British Journal of Radiology (2003) 76, 866-874
© 2003 British Institute of Radiology
doi: 10.1259/bjr/99569888
Small "indeterminate" lesions on CT of the liver: a follow-up study of stability
P J Robinson, MB, FRCP, FRCR1,
P Arnold, BSc1 and
D Wilson, MSc2
1 Clinical Radiology Research Unit and 2 Medical Physics Department, St James's University Hospital, Beckett Street, Leeds LS9 7TF, UK
 |
Abstract
|
|---|
Distinguishing between small benign malformations in the liver and early metastatic disease remains difficult. We identified a group of 115 patients with known or suspected malignant disease who had "indeterminate" small liver lesions and who underwent 2–16 CT examinations (median 5) over a follow up period of 6–60 months (median 16). The size, shape, edge, homogeneity and attenuation of each of these lesions was assessed. The lesions were classified by their behaviour on follow up CT as either stable (79%) or unstable (21%). The unstable lesions (n=62) included 37 that grew larger and 25 that became smaller or disappeared in patients undergoing anti-tumour therapy. Image features predictive of stable behaviour were small size and sharp edge. Heterogeneity and soft tissue attenuation were significantly associated with unstable behaviour, but these features were seen in only a small minority of cases. Shape had no predictive value. A logistic regression model was constructed based on size and edge to allow an estimate to be made of the likelihood of an individual lesion being unstable. In patients with known or suspected malignant disease, the majority of isolated small liver lesions found on CT are benign. Although size under 5 mm and a sharp margin are favourable features, this appearance does not exclude malignancy.
|
Introduction
|
|---|
The early detection of liver metastases is one of the main objectives of staging CT examinations in patients with suspected or established malignancy. Metastases larger than 15 mm diameter can usually be distinguished from benign lesions in the same size range, but the discrimination becomes more problematic with smaller lesions. The difficulties of characterizing small lesions arise from two technical limitations. First, partial volume averaging reduces the contrast between normal and abnormal tissues, and this effect increases with decreasing size of lesions, so that (for example) a lesion with a diameter equal to half the CT slice thickness will have apparent contrast of about one third of its actual contrast. Second, tumours at an early stage of growth (up to 1–2 cm) are usually solid, so they have less intrinsic contrast with surrounding liver than larger lesions which typically show some degree of central necrosis.
Previous studies have shown that small lesions which are discovered incidentally during the investigation of patients who are not known to have malignant disease are almost always benign [1]. Even in patients with established malignancy, the majority of small lesions discovered on CT are benign [1, 2], but a small proportion of these lesion will be metastatic. In the current study we identified a cohort of patients with suspected or established malignant disease who underwent sequential CT examinations over a prolonged period. The objective was to correlate the imaging appearances of the small lesions with their subsequent behaviour.
|
Patients and methods
|
|---|
By retrospective review of the radiology departmental records over a 5-year period, a group of patients who had undergone at least two contrast-enhanced CT examinations of the liver, with at least 6 months between first and last examinations, was identified. The CT examinations used dynamic incremental acquisition of contiguous 8 mm or 10 mm slices, after injection of 1.5 ml kg-1 bodyweight of contrast medium containing 300 mg ml-1 of iodine. The reports of these examinations were searched to select out those that contained reference to small (15 mm or less) lesions. Patients were excluded if the reports also mentioned the presence of large or multiple tumours, or if review of their initial CT images revealed other lesions larger than 15 mm. 115 patients (84 females, 31 males) fulfilled the selection criteria of two or more contrast enhanced CT examinations separated by at least 6 months, with one or more small lesion(s) being reported as the only liver abnormality. The hard copy CT images of all examinations on all of these patients were re-reported, initially by a single observer (PA).
The following characteristics of each lesion were recorded: size (maximum diameter in millimetres), shape (round or irregular), edge (sharp or unsharp), attenuation (water or soft tissue attenuation, visually assessed), internal structure (homogeneous or heterogeneous). These lesion characteristics were recorded on the first CT study for each patient, and subsequent studies were then examined to identify changes in the lesion. On the basis of serial CT images, each lesion was characterized as either stable or unstable. The unstable group included those lesions which disappeared or diminished in size over the period of observation, as well as those that enlarged. Where lesions were thought to have changed in size, the CT images were reviewed by a second observer and then by both readers in consensus. A change in size was defined as a difference in at least 25% in the maximum diameter of the lesion. Particular attention was paid to ensure that any differences in slice registration in follow-up examinations were recognised. Cases in which slight differences in registration or partial volume effects were believed to explain apparent changes in lesion size were recorded as showing no change. Consensus review was undertaken to establish whether the lesions had increased in size, decreased or disappeared, and if so to determine whether any technical variables (slice registration, respiratory artefact, windowing) could account for the apparent change.
