Linda Moy

Supervised learning from multiple experts: whom to trust when everyone lies a bit

We describe a probabilistic approach for supervised learning when we have multiple experts/annotators providing (possibly noisy) labels but no absolute gold standard. The proposed algorithm evaluates the different experts and also gives an estimate of the actual hidden labels. Experimental results indicate that the proposed method is superior to the commonly used majority voting baseline.

Intravoxel incoherent motion imaging of tumor microenvironment in locally advanced breast cancer

Diffusion‐weighted imaging plays important roles in cancer diagnosis, monitoring, and treatment. Although most applications measure restricted diffusion by tumor cellularity, diffusion‐weighted imaging is also sensitive to vascularity through the intravoxel incoherent motion effect. Hypervascularity can confound apparent diffusion coefficient measurements in breast cancer. We acquired multiple b‐value diffusion‐weighted imaging at 3 T in a cohort of breast cancer patients and performed biexponential intravoxel incoherent motion analysis to extract tissue diffusivity (Dt), perfusion fraction (fp), and pseudodiffusivity (Dp). Results indicated significant differences between normal fibroglandular tissue and malignant lesions in apparent diffusion coefficient mean (±standard deviation) values (2.44 ± 0.30 vs. 1.34 ± 0.39 μm2/msec, P < 0.01) and Dt (2.36 ± 0.38 vs. 1.15 ± 0.35 μm2/msec, P < 0.01). Lesion diffusion‐weighted imaging signals demonstrated biexponential character in comparison to monoexponential normal tissue. There is some differentiation of lesion subtypes (invasive ductal carcinoma vs. other malignant lesions) with fp (10.5 ± 5.0% vs. 6.9 ± 2.9%, P = 0.06), but less so with Dt (1.14 ± 0.32 μm2/msec vs. 1.18 ± 0.52 μm2/msec, P = 0.88) and Dp (14.9 ± 11.4 μm2/msec vs. 16.1 ± 5.7 μm2/msec, P = 0.75). Comparison of intravoxel incoherent motion biomarkers with contrast enhancement suggests moderate correlations. These results suggest the potential of intravoxel incoherent motion vascular and cellular biomarkers for initial grading, progression monitoring, or treatment assessment of breast tumors. Magn Reson Med, 2011.

Is breast MRI helpful in the evaluation of inconclusive mammographic findings

OBJECTIVE The purpose of this study was to evaluate the usefulness of MRI of the breast in cases in which mammographic or sonographic findings are inconclusive. MATERIALS AND METHODS We retrospectively reviewed images from 115 MRI examinations of the breast performed from 1999 to 2005 for the indication of problem-solving for inconclusive findings on a mammogram. Forty-eight of the 115 women (41.8%) were at high risk. We discerned whether sonography or MRI was used as an adjunctive tool and correlated the findings with those in the pathology database. RESULTS The equivocal findings most frequently leading to MRI were asymmetry and architectural distortion. No suspicious MRI correlate was found in 100 of 115 cases (87%). These cases were found stable at follow-up mammography or MRI after a mean of 34 months. Fifteen enhancing masses (13%) that corresponded to the mammographic abnormality were seen on MR images. All masses identified at MRI were accurately localized for biopsy, and six malignant lesions were identified. Four of six malignant tumors were seen in one mammographic view only; two were seen on second-look ultrasound images. MRI had a sensitivity of 100% and compared with mammography had significantly higher specificity (91.7% vs 80.7%, p = 0.029), positive predictive value (40% vs 8.7%, p = 0.032), and overall accuracy (92.2% vs 78.3%, p = 0.0052). Eighteen incidental lesions (15.7%) were detected at MRI, and all were subsequently found benign. CONCLUSION We found breast MRI to be a useful adjunctive tool when findings at conventional imaging were equivocal. Strict patient selection criteria should be used because of the high frequency of incidental lesions seen on MR images.

Improving Specificity of Breast MRI Using Prone PET and Fused MRI and PET 3D Volume Datasets

