Chest Imaging • Original Research

Satoh et al.
DWI of Malignant and Benign Pulmonary Nodules

Chest Imaging
Original Research

Can Malignant and Benign

Pulmonary Nodules Be
Differentiated with Diffusion-
Weighted MRI?
Shiro Satoh1 OBJECTIVE. The objective of our study was to evaluate whether diffusion-weighted
Yoshio Kitazume1 imaging (DWI) with a high b factor can be used to differentiate malignancies from benign
Shinichi Ohdama2,3 pulmonary nodules.
Yuji Kimula4,5 MATERIALS AND METHODS. This study included 54 pulmonary nodules (≥ 5 mm in
Shinichi Taura1 diameter) in 51 consecutive patients (37 men, 14 women; mean age, 65.7 years; age range, 31–88
years). Thirty-six (67%) of the 54 pulmonary nodules were malignant, and 18 (33%) were be-
Yasuyuki Endo 2,6
nign. Two radiologists independently reviewed the signal intensity of the nodules on DWI with
Satoh S, Kitazume Y, Ohdama S, Kimula Y, Taura a b factor of 1,000 s/mm2 using a 5-point rank scale without knowledge of clinical data. This
S, Endo Y scale was based on the following scores: 1, nearly no signal intensity; 2, signal intensity between
1 and 3; 3, signal intensity almost equal to that of the thoracic spinal cord; 4, higher signal in-
tensity than that of the spinal cord; and 5, much higher signal intensity than that of the spinal
cord. The Mann-Whitney U test and the receiver operating characteristic (ROC) curve were
used to calculate the difference between the scores of malignant and benign nodules.
RESULTS. On DWI, the mean score of malignant pulmonary nodules (4.03 ± 1.16 [SD]) was
significantly higher (p < 0.01) than that of benign nodules (2.50 ± 1.47), with an area under the
ROC curve of 0.796 (95% CI, 0.665–0.927). When a score of 3 was considered as a threshold, the
sensitivity, specificity, and accuracy were 88.9% (95% CI, 78.6–99.2%), 61.1% (38.6–83.6%), and
Keywords: b factor, diffusion-weighted imaging, lung 79.6% (68.9–90.3%), respectively. Three small metastatic nodules (13, 16, and 20 mm) and one
cancer, MRI, pulmonary nodules bronchioloalveolar carcinoma scored 1 or 2 on the 5-point rank scale. Three granulomas, two ac-
tive inflammatory lung nodules, and one fibrous nodule scored 4 or 5.
DOI:10.2214/AJR.07.3133
CONCLUSION. The signal intensity of pulmonary nodules may be useful for malignant
Received September 10, 2007; accepted after revision and benign differentiation on DWI. However, the interpretation of small metastatic nodules,
February 15, 2008. nonsolid adenocarcinoma, some granulomas, and active inflammatory nodules should be ap-
1
proached with caution.
Department of Radiology, Ohme Municipal General
Hospital, 4-16-5, Higashi-Ohme, Ohme City, Tokyo

198-0042, Japan. Address correspondence to S. Satoh.
2
Department of Pulmonary Medicine, Ohme Municipal
General Hospital, Ohme City, Tokyo, Japan.
3 methods used to examine solitary pulmonary
Present address: Department of Pulmonary Medicine,
National Printing Bureau Tokyo Hospital, Tokyo, Japan.
4
Department of Pathology, Ohme Municipal General
Hospital, Ohme City, Tokyo, Japan.
5
Present address: Department of Pathology, Kurashiki
Medical Center, Kurashiki, Japan.
6
Present address: Department of Pulmonary Medicine,
Graduate School of Tokyo Medical and Dental
University, Tokyo, Japan.
AJR 2008; 191:464–470
0361–803X/08/1912–464
© American Roentgen Ray Society
A
solitary pulmonary nodule is a
common finding on chest radiog-
raphy. PET with 18F-FDG and
CT are two common noninvasive
nodules. FDG PET, which is based on the
metabolic uptake of FDG, has been reported
to increase the diagnostic accuracy of benign
and malignant nodule differentiation [1, 2].
However, because FDG PET shows an in-
creased uptake in lung tissues with active
inflammation or benign nodules, interpreta-
tion should be approached with caution [3].
Morphologic analysis based on the assess-
ment of size, shape, and internal characteris-
tics using CT has been the mainstay in evalu-
ating pulmonary nodules. A nodule having a
corona radiata appearance is likely to be ma-
lignant, whereas the presence of intranodular
fat is a reliable indicator of a hamartoma [4].
In general, the smaller the nodule, the more
likely it is to be benign, especially nodules
less than 5 mm in diameter [4, 5]. Findings of
both calcification and lack of growth for at
least 2 years are generally accepted as reliable
signs of a benign nodule, but other findings
have not proven useful for malignant and be-
nign differentiation [4]. The evaluation of tu-
mor vascularity using contrast-enhanced CT
has proven to be useful for distinguishing ma-
lignant nodules from benign nodules [6]. In
particular, the absence of significant lung
nodule enhancement on CT is strongly predic-
tive of benignancy [7]. However, when the
nodules had significant enhancement, some
overlap was found especially between active
granulomas or hypervascular benign tumors
and malignant nodules [8, 9].

