CAD Analysis of Breast MRI May Predict Distant Metastases

Becky McCall

March 14, 2011

March 14, 2011 (Vienna, Austria) — The results of 2 studies using computer-assisted diagnosis (CAD) to investigate the prediction of lymph node involvement and breast cancer metastases in breast cancer and to allow the depiction of morphologic characteristics were presented here at the European Congress of Radiology 2011.

The studies showed that the morphologic characteristics of skin thickening and internal enhancement, as well as the increased washout identified by CAD, had prognostic value in predicting lymph node metastases (LNM) and metachronous distant metastases, respectively, in breast cancer.

Matthias Dietzel, MD, from the Institute of Diagnostic and Interventional Radiology, Friedrich-Schiller-University, Jena, Germany, was principal investigator of the 2 studies. He explained that magnetic resonance imaging (MRI) correlates with histopathologic features and is usually used to differentiate between benign and malignant lesions.

In these studies, he asked whether MRI could also be used to identify prognostic features.

"If this is so, we could then use a standard MRI examination as a prognostic tool, which would not incur extra scanning time or associated costs. . . .  Such findings could then be included in a standard radiological report. Our method involves a single MRI exam, as currently used worldwide, without any fancy scanning or protocols. This is the beauty of this method — it takes 10 minutes," Dr. Dietzel told Medscape Medical News.

"According to our results, the features used for standard diagnosis can also be used to obtain prognostic information. To date, this has not been well understood or investigated," he added.

In the first study, Dr. Dietzel aimed to correlate a detailed catalogue of 17 descriptors in magnetic resonance mammography with the presence of LNM, and to identify a combination of such descriptors that can be used with a decision tree to predict LNM.

"The end point of the study investigating individual descriptors was the presence of [LNM], which is a very important prognostic factor known to decrease breast cancer survival and disease-free survival," said Dr. Dietzel.

Seventeen previously published descriptors were assessed using MRI, and included blooming sign, shape of lesion, skin thickening, destruction of nipple line, vessels, internal septation, signal intensity, edema, washin, and washout. Of all the primary invasive cancers investigated, there were 253 cancers without and 97 with LNM.

On analysis, the researchers found that 7 of the 17 descriptors were significantly associated with LNM; the majority (6 of 7) had not been used for this purpose previously. Skin thickening and internal enhancement were found to be the most accurate descriptors (P < .001; diagnostic odds ratio, 5.9 and 13.7, respectively); and the decision tree identified skin thickening plus destruction of nipple line as a useful descriptor combination, which raised the probability of the presence of LNM by 40 % (P < .001).

In the case of the absence of skin thickening, edema, and irregular margins, the likelihood of LNM was 0% (P < .05).

The second study presented by Dr. Dietzel investigated prognostic factors in terms of distant metastases from a primary breast tumor and tumor vascularization, as assessed by breast MRI and CAD. Technically, breast MRI used repetitive T1-weighted sequences both before and after the application of a contrast agent.

Fifty-nine female patients (mean age, 56 years) participated in the study. Patients with secondary breast cancer and known malignant neoplasms other than breast cancer were excluded. The most common cancer type was invasive ductal (81.4%), no patients had synchronous distant metastases, and the median follow-up time was 1560 days.

Conventional breast MRI was conducted by experienced radiologists (>500 magnetic resonance mammographies), using standardized protocols, according to European Society of Breast Imaging recommendations. Parameters used were basic initial enhancement, level of washout, peak enhancement, and time to peak enhancement. The standard of reference was the occurrence of distant (metachronous) metastases at follow-up.

Results showed that the level of washout was the single independent parameter that predicted the occurrence of distant metastases (P < .05). Overall accuracy of washout was determined by an area under the curve of 0.76 (standard error, 0.08).

"Interestingly, basic parameters like washout, which is an important vascular parameter, emerged as a very accurate measure that indicates whether a patient is likely to have distant metastases. In our study, we were able to identify all patients likely to develop metastases. The negative predictive value and the sensitivity was 100%; however, the specificity is still a matter of debate, at 56.6%," reported Dr. Dietzel.

He concluded that the study identified that washout, a basic MRI feature, can identify a patient's prognosis in terms of development of metachronous metastases, in addition to differentiation of benign from malignant lesions. "These results might be used for risk stratification and help in clinical decision-making," he noted.

CAD reduces radiologist or human bias, but it can only analyze dynamic features, which represent one third of interesting features, he cautioned. "Morphology of a cancer is important, not only in histology but also in radiology, and the current CAD system cannot do this very well. Our associated study used morphology, which looked for features such as edema around the lesion and invasion of the ducts to further refine diagnostic accuracy. A major result was to use morphology as a complement to CAD," said Dr. Dietzel.

"These data show that CAD and morphology are more than diagnostic tools — they can also be used for prognosis — and they are free. It's a standard examination for the patient. . . . There is no need for further postprocessing, and this can identify a high-risk cancer," concluded Dr. Dietzel.

Commenting on the work, Maximilian Reiser, MD, director of the Institute of Clinical Radiology at the Ludwig Maximilian University of Munich Medical Center, Germany, and president of the European Society of Radiology, said studies that look at prognostic value are important because they can contribute to the assessment of the biologic significance of imaging results beyond the description of morphologic alterations.

"We are investigating 'imaging biomarkers' that allow for quantitative, validated, and reproducible characterization of disease, which enables provision of information concerning prognosis, selection of adequate therapeutic measures, and therapy response," Dr. Reiser said. The study by Dr. Dietzel and colleagues is "well in line with this paradigm. Before these results can be applied in clinical practice, confirmation by other researchers and prospective randomized trials are mandatory."

Dr. Dietzel and Dr. Reiser have disclosed no relevant financial relationships.

European Congress of Radiology (ECR) 2011: Abstract B-549. Presented March 6, 2011.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.