses often have attenuation levels similar to those of blood vessels and thus are often indistinguishable from vessels of similar size [12].
Particularly with regard to follow-up studies of bronchial carcinomas, the HASTE MR sequence achieves sufficient lesion contrast even without the IV administration of a paramagnetic contrast agent. Beyond a considerable cost advantage, not requiring a systemic contrast agent eliminates any risk of anaphylaxis and shortens examination times.
Compared with other MR sequences, rapid data acquisition represents another crucial factor predisposing the HASTE sequence for pulmonary examination. A full 2D data set covering the entire chest in contiguous 5-mm sections can be collected during merely two breath-holds, each lasting less than 20 sec. To compensate for the poor resolution in the z-axis, defined by the section thickness of 5 mm, we collected image sets in both the axial and coronal planes.
One issue not addressed by this study relates to lesion characterization. With CT, the interpreter focuses on the presence of calcifications, which, apart from few exceptions, denotes a benign cause. Calcifications are void of signal in the MR experiment and are therefore not seen at all. Hence, densely calcified and thus likely benign nodules may be missed altogether on MRI without grave consequence. Characterization of merely partially calcified lesions, on the other hand, will undoubtedly be rendered more difficult on MRI compared with CT. In view of the limited spatial resolution, basing the differentiation on morphologic criteria is not likely to be helpful either. The analysis of signal properties or enhancement profiles may aid in this regard. At this time, however, no data that support this view exist, to our knowledge.
A weakness of the HASTE sequence is the difficulty in delineating small nodules in the presence of orthostatic fluid accumulations at the bases of the lungs. Although CT is not immune to these problems either, multiple lesions seen unequivocally on CT were clearly missed on MRI. Furthermore, there were many false-positive findings on MRI along the lung bases. This problem may have been accentuated by the severely ill and more often bedridden oncologic patient population examined in this study. Healthy individuals undergoing screening for possible pulmonary masses would exhibit far less orthostatic fluid. If orthostatic fluid collections were still identified, an easy, but time-consuming, solution would involve examining the patients a second time in the prone position.
A direct comparison of the CT and MRI protocols immediately brings to light the far inferior spatial resolution of the HASTE images. The in-plane resolution was limited to 1.3 × 2.4 mm. In addition to image artifacts from respiratory and insufficiently ECG-compensated cardiac motion, limited resolution is the main reason for the poor performance of the HASTE sequence with regard to the detection of small nodules, those with a diameter of less than 5 mm. Although relevant in an oncologic patient population, such small nodules seem to play a less dominant role in the setting of pulmonary screening. Some experts suggest that lesions should exceed 5-7 mm for follow-up to be cost-effective [7, 24-26], even though this approach is controversial.
The data of this study suggest that the HASTE MR sequence is a robust method to use to reliably detect or exclude pulmonary mass lesions of 5 mm and larger and to achieve results comparable to those of standard CT [27-29]. For pulmonary malignancies in this size category, HASTE MR sequence thus can be readily used as a harmless imaging alternative for follow-up studies.
Whether such sensitivity sufficiently meets the oncologic requirements of pulmonary screening remains the subject of controversial discussions. Clearly, this study can be viewed only as a beginning; before advocating that MRI be used for pulmonary screening, more studies focusing on asymptomatic patients at risk for developing bronchogenic carcinoma and correlation to objective data (e.g., intraoperative findings) need to be performed. At present, suspicious lesions smaller than 5 mm still need to be validated by focused (i.e., thin-slice) CT.
Although further refinement is mandatory, the results of this study suggest a future for the HASTE MR sequence, a technique that combines the advantages of tomography and lacks harmful side effects, in the detection of pulmonary nodules.
Footnote
Address correspondence to T. Schroeder ([email protected]).
References
1.
Kauczor HU, Kreitner KF. MRI of the pulmonary parenchyma. Eur Radiol 1999; 9:1755-1764
2.
Hatabu H, Chen Q, Stock KW, et al. Fast magnetic resonance imaging of the lung. Eur J Radiol 1999; 29:114-132
3.
Thompson BH, Stanford W. MR imaging of pulmonary and mediastinal malignancies. Magn Reson Imaging Clin N Am 2000; 8:729-739
4.
Schwickert HC, Thelen M, Schweden F, Kauczor HU, Kersjes W, Triebel HJ. CT and MRI in pneumology [in German]. Pneumologie 1994; 48:1-11
5.
Müller NL. Computed tomography and magnetic resonance imaging: past, present and future. Eur Respir J 2002; 35[suppl]:3S-12S
6.
Collins J. CT screening for lung cancer: are we ready yet? WMJ 2002; 101:31-34
7.