464 AJR:191, August 2008

 DWI of Malignant and Benign Pulmonary Nodules
Although MRI is used relatively infrequently
because of its comparatively high cost, it has an
inherent advantage in terms of tissue character-
ization. Indeed, some investigators have tried to
discriminate malignancy from benign lung tu-
mors by measuring their relaxation times, al-
though the results have not been satisfactory
because of a significant overlap of values [10,
11]. Tissue contrast attained using diffusion-
weighted imaging (DWI) is different from that
attained using conventional MR sequences. The
diffusion technique reflects the diffusion mo-
tion of water protons in the tissues, producing
different contrast in different kinds of tissues.
Promising results have been achieved using this
technique for differentiation between malignant
and benign nodules in the liver [12, 13], bone
marrow [14], and head and neck [15]. However,
to our knowledge, there have been no reports
about DWI applied to pulmonary nodules for
differentiating malignancy from benignancy
because the images in this area are likely to
have susceptibility artifacts [16].
Takahara et al. [17] introduced a new tech-
nique of DWI using a STIR sequence with a
high b factor and free breathing with 10 excita-
tions and concluded that this technique allows
screening for malignancies in the entire body.
Their results showed that a free-breathing tech-
nique with 10 excitations and STIR produced a
higher contrast-to-noise ratio and good fat sup-
pression compared with a breath-hold tech-
nique with 2 excitations and a chemical shift–
selective pulse [17, 18]. The purpose of our
study was to perform DWI of pulmonary nod-
ules and evaluate whether DWI can be used to
differentiate malignant from benign nodules
and to analyze pulmonary nodules that are dif-
ficult to characterize as benign or malignant.
Materials and Methods
Our institutional review board approved this
study, and informed consent was obtained from
all patients.
Subjects
In this study, between November 2004 and April
2006, patients were selected according to the
following criteria: one or two pulmonary nodules
or masses detected on CT that needed further
evaluation; absence of calcification or definite fat
attenuation of the nodule or mass on CT; absence of
histologic diagnosis; absence of history of immuno­
deficiency; and ability to undergo the procedure.
Single-detector CT (X-Vigor, Toshiba Medical)
was performed with the patient in the supine
position. Contiguous 10-mm scans were obtained
during breath-holding and full inspiration from the
AJR:191, August 2008 465
lung apex to the lung base in all patients. The tube
current was 120 kVp at 200 mA.
This study included 54 pulmonary nodules in
51 consecutive patients (37 men, 14 women; mean
age, 65.7 years; age range, 31–88 years) who
underwent DWI in the department of radiology in
our hospital. Three patients had two pulmonary
nodules, and the remaining 48 patients had solitary
pulmonary nodules.
MRI Techniques
All MR examinations were performed with a
commercially available 1.5-T whole-body MR unit
(Intera NovaDual, Philips Healthcare) with a
maximum gradient strength of 33 mT/m and a slew
rate of 160  mT/m/s using a sensitivity-encoding
(SENSE) body coil. All patients were examined in
the supine position throughout the examination.
Before DWI, dual-echo T1-weighted fast-field echo
MRI or T1-weighted turbo spin-echo MRI was
performed. The dual-echo T1-weighted fast-field
echo sequence was performed with the following
parameters: TR range/opposed-phase TE, in-phase
TE, 70–150/2.3, 4.6; flip angle, 80°; matrix size,
208  × 256; SENSE reduction factor, 1.7; field of view,
25–30  ×  25–30 cm; number of signals acquired, 2;
section thickness, 5 mm; section gap, 1 mm; number
of sections acquired, 18–40; and acquisition time,
18.1 seconds. MR images were obtained during
end-inspiration breath-holding. The T1-weighted
turbo spin-echo sequence was performed with
the following parameters: TR range/TE range,
330–1,000/12–15; echo-train length, 5–7; matrix
size, 208–224 × 256–512; SENSE reduction factor,
1.8; field of view, 25  ×  25  cm; number of signals
acquired, 1–2; section thickness, 5–6  mm; section
gap, 0.5–1 mm; number of sections acquired, 9–40;
and acquisition time, 17–30 seconds. MR images
were obtained during end-inspiration breath-
holding after one or two breaths.
The subsequent DWI sequence was performed
with the following parameters: TR/TE range,
infinite/50–70; b factors, 0 and 1,000 s/mm2; STIR;
matrix size, 256 × 128; half scan factor, 0.6; SENSE
reduction factor, 2; field of view, 30–40 × 30–40 cm;
number of signals acquired, 10; section thickness, 4
mm; section gap, –1  mm (overlap); number of
transverse sections acquired, 50–80; and imaging
time, 193–343 seconds. MR images were obtained
during free breathing. The motion-probing gradients
were placed in three axial directions. The echo-
planar imaging factor was 45, and water–fat shift
was 8.583 pixels.
MRI Analysis
Two radiologists, with 9 and 6 years’ experience,
respectively, worked together to retrospectively
evaluate the signal intensity of lung nodules on
DW images (b factor  =  1,000  s/mm2) and T2-
weighted images (b factor  =  0  s/mm2) in a
transverse plane using a 5-point rank scale without
knowledge of patient CT features and clinical
data; final decisions regarding the scale were
reached by consensus. We evaluated T2-weighted
images to examine how they may affect DW
images of lung nodules [19, 20].
The scale for DW images was based on the
following scores: 1, nearly no signal intensity, as
seen in an almost normal lung; 2, signal intensity
between 1 and 3; 3, signal intensity almost equal to
that of the spinal cord at the thoracic spine [15]; 4,
higher signal intensity than that of the spinal cord;
and 5, much higher signal intensity than that of the
spinal cord. The interval between 3 and 4 was nearly
the same as the interval between 2 and 3. Wang et al.
[15] suggested that the spinal cord should replace the
CSF as a reference on DW images [21].
The scale for T2-weighted images was based on
the following scores: 1, nearly no signal intensity,
as seen in an almost normal lung; 2, signal intensity
almost equal to that of the dorsal muscles; 3, lower
signal intensity between 2 and 5; 4, higher signal
intensity between 2 and 5; and 5, signal intensity
almost equal to that of CSF at the thoracic spine
[15] or a saline bag on the opposite side of a
pulmonary nodule that was detected on CT. CSF
has been used as a reference on echo-planar MR
images by some researchers [15, 21]. However, we
used a saline bag together with CSF beginning in
February 2005 because, as discussed by Wang et al.
[15], subarachnoid spaces of the thoracic spine
were sometimes difficult to evaluate as a reference
in some patients in an imaged area that was less
than 5 mm in diameter. Apparent diffusion
coefficient (ADC) maps of lung nodules were not
available because of their susceptibility artifacts.
In this study, nodules of 5 mm or larger in
diameter were included. The diameter was cal­
culated using the mean of the long- and short-axis
diameters of nodules or masses on T1-weighted
images. The anatomic distribution of pulmonary
nodules was classified as the upper or lower lung
zone from the carina.
Final Diagnosis
The final diagnoses were made histologically or
clinically. Without knowledge of the results of the
5-point rank scale, one pathologist interpreted the
histologic diagnoses of specimens from sur­gery or
biopsy. Clinical diagnoses were made by pulmon­
ologists who were unaware of the results of the
5-point rank scale using clinical data and results of
radiologic follow-up studies. Nodules or masses
were classified as granulomas if a diagnosis was
confirmed histologically or bacteriologically (e.g.,
tuberculoma) or if there was radiologic evidence of
 Satoh et al.