The case notes of the patients were reviewed to identify episodes of treatment – either surgery, radiotherapy or chemotherapy. The histology of primary tumours was recorded when known, liver histology when available and treatment outcomes were also recorded.
Association between the biological behaviour (stable or unstable) of the lesions, and each of the individual imaging characteristics was measured by chi squared testing. The data were also entered into a multiple logistic regression model to determine which characteristics were predictive of lesion outcome. The potential predictive factors entered into the analysis were size, edge, shape and homogeneity. For this purpose the size data were categorised into three groups – 5 mm or less, >5 mm up to 10 mm, and >10 mm up to 15 mm. Attenuation was not included in this analysis because it was not independent of homogeneity. A model was formed to estimate the probability that a lesion was unstable on the basis of any statistically significant predictive factors.
|
Results
|
|---|
Patients
The mean number of CT examinations per patient was 5.1 (range 2–16, median 5) over a mean duration of 19.3 months (range 6 months to 5 years, median 16 months). Of the 115 patients studied, 105 (77 females, 28 males) had a known malignancy, including 30 with ovarian carcinoma, 33 with bowel cancer, and 42 with other primary malignancies. In 10 cases (7 females, 3 males) malignant pathology was initially suspected but never confirmed. 13 patients had unstable lesions only, 83 patients had stable lesions only, and 19 had both stable and unstable lesions (Figure 1
), so in total 32 patients had unstable and 102 had stable lesions.
View larger version (108K):
[in this window]
[in a new window]
|
Figure 1. Mixed stable and unstable lesions in a patient with oesophageal carcinoma. The small lesion seen on the initial CT (a) remained unchanged throughout follow up, but the final CT study (c) also showed several liver metastases with similar appearances to those of the stable lesion.
|
|
All lesions
In total, 296 lesions were identified (226 in females, 70 in males). Of these, 234 lesions (79%) were stable, (179 in females, 55 in males) whilst 62 lesions (21%) changed over time and were classed as unstable (47 in females, 15 in males).
Stable lesions
Of the 179 stable lesions found in 76 female patients, 91% occurred in patients with a known malignancy and 9% in patients with no malignancy confirmed (Figures 2 and 3). Of the 55 stable lesions found in 26 male patients, 88% occurred in patients with known malignancy and 12% in patients with no malignancy confirmed. These proportions were not significantly different from the proportions of patients in the study with benign and malignant disease, i.e. the mean number of lesions per patient was similar, whether the underlying pathology was benign or malignant.
View larger version (92K):
[in this window]
[in a new window]
|
Figure 2. Typical appearance of a stable lesion with a well-defined edge and homogeneous low attenuation in a patient with colonic cancer.
|
|
View larger version (111K):
[in this window]
[in a new window]
|
Figure 3. A small but sharply defined lesion (arrow) which remained stable over 9 months in a patient with ovarian carcinoma—note ascites on the initial CT.
|
|
Unstable lesions
Of the 62 unstable lesions, 47 were found in female patients, 36% of whom had ovarian tumours and 23% large bowel tumours. 15 unstable lesions were found in 10 male patients, 9 of whom had primary bowel cancer.
Of these 62 lesions, 37 enlarged during the period of observation (Figure 4). 25 lesions showed a convincing size reduction or disappeared on sequential studies with similarly registered slices (Figure 5). Of these shrinking or disappearing lesions, 24 of 25 were in patients undergoing chemotherapy, whilst the remaining lesion was in a patient on Interferon maintenance treatment. Of these 25, 17 lesions in 10 patients disappeared between serial studies, including five of 2–5 mm in size and 12 between 5 mm and 10 mm. Four of the 25 lesions (in two patients) decreased in size before disappearing; three of these were between 5 mm and 10 mm, and one was 15 mm. Three lesions (in three patients) first increased in size then reduced and disappeared; their initial sizes were 3 mm, 4 mm, and 6 mm. One lesion disappeared, reappeared and then disappeared again.
