| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Letters to the Editor |
Department of Radiology, Gifu University Hospital, 1–1 Yanagido, Gifu 501-1193, Japan. e-mail: masa-gif@umin.ac.jp
Editor:
Dr Hussain and colleagues are to be congratulated for their splendid article on the evaluation of the diagnostic value of T2-weighted magnetic resonance (MR) imaging for the detection and characterization of focal lesions in the cirrhotic liver in the March 2004 issue of Radiology (1). It was concluded that T2-weighted MR imaging does not provide added diagnostic value for the detection and characterization of focal lesions in cirrhosis. This conclusion appears somewhat misleading, despite their supplementary statement in the last paragraph of the discussion that the authors still include T2-weighted MR imaging in their liver protocol.
We take issue with several methodologic points. First, as is common in receiver operating characteristic (ROC) curve analysis, blinded observers should be instructed to assign a confidence level or a percentage for the presence of disease—that is, lesion(s) was (were) definitely absent, probably absent, probably present, or definitely present when using a four-point scale. However, observers in the study in question were instructed to assign a confidence level based on a four-point scale corresponding with the order: definitely benign, probably benign, possibly malignant, or definitely malignant. It is our understanding that the ROC method is valid only when observers’ confidence levels are recorded for lesion detection, and that it is invalid for lesion characterization (2).
Moreover, observer performances were evaluated by using 55 pathologically proved malignant and benign lesions, but the sizes of these lesions were somewhat larger than those often found to be problematic in the clinical setting. Indeed, gadolinium-enhanced MR imaging may have been sufficient for the diagnosis of these lesions. Given the sizes of these lesions, it is no surprise that T2-weighted MR images provided little added value. In addition, observers in the study detected a total of 107–113 lesions in the liver during image review, but no observer performance study was performed for them all. Small lesions (7 mm or smaller) are often found on gadolinium-enhanced images in daily practice, and it is often impossible to differentiate hepatocellular lesions (eg, hepatocellular carcinomas [HCCs], dysplastic nodules, regenerating nodules) from nonhepatocellular lesions (eg, metastases, cavernous hemangiomas, cysts) on gadolinium-enhanced MR images alone, because the signal intensities of small lesions are readily affected by their locations between sections and by the resultant volume averaging effect. However, T2-weighted MR imaging is often robust and helpful for discrimination of small hemangiomas or cysts from hepatocellular lesions in cirrhosis, which significantly increases hepatocellular lesion detection specificity in cirrhosis (3,4). We were disappointed that this important role of T2-weighted MR imaging was not fairly reflected by the concluding statement.
Further, no mention was made of early-enhancing pseudolesions in cirrhosis, which often cause dilemmas in the differential diagnosis of hypervascular tumors and benign perfusion abnormalities. Early-enhancing pseudolesions are known to be associated with arterioportal shunting, portal venous obstruction, non–portal splanchnic vein drainage, or compression, and the discrimination of small rounded early-enhancing pseudolesions from a malignant tumor is occasionally difficult (5,6). T2-weighted MR images often help radiologists to discriminate pseudolesions because most are isointense on T2-weighted images (6).
Finally, no description of any detailed parameters of T2-weighted MR imaging was provided, such as number of signals acquired, acquisition time, or section thickness. If MR examination time is shortened to increase patient throughput, the number of signals acquired might be reduced to the detriment of image quality. We raise this topic because the image quality of the T2-weighted MR images shown in their figures appeared to be somewhat lower than should be achieved with a modern MR imager. Thus, we are firmly of the opinion that optimally obtained T2-weighted MR images work rather better than indicated.
To conclude, we respectfully suggest that the most important issues for discussion concerning the additional value of T2-weighted MR imaging for the diagnosis of focal lesions in cirrhosis include performing valid observer performance studies, incorporating observer performance for small nonhepatocellular lesions, and maintaining the image quality of T2-weighted MR images.
REFERENCES
Department of Radiology/MRI, University of Michigan Health System, 1500 East Medical Center Drive, UH B2A209K, Ann Arbor, MI 48109. e-mail: hhussain@umich.edu
We thank Dr Kanematsu and colleagues for their interest in our article (1) and their insightful comments. We would like to respond to their comments as follows.
With regard to the use of ROC curves for lesion characterization, an ROC curve is a plot of sensitivity versus 1 minus specificity for all the possible thresholds an observer could make. The threshold is the boundary that separates a hypothetical binary decision, such as diseased versus healthy.