no growth during at least 2 years of follow-up [7]. TABLE 1: Diagnoses, Scores, and Mean Size of 54 Pulmonary Nodules
The diagnoses of pulmonary nodules were class­ Evaluated on Diffusion-Weighted Imaging
ified as benign if the nodules were established with 5-Point Rank Scalea Nodule Size (mm)
radiologic follow-up studies that revealed disappear­ Diagnosis of Pulmonary Nodule
(No. of Nodules) 1 2 3 4 5 Mean ± SD Range
ance or significant regression of the nodules after
initiation of antibacterial or steroid therapy [15]. Primary lung cancer (31) 43.4 ± 25.1 13–132
  Squamous cell carcinoma (14) 3 3 8 46.4 ± 20.8 23–96
Statistical Analysis   Adenocarcinoma (10) 1 2 2 5 38.1 ± 35.4 13–132
The Mann-Whitney U test was used to calculate
  Small cell carcinoma (6) 1 5 41.0 ± 13.7 29–66
the difference in the median score on the 5-point
rank scale of DW images and T2-weighted images   Pulmonary MALT lymphoma (1) 1 68
between malignant and benign nodules. A p value Metastatic lung cancer (5) 16.4 ± 5.22 10–23
of less than 0.05 was considered to indicate a   Lung (1) 1 23
significant difference. We used the receiver
  Esophagus (1) 1 10
operating characteristic (ROC) curve to evaluate
the diagnostic capability of the 5-point rank scale   Pharynx (1) 1 13
for differentiation between malignant and benign   Breast (1) 1 16
lesions. We determined the threshold score   Urinary bladder (1) 1 20
showing the highest accuracy. Scores equal to the Benign lung lesions (18) 22.2 ± 15.2 7–60
threshold and higher were considered to indicate
  Granuloma (5) 2 2 1 16.8 ± 7.50 7–28
malignant pulmonary nodules, and scores below
the threshold were considered to indicate benign   Tuberculoma (3) 1 1 1 21.3 ± 1.53 20–23
lung nodules. Statistical analyses were performed   Hamartoma (2) 1 1 10.5 ± 4.95 7, 14b
with SPSS software (version 13.0, SPSS).   Fibrous nodule (2) 1 1 13.0 ± 8.49 7, 19b
  Lung abscess (1) 1 47
Results
Final Diagnosis   Organizing pneumonia (1) 1 60
In 44 of the 54 nodules, the final diagnoses   Round atelectasis (1) 1 51
were made histologically by either surgery   Cryptococcus infection (1) 1 17
(n = 14) or biopsy (n = 30). The diagnosis of
  Inflammatory nodule (1) 1 14
five granulomas was based on radiologic fol-
low-up studies that revealed no change over 2   Bacterial pneumonia (1) 1 15
years. The diagnoses of inflammatory nod- Total (54) 7 8 8 10 21 33.8 ± 23.7 7–132
ule (n  =  1), lung abscess (n  =  1), bacterial Note—Pulmonary MALT lymphoma = B-cell lymphoma of mucosa-associated lymphoid tissue.
aScale was based on the following scores: 1, nearly no signal intensity; 2, signal intensity between 1 and 3; 3,
pneumonia (n = 1), and organizing pneumo-
nia (n = 1) were established, revealing disap- signal intensity almost equal to that of the thoracic spinal cord; 4, higher signal intensity than that of the spinal
cord; and 5, much higher signal intensity than that of the spinal cord.
pearance of the nodules after initiation of bBoth values are listed rather than a range.