View larger version (247K):
[in this window]
[in a new window]
|
Figure 4. Typical unstable small lesion in a patient with ovarian cancer. At staging (a), the subcentimetre low attenuation lesion (arrow) was regarded as indeterminate. Interval studies over the next 4 years showed gradual progression of disease with the development of surface lesions (b–f).
|
|
View larger version (137K):
[in this window]
[in a new window]
|
Figure 5. "Disappearing" lesion in a patient with colon cancer. CT at staging showed an indeterminate small lesion (a) which remained visible on follow up CT up to 17 months (b, c) but on later follow up at 20, 25, 27 and 33 months, the lesion was undetectable (d–h). Note g and h are contiguous slices from the final examination.
|
|
Relation to primary pathology
Unstable lesions were found in 14 of 33 (42.4%) patients with bowel primaries, 8 of 30 (26.7%) patients with ovarian primaries, 10 of 42 (23.8%) with all other malignancies combined, and in 32 of 115 (27.8%) in the group as a whole. No unstable lesions were found in patients who did not have malignant disease.
Statistical analysis—individual imaging features
The association between the imaging characteristics of individual lesions and their subsequent behaviour is illustrated in Tables 1–5. Lesion size predicted outcome well (p<0.001) with larger lesions being more likely to be unstable. A sharp edge was significantly associated with stability (p=0.018) and whilst heterogeneity was significantly associated with subsequent instability (p=0.017), only a small proportion of lesions (5.7%) were heterogeneous. Soft-tissue or mixed attenuation was predictive of unstable behaviour (p=0.002) but again this feature was detected only in a minority of cases (17.6% of all lesions). Shape had no predictive value.
Statistical analysis—logistic regression
Size (p<0.001) and edge (p=0.002) were found to be the only significant predictors of lesion outcome in the model. Thus, the model formed was used to estimate the probability that a lesion was unstable on the basis of size and edge. The probabilities for unstable behaviour for all combinations of size and edge are shown in Table 6.
View this table:
[in this window]
[in a new window]
|
Table 6. Model derived by multiple logistic regression analysis showing percentage probability of a lesion being malignant using the characteristics of size and edge as predictive factors
|
|
|
Discussion
|
|---|
Small patches of skin pigmentation, adenomatous nodules in the thyroid, "unidentified bright objects" in the brain, small cysts in the kidney, sub-centimetre polyps in the bowel, and small granulomas in the lung are all common incidental findings, particularly in older patients. The liver likewise is a common site for small cysts, haemangiomas, microhamartomas and other benign malformations, which become more prevalent with increasing age and pose no threat to health, but create difficulty in the early diagnosis of significant pathologies, particularly malignancy. We need to characterize these lesions in order to avoid inappropriate diagnosis, and particularly in patients with known malignancy, to avoid unnecessary surgery or chemotherapy.
The conventional approach to differential diagnosis of focal liver disease is to try to distinguish between benign and malignant lesions. When imaging characteristics are regarded as indeterminate, guided biopsy or surgical removal of the lesions may be the only definitive diagnostic approach. Since patients undergoing CT of the liver–whether or not they have malignant disease–are quite often found to have small lesions, surgical removal is inappropriate owing to its morbidity, invasiveness and costs. In expert hands, ultrasound guided biopsy has been shown to be highly successful in obtaining positive histology from lesions in the size range of 9–15 mm [3]. However, improving CT techniques, particularly multidetector CT using thin "slice" reconstructions, now allow us to detect increasing numbers of smaller and smaller lesions, so biopsy becomes impractical. MRI, using heavily T2 weighted sequences, is particularly sensitive to the high water content of small cysts, and also allows the differentiation of haemangiomas from metastases in the majority of cases. It appears likely that increasing access to MRI will offer a potentially effective method for discriminating small benign malformations from metastases, but at present the availability of MR scanning time is insufficient to deal with the large number of patients found to have indeterminate lesions on staging and screening examinations.
Taking a pragmatic approach, we may characterize the lesions according to their behaviour on sequential imaging studies rather than on their pathological characteristics, i.e. a possible classification would be "stable" or "unstable", as a surrogate for histologically benign or malignant. In previous studies it has been assumed that lesions that grow over time are malignant, whilst those which remain the same size are benign, although this approach risks categorising as benign some patients with stable malignancy. The current study adds a further observation–some lesions also become smaller or disappear, and these should also be regarded as "unstable"–i.e. probably histologically malignant. It is expected that chemotherapy may lead to metastases becoming smaller, but it is also apparent from our study that some small lesions may shrink below the threshold of visibility and become undetectable on examinations repeated after treatment. In some of our patients such lesions recurred later at the same site, suggesting that small foci of viable tumour remained at the original sites of metastasis, whilst in other cases new lesions appeared elsewhere in the liver, indicating a second crop of metastatic tumours.