ROC analysis originated in signal detection theory and has found much use in radiologic sciences. In the detection of disease, one can think of the population being studied as partitioned into two subpopulations, one without disease and one with disease. Some diagnostic test is applied to a random sample from both populations. When comparing two imaging modalities, the difference in the area under the ROC curve is used as a relative measure of the abilities of the different modalities to allow the two subpopulations to be distinguished. As such, ROC methodology has a greater applicability than simply detection of the absence or presence of a lesion.
ROC analysis has been used to study the relative accuracy of biomarkers for the detection of pancreatic cancer (2), to determine if childhood basal metabolic index could predict adult obesity (3), to compare MR imaging and transrectal ultrasonography (US) for staging of prostate cancer (4), and to compare computed tomography and MR imaging for staging of lung cancer (5,6). The last two applications in particular are not lesion detection but, in a sense, lesion characterization. The issue boils down to the population being studied. If the population consists of a mix of subjects with no lesions and subjects with lesions, then lesion detection is the question to be answered with ROC analysis. If the population consists of subjects with confirmed lesions, then the question at hand would be if the modality or modalities can allow benign and malignant lesions to be distinguished or if they can help stage the extent of disease.
In our study, we included only lesions with pathologic proof of diagnosis and therefore did not include smaller non–biopsy-proved lesions for analysis. Much to our surprise, some of the larger HCCs did not show signal intensity abnormality on T2-weighted MR images. The cirrhotic liver differs significantly from the normal liver in that the signal intensity of the background liver is often heterogeneous on T2-weighted images, probably as a result of hepatic fibrosis, resulting in reduced conspicuity of small and mildly hyperintense lesions. Furthermore, unlike the noncirrhotic liver, T2-weighted MR images of the cirrhotic liver can be degraded by breathing artifacts, especially in sick patients in whom respiratory triggering fails because of their irregular breathing pattern. Ghosting artifact is also exaggerated in the presence of ascites.
MR imaging at our institution is not used as a screening modality for the detection of HCC in patients with cirrhosis; rather, it is used to assess patients suspected of having HCC because of elevated 
-fetoprotein level, a mass detected at US, or poor clinical health. Therefore, a selection bias toward sicker and less cooperative patients may have been introduced.
We agree with Drs Kanematsu and Goshima that T2-weighted MR imaging is helpful for the diagnosis of cysts and hemangiomas and have clearly mentioned in the last paragraph of the article that "the main value of T2-weighted imaging may lie in its usefulness for the diagnosis of cysts, hemangiomas, and lymphadenopathy" (1). It should be remembered, however, that hemangiomas are uncommon in cirrhosis (7). We also excluded both cysts and hemangiomas from our analysis and only included solid nodular lesions.
The purpose of our study was not to address small arterially enhancing lesions, and we have not found T2-weighted MR imaging to be particularly useful in the discrimination of pseudolesions from true lesions. Small (
2-cm) arterially enhancing HCCs have been shown in several studies to be isointense relative to adjacent liver parenchyma on T2-weighted MR images (8–10), and arterioportal shunts can be hyperintense (11). Since many of these arterially enhancing lesions are too small to be seen and sampled for biopsy with ultrasonographic guidance, management involves follow-up MR imaging to assess interval growth, which is the best indicator for malignancy (8,9).
Our parameters for T2-weighted fast spin-echo MR imaging were a repetition time of up to 5000 msec, effective echo time of 86–92 msec, echo train length of eight, section thickness of 8 mm, gap of 0–2 mm, matrix of 256 x 224–256, four signals acquired, receiver bandwidth of 31.25 kHz, acquisition time of 4–9 minutes, respiratory triggering, flow compensation, and fat suppression. We feel that we have selected appropriate parameters for respiratory-triggered T2-weighted fast spin-echo MR imaging (12), but the quality of the images is often suboptimal compared with T2-weighted MR images of the noncirrhotic liver for the reasons described earlier. Furthermore, lesions of the size we showed in our two examples of HCC (4 and 5 cm) should have been visible, even if the T2-weighted MR images are suboptimal, as we felt they were because of the reasons outlined in the figure legends.
REFERENCES
2-cm) early-enhancing hepatic lesions with serial contrast-enhanced dynamic MR imaging. Radiology 2003; 226:550-555.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
