Enhancement Type at Contrast-enhanced Mammography and Association with Malignancy.

Background Despite the increasing use of contrast-enhanced mammography (CEM), there are limited data on the evaluation of findings on recombined images and the association with malignancy. Purpose To determine the rates of malignancy of enhancement findings on CEM images in the presence or absence of low-energy findings using the Breast Imaging Reporting and Data System (BI-RADS) lexicon developed for mammography and MRI. Materials and Methods All diagnostic CEM examinations performed at one academic institution between December 2015 and December 2019 had low-energy and recombined images retrospectively. Data were independently reviewed by three breast imaging radiologists with 5-25 years of experience using the BI-RADS mammography and MRI lexicon. Outcome was determined with pathologic analysis or 1-year imaging or clinical follow-up. The χ2 and Fisher exact tests were used for analysis. Results A total of 371 diagnostic CEM studies were performed in 371 women (mean age, 54 years ± 11[SD]). Sensitivity, specificity, positive predictive value (PPV), and negative predictive value of enhancement on CEM images was 95% (104 of 109 [95% CI: 90, 98]), 67% (176 of 262 [95% CI: 61, 73]), 55% (104 of 190 [95% CI: 47, 62]), and 97% (176 of 181 [95% CI: 94, 99]), respectively. Among 190 CEM studies with enhancing findings, enhancing lesions were more likely to be malignant when associated with low-energy findings (26% vs 59%, P < .001). Among enhancement types, mass enhancement composed 71% (99 of 140) of all malignancies with PPV of 63% when associated with low-energy findings. Foci, non-mass enhancement, and mass enhancement without low-energy findings had PPV of 6%, 24%, and 38%, respectively. Neither background parenchymal enhancement nor density was associated with enhancement type (P = .19 and P = .28, respectively). Conclusion Mass enhancement on recombined images using CEM was most commonly associated with malignancy, especially when associated with low-energy findings. Enhancement types were more likely to be benign when not associated with low-energy findings; however, they should still be viewed with suspicion, given the high association with malignancy. © RSNA, 2022 Online supplemental material is available for this article.

Breast MRI for Evaluation of Response to Neoadjuvant Therapy.

Neoadjuvant therapy is increasingly being used to treat early-stage triple-negative and human epidermal growth factor 2-overexpressing breast cancers, as well as locally advanced and inflammatory breast cancers. The rationales for neoadjuvant therapy are to shrink tumor size and potentially decrease the extent of surgery, to serve as an in vivo test of response to therapy, and to reveal prognostic information for the patient. MRI is the most accurate modality to demonstrate response to therapy and to help ensure accurate presurgical planning. Changes in lesion diameter, volume, and enhancement are used to predict complete response, partial response, or nonresponse to therapy. However, residual disease may be overestimated or underestimated at MRI. Fibrosis, necrotic tumors, and residual benign masses may be causes of overestimation of residual disease. Nonmass lesions, invasive lobular carcinoma, hormone receptor-positive tumors, nonconcentric shrinkage patterns, the use of antiangiogenic therapy, and late-enhancing foci may be causes of underestimation of residual disease. In patients with known axillary lymph node metastasis, neoadjuvant therapy may be followed by targeted axillary dissection to avoid the potential morbidity associated with an axillary lymph node dissection. Diffusion-weighted imaging, radiomics, machine learning, and deep learning methods are under investigation to improve MRI accuracy in predicting treatment response.©RSNA, 2021.

An Efficient Deep Neural Network to Classify Large 3D Images With Small Objects.