As the study population was relatively small, there is insufficient data to draw conclusions about the significance of lesions found in patients with different primary malignancies. In the whole group, the incidence of unstable lesions was 27.8%, in patients with colorectal primaries it was 42.4%, with ovarian primaries 26.7%, and with all other malignancies combined 23.87%. In patients with no known malignancy, none of the lesions were unstable. It may be that instability is more likely in patients with primary bowel lesions, but the number of patients enrolled in the study is too small to draw a firm conclusion.
The two imaging features which most strongly influence the conspicuity of liver lesions on CT are the liver-lesion contrast and the sharpness of the interface between the two. It would be expected that liver cysts, which typically show a sharp interface and water attenuation, would be the lesions most easily visible at a small size. This might explain why the single most useful predictor of the behaviour of a lesion in the current series was its size at the time of initial presentation; the smaller the lesion, the less likely it was to be unstable. The apparent sharpness of the edge of a lesion is also influenced by observed contrast, so with lesions of smaller diameter than the slice thickness, when contrast is degraded by volume averaging, it becomes increasingly difficult to identify the true sharpness of the edge of a lesion. This difficulty explains the relatively large proportion of stable lesions which were observed to have unsharp margins. Even so, in the current series a sharp margin was significantly associated with stable behaviour. Attenuation was not measured in the current study owing to the difficulties associated with partial volume averaging, and it is perhaps not surprising that visual assessment of non-watery attenuation, although associated with unstable lesion behaviour, was an inconstant feature. The shape of lesions was also unhelpful in predicting their behaviour. Experimental work has shown that micro-metastases typically take up a spherical shape, but many of the stable lesions in this series were also spherical. Pathologically, small cysts are usually spherical, but biliary microhamartomas (von Meyenberg complexes) are often linear, angular or irregular in shape [4].
This study suffers from several limitations. The patients were selected retrospectively on the basis of the CT reports, so we included only those patients in whom small lesions were recognised by the radiologist, and who then underwent further CT examinations. No conclusions can be drawn from these data about the incidence of small lesions in the general population, but our aim was to identify those patients in whom the interpretation of liver CT is most problematic. For the same reason we excluded patients who had large lesions or a mixture of large and small lesions.
Although all the images of all the CT examinations were reviewed for this study, the initial review was made by a single observer and only those cases in which the lesions were thought to have changed over time were reviewed again by a second observer, and then finally in consensus by both observers. We did not attempt to measure the effect of observer variation on our classification of lesions so the results may harbour a margin of error from this source. However, both observers had extensive experience in liver CT, and both had taken part in previous studies of lesion detection which incorporated alternative free response receiver operating characteristic (AFROC) analysis, in which the variation between these observers was shown to be small [5].
The image characteristics were determined only by visual observation; no attempt was made to quantitate the attenuation of lesions in Hounsfield units (HU), the heterogeneity of lesions (distribution of HU values) or the edge sharpness (rate of change of HU at the lesion margin). Although it could be argued that such measurements would define the characteristics of liver lesions more clearly, a subjective element is still involved in recording the measurements. Furthermore, with lesions smaller than the slice thickness, measurements of attenuation become unreliable, particularly in lesions with an irregular shape. The visual approach we used reflects routine clinical practice.
The CT technique which was used is historic; recent developments in technology have improved the detection of small liver lesions. Using single pass helical CT, overlapping reconstructions show more small lesions than contiguous slices [6]. A multidetector CT study using an effective beam width of 2.5 mm showed that as the same volume data was reconstructed into increasingly thinner slices (from 10 mm down to 2.5 mm), there were corresponding increases in the number of small lesions detected [7]. The current study used what is now an outdated technique of contiguous 8–10 mm slices with dynamic incremental acquisition, so our sensitivity for small lesions is likely to be less than that achievable with newer technology. However, recent experience with multidetector, multislice CT shows that improved resolution leads to the discovery of "indeterminate" lesions of still smaller size, so the nature of the diagnostic problem remains the same, only the size range has been shifted down a little. The use of dynamic incremental rather than helical CT allows more opportunity for apparent changes in lesions on consecutive studies resulting from mis-registration and varying partial volume effects. The observers were acutely aware of this when reviewing consecutive studies and made every effort to identify the lesions in relation to anatomic landmarks within the liver.
