Liver MRI CAD speeds workflow, helps standardize interpretation

Magnetic resonance imaging is an established and important modality for the evaluation, assessment, and management of liver disease. It is one of the most sensitive tests for the detection and characterization of liver disease^1-4 and provides detailed anatomic information regarding the relationship of liver lesions to major vascular and biliary structures,⁵ significantly aiding in surgical planning.⁶

At our university-based practice, which is a major liver transplant center, we primarily evaluate patients with chronic cirrhosis being screened for hepatocellular carcinoma (HCC), being followed postsurgery or postablation for HCC, undergoing pretransplant evaluation, or awaiting liver transplant. Secondarily, we evaluate patients with liver lesions discovered on ultrasound or CT that need further characterization. Our clinical service currently performs approximately 100 liver MRI studies per month, and we find ourselves evaluating an increasing number of patients with MRI. One of the key growth drivers in the demand for liver MRI is an increasing awareness among clinicians and patients concerning the radiation risks associated with multiple CT examinations.^7,8

BENIGN AND MALIGNANT ENTITIES

The liver encompasses a wide range of benign and malignant entities, making liver MRI challenging to interpret. In a typical study, there are often 12 to 18 different sequences to review9 (multiplanar T2 and T2 fat-suppressed, in- and out-of-phase, diffusion with apparent diffusion coefficient [ADC] maps, multiplanar dynamic contrast-enhanced, and, in some cases, additional postcontrast delayed). When there are prior studies for comparison, at least portions of these must be viewed and, in many cases, manually aligned with the current study in order to determine disease progression and/or find and evaluate any lesions targeted for follow-up. In addition, image misregistration (i.e., liver in a different position on a different sequence, particularly on the dynamic series) further complicates the interpretation, especially when trying to detect small, subtle, or capsular lesions and look for enhancement in T1-bright lesions such as ablation sites (Figure 1).

Adding to the complexity, liver MRI studies often contain huge amounts of data, especially as MR imaging advances to include new techniques such as diffusion with ADC maps, perfusion imaging, and MR elastography.^10,11 Hence liver MRI studies often exceed 1500 images. The workstation/PACS management of these data sets can be time-consuming and complicated.

Additionally, there is significant (and often unrecognized) quantifiable information available in liver MRI, such as contrast enhancement metrics (e.g., wash-in and wash-out slopes, relative enhancement), ADC values, fat content, and iron quantification. These are often neglected in day-to-day liver MRI interpretation, largely because radiologists must spend extra time to extract such quantifiable information from the study-if doing so is even feasible, given their PACS environment.

Although the “traditional” pulse sequences (T1-weighted, T2-weighted, contrast-enhanced) have been refined and generally standardized over the years,⁹ body radiologists have, in general, been interpreting MRI studies in the same manner for years, with the exception of using PACS instead of film. Current practice includes browsing through all the images/sequences, finding abnormalities, correlating these abnormalities across different sequences and comparisons, and then generating a report with a rendered diagnosis or differential.

Lesion contrast enhancement and dynamic enhancement properties are usually determined by direct visualization of the gray-scale images, with only occasional use of a quantitative region of interest (ROI) to look at signal intensities. Images and dynamic series are evaluated by scrolling back and forth, and liver and lesion size measurements are manually
calculated.

While interpretation has been performed in the same manner for years, how images are organized and displayed is much more variable. Different radiologists use different hanging protocols that vary between exams and MR machines. In addition, because MR by nature lends itself to repeating, altering, or renaming sequences, hanging protocols often do not work well, and considerable time is spent moving the desired pulse sequence to the desired window for evaluation.

This lack of viewing layout standardization can waste valuable time. Furthermore, there can be significant variability in how liver MRI studies are reported. For example, I find some radiologists inconsistent in labeling the Couinaud segment in which a lesion is located, making follow-up confusing or more time-consuming at best, particularly when multiple lesions are present that must be individually cataloged. All of these inconsistencies can be considerable, making standardization a worthy goal.

AUTOMATED POSTPROCESSING

In my experience, consistent, automated postprocessing of studies not only provides standardization, it also reduces the amount of time spent analyzing the study. An equally important value of standardization is the increased availability of quantifiable data. In late 2009, we added CAD, or computer-aided detection (CADstream, Merge Healthcare, Hartland, WI), to our liver MRI program. This has allowed us to review our MRI exams in a more quantifiable, standardized manner and improve reporting for more comprehensive patient management.

I have experienced several workflow enhancements using CAD. Because studies are preprocessed according to our site-specific protocols, I can customize the interface itself to enhance interpretability while toggling between multiple screen layouts (including comparisons). The screens remain centered when moving from one layout to another, which is helpful, for example, when I find a lesion on A T2-weighted image and want to immediately view the dynamic contrast or diffusion images at that location. Furthermore, I have instant access to kinetic data (presented as signal-intensity graphs, Figure 2), liver and lesion size information, ADC maps, and liver segmentation with subsequent lesion localization for reporting (Figure 3). The kinetic data are automatically registered to correct for movement. This is essential for viewing subtraction images, particularly in a previously treated liver where lesions are often bright on T1-weighted images and any misregistration can appear as focal edge enhancement (Figure 1).

CAD enables automated cataloguing of lesions. A separate portfolio (set of images) is generated showing each lesion with size, kinetics, ADC, and segmental localization. This is particularly helpful when comparing serial exams in a complex cirrhotic patient in whom we are following multiple suspicious lesions, as visualized data are often easier to comprehend than a written report. Cataloguing also provides a solid foundation for a potential research database.

Because the use of CAD for liver MRI is new, research in this area is limited. In a study of 52 patients, physicians in Japan found magnetic resonance CAD is useful for diagnosing fibrosis of the liver.¹² Research presented at the Second International Conference of Imaging and Informatics showed that CAD is useful for diagnosing liver cancer, and that employing CAD in clinical practice could be expected to reduce mortality caused by liver cancer in Asian countries.¹³

Preliminary findings such as this with liver MRI CAD are encouraging, and I am intrigued by the potential for liver CAD to produce clinical benefits similar to those we have seen with breast MRI CAD. For example, CAD for breast MRI has been demonstrated to increase specificity by as much as 50% with no decrease in sensitivity.¹⁴ In addition, it can improve discrimination between benign and malignant lesions when compared with initial interpretation by radiologists.¹⁵ These are important advances for breast MRI, and I believe that with the development of the Liver Imaging Reporting and Data System (LI-RADS),¹⁶ we will see increased research in liver MRI and the application of CAD to assist radiologists in interpretation and standardization of these important studies.