antibacterial or steroid therapy. Tuberculosis

was diagnosed in the one remaining patient
by gastric fluid analysis revealing Mycobac-
terium tuberculosis organisms.
The diagnoses of the 54 nodules are listed
in the Table 1. Thirty-six nodules of 34 pa-
tients were malignant and 18 nodules of 18
patients were benign. One patient had one
malignant nodule and one benign nodule.
MRI Analysis
The scores on the 5-point rank scale used
to evaluate malignant and benign pulmonary
nodules on DW images were 4.03 ± 1.16 and
2.50 ± 1.47 (mean ± SD), respectively (Figs. 1
and 2). The scores on the 5-point rank scale
used to evaluate malignant and benign pul-
monary nodules on T2-weighted images
were 3.36  ±  0.64 and 2.94  ±  0.94 respective­-
ly. The mean nodular size was 33.8  ±  23.7
mm. Twenty-five pulmonary nodules were
located in the upper lung zone from the cari-
na, whereas 29 nodules were in the lower
lung zone.
Four of five pulmonary metastases scored 2
or 3 on the scale and were 20 mm or smaller.
One primary lung adenocarcinoma scoring 1
on the 5-point rank scale was a well-differen-
tiated adenocarcinoma with a bronchioloal-
veolar carcinoma of 20  mm diagnosed histo-
logically by surgery (Fig. 3). Six of 18 benign
pulmonary nodules scored 4 or 5 on the scale.
One lung abscess (47 mm) and one granulo-
ma (15 mm) scored 5 on the scale, whereas
two granulomas (28 and 17 mm), one active
tuberculoma (23  mm), and one fibrous nod-
ule (19 mm) scored 4 on the scale (Fig. 4).
Statistical Analysis
On DW images, the scores on the 5-point
rank scale of malignant pulmonary nodules
were significantly higher than those of be-
nign nodules (p  <  0.01), and the area under
the ROC curve was 0.796 (95% CI, 0.665–
0.927) (Fig. 5). The highest accuracy was
obtained when a score of 3 was considered as
the threshold. When scores of 3 or more were
estimated to be malignant and scores of 1
and 2 were benign, the sensitivity, specificity,
and accuracy were 88.9% (95% CI, 78.6–
99.2%), 61.1% (38.6–83.6%), and 79.6%
(68.9–90.3%), respectively. Among nodules
of 30 mm or smaller (malignancy, n = 16; be-
nignancy, n = 15), the scores on the 5-point
rank scale of malignancies on DW images
were significantly higher than those of be-
nign nodules (p < 0.05; U value [Mann-Whit-
ney U test], 171.5; significant points of a two-
tailed U value, p < 0.05, were a lower value of
70 and an upper value of 170), and the area
under the ROC curve was 0.715 (95% CI,