MRI is a sensitive method for detecting invasive breast cancer, but it lacks specificity. To examine the effect of combining PET with MRI on breast lesion characterization, a prototype positioning device was fabricated to allow PET scans to be acquired in the same position as MRI scans—that is, prone. Methods: To test the hypothesis that fusion of 18F-FDG PET and MRI scans improves detection of breast cancer, 23 patients with suspected recurrent or new breast cancer underwent a routine whole-body PET scan, a prone PET scan of the chest, and a routine breast MRI scan. The attenuation-corrected prone PET and MRI datasets were registered twice by different operators. The fusion results were judged for quality by visual inspection and statistical analysis. A joint reading of the MRI and PET scans side by side and integrated images was performed by a nuclear medicine physician and a radiologist. Sensitivity and specificity of MRI and combined MRI and PET scans were calculated on the basis of pathology reports or at least 1 y of clinical and radiologic follow-up. Results: All fusions were verified to be well matched using specific anatomic criteria. A total of 45 lesions was assessed. Lesion size range was 0.6 to 10.0 cm. Of the 44 breasts examined, 29 were suspicious for cancer, of which 15 were found to be positive on surgical excision. In lesion-by-lesion analysis, sensitivity and specificity of MRI alone were 92% and 52%, respectively; after MRI and PET fusion, they were 63% and 95%, respectively. The positive predictive value and the negative predictive value for MRI alone were 69% and 85%, respectively; after MRI and PET fusion, they were 94% and 69%, respectively. Conclusion: Acquisition of prone PET scans using the new positioning device permitted acquisition of prone scans suitable for fusion with breast MRI scans. Fused PET and MRI scans increased the specificity of MRI but decreased the sensitivity in this small group of patients. Additional data are needed to confirm the statistical significance of these preliminary findings.

Comparison of fitting methods and b‐value sampling strategies for intravoxel incoherent motion in breast cancer

To compare fitting methods and sampling strategies, including the implementation of an optimized b‐value selection for improved estimation of intravoxel incoherent motion (IVIM) parameters in breast cancer.

Role of Fusion of Prone FDG‐PET and Magnetic Resonance Imaging of the Breasts in the Evaluation of Breast Cancer

Abstract:  The purpose of this study is to report further about the statistically significant results from a prospective study, which suggests that fusion of prone F‐18 Fluoro‐deoxy‐glucose (FDG) positron emission tomography (PET) and magnetic resonance (MR) breast scans increases the positive predictive value (PPV) and specificity for patients in whom the MR outcome alone would be nonspecific. Thirty‐six women (mean age, 43 years; range, 24–65 years) with 90 lesions detected on MR consented to undergo a FDG‐PET scan. Two blinded readers evaluated the MR and the computer tomography (CT) attenuation‐corrected prone FDG‐PET scans side‐by‐side, then after the volumes were superimposed (fused). A semiautomatic, landmark‐based program was used to perform nonrigid fusion. Pathology and radiologic follow‐up were used as the reference standard. The sensitivity, specificity, PPV, negative predictive value (NPV), and accuracy (with 95% confidence intervals) for MR alone, FDG‐PET alone, and fused MR and FDG‐PET were calculated. The median lesion size measured from the MR was 2.5 cm (range, 0.5–10 cm). Histologically, 56 lesions were malignant, and 15 were benign. Nineteen lesions were benign after 20–47 months of clinical and radiologic surveillance. The sensitivity of MR alone was 95%, FDG‐PET alone was 57%, and fusion was 83%. The increase in PPV from 77% in MR alone to 98% when fused and the increase in specificity from 53% to 97% were statistically significant (p < 0.05). The false‐negative rate on FDG‐PET alone was 26.7%, and after fusion this number was reduced to 9%. FDG‐PET and MR fusions were helpful in selecting which lesion to biopsy, especially in women with multiple suspicious MR breast lesions.

Ductal Carcinoma in Situ of the Breasts: Review of MR Imaging Features

The incidence of ductal carcinoma in situ (DCIS) has increased over the past few decades and now accounts for over 20% of newly diagnosed cases of breast cancer. Although the detection of DCIS has increased with the advent of widespread mammography screening, it is essential to have a more accurate assessment of the extent of DCIS for successful breast conservation therapy. Recent studies evaluating the detection of DCIS with magnetic resonance (MR) imaging have used high spatial resolution techniques and have increasingly been performed to screen a high-risk population as well as to evaluate the extent of disease. This work has shown that MR imaging is the most sensitive modality currently available for identifying DCIS and is more accurate than mammography in evaluating the extent of DCIS. MR imaging is particularly sensitive for identifying high-grade and intermediate-grade DCIS. DCIS may have variable morphologic features on MR images, with non-mass enhancement morphology being the most common manifestation. Less commonly, DCIS may also manifest as a mass on MR images, in which case it is most likely to be irregular. The kinetics of DCIS are also variable, with fast uptake and a plateau curve reported as the most common kinetic pattern. Additional MR imaging tools such as diffusion-weighted imaging and quantitative kinetic analysis combined with the benefit of high field strength, such as 3 T, may increase the sensitivity and specificity of breast MR imaging in the detection of DCIS.