Finally, histology of the lesions was only obtained in a few cases. We would suggest that when the terms "benign" and "malignant" are used to describe histological appearances, they are used as a metaphor for the behaviour of the lesion. A small lesion which remains stable over a prolonged period of observation may be regarded as benign. In this series, the average duration of observation was 19.3 months but in a few patients the interval between first and last examinations was only 6 months, and in such cases a degree of uncertainty about long-term behaviour must remain.
Alternative methods of imaging may be more decisive. Positron emission tomography (PET) using fluorine-18-deoxyglucose may be more accurate than CT in detecting colorectal liver metastases [8] but the sensitivity of PET for detecting small lesions remains to be established. In our own experience heavily T2 weighted MRI usually discriminates between benign and malignant liver lesions [9], but no systematic pathological studies have yet been published to confirm this impression.
In conclusion, as in earlier studies [1, 2] our study has shown that the majority of small "indeterminate" liver lesions found on CT in patients with known malignant disease are benign, and that such lesions are virtually always benign when discovered in patients with no primary malignancy. The characteristics of size and edge are significant in predicting the future behaviour of these small lesions. Lesions that become smaller or disappear during treatment should be regarded as malignant, in addition to those which grow larger. A logistic regression model is proposed, using lesion size and edge to predict the probability of unstable behaviour.
Received for publication October 31, 2002.
Revision received April 15, 2003.
Accepted for publication July 24, 2003.
|
References
|
|---|
- Jones EC, Chezmar JL, Nelson RC, Bernardino ME. The frequency and significance of small (less than or equal to 15 mm) hepatic lesions detected by CT. AJR Am J Roentgenol 1992;158:535–9.[Abstract/Free Full Text]
- Schwartz LH, Gandras EJ, Colangelo SM, Ercolani MC, Panicek DM. Prevalence and importance of small hepatic lesions found at CT in patients with cancer. Radiology 1999;210:71–4.[Abstract/Free Full Text]
- Middleton WD, Hiskes SK, Teefey SA, Boucher LD. Small (1.5 cm or less) liver metastases: US-guided biopsy. Radiology 1997;205:729–32.[Abstract/Free Full Text]
- Lou TY, Itai Y, Eguchi N, Kurosaki Y, Onaya H, Ahmadi Y, et al. Von Meyenburg complexes of the liver: imaging findings. J Comput Assist Tomogr 1998;22:372–8.[CrossRef][Medline]
- Scott DJ, Guthrie JA, Arnold P, Ward J, Atchley J, Wilson D, et al. Dual phase helical CT versus portal venous phase CT for the detection of colorectal liver metastases: correlation with intra-operative sonography, surgical and pathological findings. Clin Radiol 2001;56:235–42.[CrossRef][Medline]
- Urban BA, Fishman EK, Kuhlman JE, Kawashima A, Hennessey JG, Siegelman SS. Detection of focal hepatic lesions with spiral CT: comparison of 4- and 8-mm interscan spacing. AJR Am J Roentgenol 1993;160:783–5.[Abstract/Free Full Text]
- Weg N, Scheer MR, Gabor MP. Liver lesions: improved detection with dual-detector-array CT and routine 2.5-mm thin collimation. Radiology 1998;209:417–26.[Abstract/Free Full Text]
- Zhuang H, Sinha P, Pourdehnad M, et al. The role of positron emission tomography with fluorine-18-deoxyglucose in identifying colorectal cancer metastases to liver. Nuc Med Commun 2000;21:793–8.
- Robinson PJ. Imaging liver metastases: current limitations and future prospects. Br J Radiol 2001;73:234–41.
This article has been cited by other articles:
|
|
|
|
 
S. A. Patterson, H. I. Khalil, and D. M. Panicek
MRI evaluation of small hepatic lesions in women with breast cancer.
Am. J. Roentgenol.,
August 1, 2006;
187(2):
307 - 312.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
H. I. Khalil, S. A. Patterson, and D. M. Panicek
Hepatic Lesions Deemed Too Small to Characterize at CT: Prevalence and Importance in Women with Breast Cancer
Radiology,
June 1, 2005;
235(3):
872 - 878.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Introduction