466 AJR:191, August 2008

 DWI of Malignant and Benign Pulmonary Nodules
A B
C D
A B
C D
AJR:191, August 2008 467
Fig. 1—Images of small cell lung carcinoma in right
upper lobe in 88-year-old man in whom diagnosis was
true-positive.
A, Transverse T1-weighted image (TR/TE, 150/4.6)
shows mass (arrow) in right upper lobe.
B, Transverse diffusion-weighted (DW) echo-planar
image (3,084/70) obtained with b factor of 1,000
s/mm2 shows mass (arrow) with very high signal
intensity compared with spinal cord; it scored 5 on
5-point rank scale. Spinal cord scored 3 on 5-point
rank scale on DW images obtained with b factor of
1,000 s/mm2 .
C, Transverse T2-weighted echo-planar image
(3,084/70) obtained with b factor of 0 s/mm2 shows
mass (arrow) with slightly low signal intensity
compared with CSF or saline bag and high signal
intensity compared with dorsal muscle; it scored 4
on 5-point rank scale. CSF or saline bag scored 5 on
5-point rank scale on T2-weighted image obtained
with b factor of 0 s/mm2 . Dorsal muscle scored 2 on
5-point rank scale on T2-weighted images obtained
with b factor of 0 s/mm2 .
D, Photomicrograph of surgically resected specimen
shows small cell lung carcinoma. Tumor cells are
densely packed, with scant cytoplasm. (H and E, × 40)
Fig. 2—Images of round atelectasis in right lower
lobe in 68-year-old man in whom diagnosis was true-
negative.
A, Coronal T1-weighted image (TR/TE, 119/4.6) shows
mass (arrow) in right lower lobe.
B, Transverse diffusion-weighted (DW) image
(4,654/70) obtained with b factor of 1,000 s/mm2
shows mass (arrow) with slightly lower signal
intensity compared with spinal cord; it scored 2 on
5-point rank scale. Spinal cord scored 3 on 5-point
rank scale on DW images obtained with b factor of
1,000 s/mm2 .
C, Transverse T2-weighted image (4,654/70) obtained
with b factor of 0 s/mm2 shows mass (arrow) with
slightly high signal intensity compared with dorsal
muscle; it scored 3 on 5-point rank scale. Small pleural
effusion is evident in posteromedial vicinity of mass.
Dorsal muscle scored 2 on 5-point rank scale on T2-
weighted images obtained with b factor of 0 s/mm2.
D, Photomicrograph of CT-guided core lung
biopsy specimen shows slight inflammatory cell
accumulation in alveolar septa and no neoplastic
tissue. (H and E, × 4)
 Satoh et al.

A B C
Fig. 3—Images of adenocarcinoma in left upper lobe in 53-year-old man in whom diagnosis was false-negative.
A, Transverse T1-weighted image (TR/TE, 148/4.6) shows nodule (arrow) in left upper lobe.
B, Transverse diffusion-weighted (DW) image (3,602/50) obtained with b factor of 1,000 s/mm2 shows nodule (arrow) with very low signal intensity similar to that of
surrounding more-normal lung; it scored 1 on 5-point rank scale. More-normal lung scored 1 on 5-point rank scale on DW images obtained with b factor of 1,000 s/mm2 .
C, Photomicrograph of surgically resected specimen shows well-differentiated adenocarcinoma of lung. Cuboidal to columnar cells grow along alveolar walls in lepidic
fashion. (H and E, ×10)