Current Status of Hybrid PET/MRI in Oncologic Imaging

OBJECTIVE This review article explores recent advancements in PET/MRI for clinical oncologic imaging. CONCLUSION Radiologists should understand the technical considerations that have made PET/MRI feasible within clinical workflows, the role of PET tracers for imaging various molecular targets in oncology, and advantages of hybrid PET/MRI compared with PET/CT. To facilitate this understanding, we discuss clinical examples (including gliomas, breast cancer, bone metastases, prostate cancer, bladder cancer, gynecologic malignancy, and lymphoma) as well as future directions, challenges, and areas for continued technical optimization for PET/MRI.

Outcome of High-Risk Lesions at MRI-Guided 9-Gauge Vacuum- Assisted Breast Biopsy

OBJECTIVE The purposes of this study were to determine the frequency of underestimation of high-risk lesions at MRI-guided 9-gauge vacuum-assisted breast biopsy and to determine the imaging and demographic characteristics predictive of lesion upgrade after surgery. MATERIALS AND METHODS We retrospectively reviewed consecutively detected lesions that were found only at MRI and biopsied under MRI guidance from May 2007 to April 2012. Imaging indications, imaging features, and histologic findings were reviewed. The Fisher exact test was used to assess the association between characteristics and lesion upgrade. Patients lost to follow-up or who underwent mastectomy were excluded, making the final study cohort 140 women with 151 high-risk lesions, 147 of which were excised. RESULTS A database search yielded the records of 1145 lesions in 1003 women. Biopsy yielded 252 (22.0%) malignant tumors, 184 (16.1%) high-risk lesions, and 709 (61.9%) benign lesions. Thirty of the 147 (20.4%) excised high-risk lesions were upgraded to malignancy. The upgrade rate was highest for atypical ductal hyperplasia, lobular carcinoma in situ, and radial scar. No imaging features were predictive of upgrade. However, there was a significantly higher risk that a high-risk lesion would be upgraded to malignancy if the current MRI-detected high-risk lesion was in the same breast as a malignant tumor previously identified in the remote history, a recently diagnosed malignant tumor, or a high-risk lesion previously identified in the remote history (p = 0.0001). The upgrade rate was significantly higher for women with a personal cancer history than for other indications combined (p = 0.0114). CONCLUSION The rate of underestimation of malignancy in our series was 20%. No specific imaging features were seen in upgraded cases. Surgical excision is recommended for high-risk lesions found at MRI biopsy and may be particularly warranted for women with a personal history of breast cancer.

Microinvasive ductal carcinoma in situ: Clinical presentation, imaging features, pathologic findings, and outcome

OBJECTIVE The purpose of our study was to describe the clinical features, imaging characteristics, pathologic findings and outcome of microinvasive ductal carcinoma in situ (DCISM). MATERIALS AND METHODS The records of 21 women diagnosed with microinvasive ductal carcinoma in situ (DCISM) from November 1993 to September 2006 were retrospectively reviewed. The clinical presentation, imaging and histopathologic features, and clinical follow-up were reviewed. RESULTS The 21 lesions all occurred in women with a mean age of 56 years (range, 27-79 years). Clinical findings were present in ten (48%): 10 with palpable masses, four with associated nipple discharge. Mean lesion size was 21mm (range, 9-65mm). The lesion size in 62% was 15mm or smaller. Mammographic findings were calcifications only in nine (43%) and an associated or other finding in nine (43%) [mass (n=7), asymmetry (n=1), architectural distortion (n=1)]. Three lesions were mammographically occult. Sonographic findings available in 11 lesions showed a solid hypoechoic mass in 10 cases (eight irregular in shape, one round, one oval). One lesion was not seen on sonography. On histopathologic examination, all lesions were diagnosed as DCISM, with a focus of invasive carcinoma less than or equal to 1mm in diameter within an area of DCIS. Sixteen (76%) lesions were high nuclear grade, four (19%) were intermediate and one was low grade (5%). Sixteen (76%) had the presence of necrosis. Positivity for ER and PR was noted in 75% and 38%. Nodal metastasis was present in one case with axillary lymph node dissection. Mean follow-up time for 16 women was 36 months without evidence of local or systemic recurrence. One patient developed a second primary in the contralateral breast 3 years later. CONCLUSION The clinical presentation and radiologic appearance of a mass are commonly encountered in DCISM lesions (48% and 57%, respectively), irrespective of lesion size, mimicking findings seen in invasive carcinoma. Despite its potential for nodal metastasis (5% in our series), mean follow-up at 36 months was good with no evidence of local or systemic recurrence at follow-up. Knowledge of these clinical and imaging findings in DCISM lesions may alert the clinician to the possibility of microinvasion and guide appropriate management.