Hillman BJ. Economic, legal, and ethical rationales for the ACRIN national lung screening trial of CT screening for lung cancer. Acad Radiol 2003; 10:349-350
8.
Diederich S, Wormanns D, Semik M, et al. Screening for early lung cancer with low-dose spiral CT: prevalence in 817 asymptomatic smokers. Radiology 2002; 222:773-781
9.
Bader TR, Semelka RC, Pedro MS, et al. Magnetic resonance imaging of pulmonary parenchymal disease using a modified breath-hold 3D gradient-echo technique: initial observations. J Magn Reson Imaging 2002; 15:31-38
10.
Schäfer J, Vollmar J, Schick F, et al. Imaging diagnosis of solitary pulmonary nodules on an open low-field MRI system: comparison of two MR sequences with spiral CT [in German]. Rofo 2002; 174:1107-1114
11.
Ohno Y, Hatabu H, Takenaka D, et al. Solitary pulmonary nodules: potential role of dynamic MR imaging in management initial experience. Radiology 2002; 224:503-511
12.
Kersjes W, Mayer E, Buchenroth M, et al. Diagnosis of pulmonary metastases with turbo-SE MR imaging. Eur Radiol 1997; 7:1190-1194
13.
Goyen M, Laub G, Ladd ME, et al. Dynamic 3D MR angiography of the pulmonary arteries in under four seconds. J Magn Reson Imaging 2001; 13:372-377
14.
Ruehm SG, Goyen M, Quick HH, et al. Whole-body MRA on a rolling table platform (AngioSURF) [in German]. Rofo 2000; 172:670-674
15.
Kauczor HU, Kreitner KF. Contrast-enhanced MRI of the lung. Eur J Radiol 2000; 34:196-207
16.
Kauczor HU, Heussel CP, Schreiber WG, Kreitner KF. New developments in MRI of the thorax [in German]. Radiologe 2001; 41:279-287
17.
Hatabu H, Stock KW, Sher S, et al. Magnetic resonance imaging of the thorax: past, present, and future. Radiol Clin North Am 2000; 38:593-620
18.
Semelka RC, Kelekis NL, Thomasson D, et al. HASTE MR imaging: description of technique and preliminary results in the abdomen. J Magn Reson Imaging 1996; 6:698-699
19.
Haddad JL, Rofsky NM, Ambrosino MM, et al. T2-weighted MR imaging of the chest: comparison of electrocardiograph-triggered conventional and turbo spin-echo and nontriggered turbo spin-echo sequences. J Magn Reson Imaging 1995; 5:325-329
20.
Hatabu H, Gaa J, Tadamura E, et al. MR imaging of pulmonary parenchyma with a half-Fourier single-shot turbo spin-echo (HASTE) sequence. Eur J Radiol 1999; 29:152-159
21.
Lutterbey G, Leutner C, Gieseke J, et al. Detection of focal lung lesions with magnetic resonance tomography using T2-weighted ultrashort turbo-spin-echo-sequence in comparison with spiral computerized tomography [in German]. Rofo 1998; 169:365-369
22.
Leutner C, Gieseke J, Lutterbey G, et al. MRT versus CT in the diagnosis of pneumonias: an evaluation of a T2-weighted ultrafast turbo-spin-echo sequence (UTSE) [in German]. Rofo 1999; 170:449-456
23.
Metz CE, Herman BA, Shen JH. Maximum likelihood estimation of receiver operating characteristic (ROC) curves from continuously-distributed data. Stat Med 1998; 17:1033-1053
24.
Hall FM. Screening for lung cancer: been there and done that. Radiology 2002; 224:928-929
25.
Marshall D, Simpson KN, Earle CC, et al. Potential cost-effectiveness of one-time screening for lung cancer (LC) in a high risk cohort. Lung Cancer 2001; 32:227-236
26.
McLoud TC. Imaging techniques for diagnosis and staging of lung cancer. Clin Chest Med 2002; 23:123-136
27.
Galus M. Preventive medicine and screening. Cleve Clin J Med 2001; 68:82-83
28.
Marcus PM. Lung cancer screening: an update. J Clin Oncol 2001; 19[18 suppl]:83S-86S
29.
Leutner C, Schild H. MRI of the lung parenchyma [in German]. Rofo 2001; 173:168-175
Information & Authors
Information
Published In
Copyright
© American Roentgen Ray Society.
History
Submitted: May 23, 2004
Accepted: November 10, 2004
First published: November 23, 2012
Authors
Metrics & Citations
Metrics
Citations
Export Citations
To download the citation to this article, select your reference manager software.