signal intensity of the spinal cord was con-

sidered as a threshold, the highest accuracy
of 79.6% was obtained. However, small met-
astatic nodules, nonsolid adenocarcinoma,
some granulomas, active inflammatory lung
nodules, and fibrous nodules make it difficult
to differentiate between malignant and be-
nign pulmonary nodules.
Swensen et al. [7] reported that distin-
guishing between malignant and benign soli-
tary pulmonary nodules on contrast-en-
hanced CT images showed an average area
under the ROC curve of 0.831 or 0.785 (the
A B
Shiga cohort). In contrast, our results from
Fig. 4—Image of granuloma in left upper lobe in 62-year-old woman in whom diagnosis was false-positive.
A, Sagittal T1-weighted image (TR/TE, 1,000/12) shows nodule (arrow) in left upper lobe. DWI data alone showed the average area un-
B, Transverse diffusion-weighted (DW) image (9,050/50) obtained with b factor of 1,000 s/mm2 shows nodule der the ROC curve to be 0.796. Diagnosis of
(arrow) with very high signal intensity compared with spinal cord; it scored 5 on 5-point rank scale. Spinal cord the solitary pulmonary nodule is a difficult
scored 3 on 5-point rank scale on DW images obtained with b factor of 1,000 s/mm2 .
task when attempting to do so primarily from
a visual interpretation of the imaging data
0.527–0.902). The highest accuracy was specificity, and accuracy of 100%, 33.3% alone. Our results indicate that interpretation
shown when a score of 3 was considered as a (95% CI, 0–71.1%), and 88.2% (77.4–99.1%), using DWI performed with a b factor of
threshold; the sensitivity, specificity, and ac- respectively. 1,000 s/mm2 has the potential to distinguish
curacy were 75.0% (95% CI, 53.8–96.2%), On T2-weighted images, the scores on the malignant from benign pulmonary nodules.
66.7% (42.8–90.5%), and 71.0% (55.0– 5-point rank scale of malignant pulmonary According to Wang et al. [15], DW images
86.9%), respectively. However, among nod- nodules (3.36 ± 0.639) were not significantly obtained with a b factor of 1,000 s/mm2 may
ules of 20 mm or smaller (malignancy, higher (p = 0.083) than those of benign nod- be used to characterize head and neck lesions.
n = 8; benignancy, n = 12), the scores on the ules (2.94 ± 0.938). Although the area under the ROC curve for
5-point rank scale of DW images showed On DW images, no significant difference use in differentiating malignancies from be-
no significant difference between malig- in scores was found between upper and lower nign lesions was 0.87 in their results, which is
nant and benign nodules (p = 0.262; U val- lung zones in terms of malignant nodules higher than ours (0.796), this difference could
ue, 62; significant points of a two-tailed U (4.28  ±  1.10 and 3.89  ±  1.20, respectively; be explained, at least in part, by several fac-
value, p < 0.05, were a lower value of 22 and p = 0.334) or benign nodules (2.38 ± 1.51 and tors. First, Wang et al. excluded patients who
an upper value of 74). When the pulmonary 2.80 ± 1.55, respectively; p = 0.756). had a local distortion that affected the lesions
nodules were confined to more than 20 mm and had ADC maps that were suboptimal be-
in diameter (malignancy, n = 28; benignan- Discussion cause of susceptibility artifacts. In contrast,
cy, n  =  6), the area under the ROC curve On DWI, the signal intensity of pulmo- almost all of our patients had local distortion
was 0.773 (95% CI, 0.556–0.990) (Fig. 6), nary nodules may be useful for differentiat- between the pulmonary nodules and sur-
and the threshold score of 3, which showed ing malignancy and benignancy; the area rounding aerated lungs due to susceptibility
the highest accuracy, provided a sensitivity, under the ROC curve was 0.796. When the artifacts. Second, breathing artifacts were

468 AJR:191, August 2008


1.0
0.8
True-Positive Fraction
0.6
0.4
0.2
1.0
0.0 0.2 0.4 0.6 0.8
False-Positive Fraction
Fig. 5—Receiver operating characteristic (ROC)
curve of 5-point rank scale for use in differentiation
between malignant and benign pulmonary nodules.
Area under ROC curve is 0.796 (95% CI, 0.665–0.927).
minimal in head and neck lesions in their
study, whereas these artifacts could not be
avoided in pulmonary nodules because pa-
tients were free-breathing. Third, benign cys-
tic lesions were common in head and neck le-
sions in their study, and their mean ADC was
significantly larger than that of carcinomas,
whereas there were no benign cystic pulmo-
nary nodules in our study group.
The interpretation by signal intensity using
DW images of nonsolid neoplasms and active
inflammatory pulmonary nodules was diffi-
cult in our study. The differences in signal in-
tensities may reflect differences in histopatho-
logic features; malignant tumors generally
have enlarged cells and show hypercellularity
[22]. Bronchioloalveolar carcinoma, however,
did not reveal such large cells and did not
show hypercellularity. Lung abscesses and ac-
tive tuberculosis may have had numerous in-
flammatory cells gathering in the nodules and
resultant higher signal intensities, even though
they were benign pulmonary nodules.
The diagnostic information provided by
DW images and ADC maps is not identical.
The signal intensity on DW images is mainly
influenced by T2 relaxation and the ADC
[19, 20]. Increases in T2, such as those typi-
cally seen with brain infarcts, cause an in-
creased signal intensity of lesions on DW
images [19, 20, 23]. This may be explained
by a simultaneous increase in cytotoxic and
vasogenic neuronal edema [24]. Clinically,
the increase in mean signal intensity on T2-
weighted images corresponds well with the
typical development of brain swelling. How-
ever, in our study, the signal intensity on T2-
weighted MR images could not be used to
AJR:191, August 2008 469
DWI of Malignant and Benign Pulmonary Nodules
1.0
0.8
True-Positive Fraction
0.6
0.4
0.2
1.0
1.0 0.0 0.2 0.4 0.6 0.8
False-Positive Fraction
Fig. 6—When pulmonary nodules were confined to
more than 20 mm in diameter (malignancy, n = 28;
benignancy, n = 6), area under ROC curve was 0.773
(95% CI, 0.556–0.990).
distinguish malignancy from benign pulmo-
nary nodules. Unlike brain infarcts, the T2
effect may not substantially influence DW
images in lung nodule differentiation. Al-
though we cannot fully explain this differ-
ence between brain infarcts and lung neo-
plasms in our results, the main reason may
be that lung neoplasms contain little edema
or scant tissue swelling.
Takahara et al. [17] reported on whole-
body DWI using a background body-signal
suppression technique. This technique used
free breathing, STIR, and a high-resolution
3D display. They compared their DWI tech-
nique with a technique involving a sequence
of breath-holding with 2 excitations and a
chemical shift–selective pulse and concluded
that free breathing with 10 excitations and
STIR had a higher contrast-to-noise ratio than
the other technique and had good fat suppres-
sion [17, 18]. However, further study may be
required to confirm their conclusions.
The differences in mean signal intensities
between malignant and benign pulmonary
nodules were slightly closer in the lower lung
zone than those in the upper lung zone. These
results indicate that breathing artifacts, mainly
caused by diaphragmatic motion during free
breathing, could close the difference in signal
intensity between malignant and benign pul-
monary nodules. We recommend obtaining
DW images using respiratory gating, especially
if there is a nodule in the lower lung zone.
When the score of 3, which was a nodule
with a signal intensity with nearly the same
signal intensity as the spinal cord, was con-
sidered as a threshold, the maximum accu-
racy obtained was 79.6%, which is not par-
ticularly high. This accuracy is low because
our study group had comparatively many be-
nign pulmonary nodules (n = 18). For exam-
ple, when the pulmonary nodules were con-
fined to more than 20 mm in diameter, the
number of benign lung nodules decreased
from 18 to six (malignancy, from 36 to 28),
and the accuracy increased from 79.6% to
88.2%. However, the area under the ROC
curve was similar (0.796 vs 0.773). It can be
shown that accuracy is equivalent to
1.0 PREVs × sensitivity + (1 – PREVs) × 
specificity,
where PREVs is the prevalence of disease in
the sample. That is, accuracy is affected not
only by the threshold chosen, but also by the
prevalence of disease in the study sample,
whereas accuracy derived from the ROC
curve indicates the inherent accuracy of a di-
agnostic test because it is affected by neither
the chosen threshold nor the prevalence of
disease in the study sample [25].
There are several limitations to this study.
First, avoiding susceptibility artifacts on
DWI of pulmonary nodules is difficult. This
is the reason we used the 5-point rank scale
rather than ADC maps. Wang et al. [15] re-
ported that they were unable to measure
ADC values of lesions located adjacent to
air-containing organs because of susceptibil-
ity artifacts. In our study, these artifacts may
have made it difficult to interpret pulmonary
nodules, resulting in a reduction of the area
under the ROC curve and of diagnostic ac-
curacy. Second, in our final diagnosis, histo-
logic and bacteriologic confirmations were
made in only 83.3% (45/54) of patients, and
the remaining cases (benignancy, n = 9) had
clinical data and radiologic follow-up confir-
mation. However, not all benign pulmonary
nodules need histologic confirmation. We in-
troduced clinical data and radiologic follow-
up confirmation to avoid a selection bias that
would exclude mainly benign pulmonary
nodules from the study sample. If the study
group includes fewer benign pulmonary nod-
ules, it is difficult to analyze benign nodules
that mimic malignancies on DWI, which was
one of the aims of our study.
In conclusion, the signal intensity of pul-
monary nodules may be useful for malignant
and benign differentiation on DW images.
However, the interpretation of signal intensity
of small metastatic pulmonary nodules, non-
solid adenocarcinoma, some granulomas, and
active inflammatory lung nodules should be

treated with caution. Unlike conventional MR
sequences, DWI is the only sequence that may
be used to help in distinguishing malignant
from benign pulmonary nodules. Further elu-
cidation of the relationship between signal in-
tensity on DWI and pulmonary nodules and
reduction of susceptibility and motion arti-
facts will be required in the future.
References
1. Yi CA, Lee KS, Kim B, et al. Tissue characteriza-
tion of solitary pulmonary nodule: comparative
study between helical dynamic CT and integrated
PET/CT. J Nucl Med 2006; 47:443–450
2. Lobrano MB, Hayman E, Dekelbaum M, Cam-
peau R. Biopsy findings in PET/CT-positive lung
lesions in a community hospital. J La State Med
Soc 2005; 157:319–324
3. Shim SS, Lee KS, Kim B, Choi JY, Chung MJ,
Lee EJ. Focal parenchymal lung lesions showing a
potential of false-positive and false-negative in-
terpretations on integrated PET/CT. AJR 2006;
186:639–648
4. Erasmus JJ, Connolly JE, McAdams HP, Roggli
VL. Solitary pulmonary nodules. Part 1. Morpho-
logic evaluation for differentiation of benign and
malignant lesions. RadioGraphics 2000; 20:43–58
5. Henschke CI, Yankelevitz DF, Naidich DP, et al.
CT screening for lung cancer: suspiciousness of
nodules according to size on baseline scans. Radi-
ology 2004; 231:164–168
6. Kim JH, Kim H, Lee K, Kim KH, Lee HL. Soli-
tary pulmonary nodules: a comparative study
evaluated with contrast-enhanced dynamic MR
imaging and CT. J Comput Assist Tomogr 2004;
28:766–775
7. Swensen SJ, Viggiano RW, Midthun DE, et al.
Lung nodule enhancement at CT: multicenter
470 AJR:191, August 2008
Satoh et al.
study. Radiology 2000; 214:73–80
8. Jeong YJ, Lee KS, Jeong SY, et al. Solitary pul-
monary nodule: characterization with combined
wash-in and washout features at dynamic multi-
detector row CT. Radiology 2005; 237:675–683
9. Schaefer JF, Vollmar J, Schick F, et al. Solitary pul-
monary nodules: dynamic contrast-enhanced MR
imaging—perfusion differences in malignant and
benign lesions. Radiology 2004; 232: 544–553
10. Ross JS, O’Donovan PB, Novoa R, et al. Magnetic
resonance of the chest: initial experience with im-
aging and in vivo T1 and T2 calculations. Radiol-
ogy 1984; 152:95–101
11. Shioya S, Haida M, Ono Y, Fukuzaki M, Ya­
mabayashi H. Lung cancer: differentiation of tu-
mor, necrosis, and atelectasis by means of T1 and
T2 values measured in vitro. Radiology 1988; 167:
105–109
12. Namimoto T, Yamashita Y, Sumi S, Tang Y,
Takahashi M. Focal liver masses: characteriza-
tion with diffusion-weighted echo-planar MR im-
aging. Radiology 1997; 204:739–744
13. Kim T, Murakami T, Takahashi S, Hori M, Tsuda
K, Nakamura H. Diffusion-weighted single-shot
echoplanar MR imaging for liver disease. AJR
1999; 173:393–398
14. Maeda M, Sakuma H, Maier SE, Takeda K. Quan-
titative assessment of diffusion abnormalities in
benign and malignant vertebral compression frac-
tures by line scan diffusion-weighted imaging.
AJR 2003; 181:1203–1209
15. Wang J, Takashima S, Takayama F, et al. Head
and neck lesions: characterization with diffusion-
weighted echo-planar MR imaging. Radiology
2001; 220:621–630
16. Matoba M, Tonami H, Kondou T, et al. Lung car-
cinoma: diffusion-weighted MR imaging—pre-
liminary evaluation with apparent diffusion coef-
ficient. Radiology 2007; 243:570–577
17. Takahara T, Imai Y, Yamashita T, Yasuda S, Nasu
S, Van Cauteren M. Diffusion weighted whole
body imaging with background body signal sup-
pression (DWIBS): technical improvement using
free breathing, STIR and high resolution 3D dis-
play. Radiat Med 2004; 22:275–282
18. Koh DM, Collins DJ. Diffusion-weighted MRI in
the body: applications and challenges in oncology.
AJR 2007; 188:1622–1635
19. Burdette JH, Elster AD, Ricci PE. Acute cerebral
infarction: quantification of spin-density and T2
shine-through phenomena on diffusion-weighted
MR images. Radiology 1999; 212:333–339
20. Eastwood JD, Engelter ST, MacFall JF, Delong
DM, Provenzale JM. Quantitative assessment of
the time course of infarct signal intensity on dif-
fusion-weighted images. Am J Neuroradiol 2003;
24:680–687
21. Tanner SF, Ramenghi LA, Ridgway JP, et al.
Quantitative comparison of intrabrain diffusion in
adults and preterm and term neonate and infants.
AJR 2000; 174:1643–1649
22. Anderson JR. Tumours: general features, types
and examples. In: Anderson JR, ed. Muir’s text-
book of pathology, 12th ed. London, England:
Edward Arnold, 1985:12.1–12.49
23. Warach S, Chien D, Li W, Ronthal M, Edelman
RR. Fast magnetic resonance diffusion-weighted
imaging of acute human stroke. Neurology 1992;
42:1717–1723
24. Lansberg MG, Thijs VN, O’Brien MW, et al. Evo-
lution of apparent diffusion coefficient, diffusion-
weighted, and T2-weighted signal intensity of
acute stroke. Am J Neuroradiol 2001; 22:637–644
25. Obuchowski NA. Fundamentals of clinical re-
search for radiologists: ROC analysis. AJR 2005;
184:364–372