3D imaging enables accurate diagnosis by providing spatial information about organ anatomy. However, using 3D images to train AI models is computationally challenging because they consist of 10x or 100x more pixels than their 2D counterparts. To be trained with high-resolution 3D images, convolutional neural networks resort to downsampling them or projecting them to 2D. We propose an effective alternative, a neural network that enables efficient classification of full-resolution 3D medical images. Compared to off-the-shelf convolutional neural networks, our network, 3D Globally-Aware Multiple Instance Classifier (3D-GMIC), uses 77.98%-90.05% less GPU memory and 91.23%-96.02% less computation. While it is trained only with image-level labels, without segmentation labels, it explains its predictions by providing pixel-level saliency maps. On a dataset collected at NYU Langone Health, including 85,526 patients with full-field 2D mammography (FFDM), synthetic 2D mammography, and 3D mammography, 3D-GMIC achieves an AUC of 0.831 (95% CI: 0.769-0.887) in classifying breasts with malignant findings using 3D mammography. This is comparable to the performance of GMIC on FFDM (0.816, 95% CI: 0.737-0.878) and synthetic 2D (0.826, 95% CI: 0.754-0.884), which demonstrates that 3D-GMIC successfully classified large 3D images despite focusing computation on a smaller percentage of its input compared to GMIC. Therefore, 3D-GMIC identifies and utilizes extremely small regions of interest from 3D images consisting of hundreds of millions of pixels, dramatically reducing associated computational challenges. 3D-GMIC generalizes well to BCS-DBT, an external dataset from Duke University Hospital, achieving an AUC of 0.848 (95% CI: 0.798-0.896).

Impact of imaging nurse navigation on breast interventions: Direct and closed-loop patient communication and documentation.

OBJECTIVE: A breast imaging nurse navigator (NN) was established with the goals to enhance the patient experience after biopsy, improve care timeliness, accuracy, and coordination, facilitate direct communication to patients, and increase care retention within our system. Our aim was to determine the impact of NN on patient care time metrics, communication, documentation, compliance, and care retention at our institution after breast biopsy. METHODS: Retrospective review of a six-month period before (5/1/17-10/31/17) and after (5/1/19-10/31/19) establishment of a nurse navigator in our breast imaging department was performed, including 498 patients in the pre-navigation (pre-NN) group and 526 patients in the post-navigation (post-NN) group. Data was gathered from the electronic medical record and collected using REDCap. RESULTS: Biopsy pathology results were communicated directly to the patient more often post-NN (71%, 374/526) compared to pre-NN (4%, 21/498) (p < 0.0001), without change in overall time of result communication (p = 0.08). Due to factors outside of imaging, most care time metrics were longer post-NN, including days from biopsy to pathology report (p < 0.001), result communication to initiation of care (p < 0.001), and biopsy to surgery (p < 0.001). There was no difference and high compliance (p = 1) and care retention (p = 0.015) in both groups. There was improved documentation of pathology results, recommendations, and communication post-NN (0/526 vs 10/498, p = 0.001). CONCLUSION: Imaging nurse navigation added greatest value by communicating breast biopsy results and recommendations directly to patients and ensuring documentation. Compliance and retention were high in both groups. Factors outside of Radiology influenced time metrics, requiring further investigation of multidisciplinary collaboration.

Differences between human and machine perception in medical diagnosis.

Deep neural networks (DNNs) show promise in image-based medical diagnosis, but cannot be fully trusted since they can fail for reasons unrelated to underlying pathology. Humans are less likely to make such superficial mistakes, since they use features that are grounded on medical science. It is therefore important to know whether DNNs use different features than humans. Towards this end, we propose a framework for comparing human and machine perception in medical diagnosis. We frame the comparison in terms of perturbation robustness, and mitigate Simpson's paradox by performing a subgroup analysis. The framework is demonstrated with a case study in breast cancer screening, where we separately analyze microcalcifications and soft tissue lesions. While it is inconclusive whether humans and DNNs use different features to detect microcalcifications, we find that for soft tissue lesions, DNNs rely on high frequency components ignored by radiologists. Moreover, these features are located outside of the region of the images found most suspicious by radiologists. This difference between humans and machines was only visible through subgroup analysis, which highlights the importance of incorporating medical domain knowledge into the comparison.

Contrast-enhanced Mammography: A Guide to Setting Up a New Clinical Program.

Contrast-enhanced mammography (CEM) is gaining rapid traction following the U.S. Food and Drug Administration approval for diagnostic indications. Contrast-enhanced mammography is an alternative form of mammography that uses a dual-energy technique for image acquisition after the intravenous administration of iodinated contrast material. The resulting exam includes a dual set of images, one that appears similar to a routine 2D mammogram and one that highlights areas of contrast uptake. Studies have shown improved sensitivity compared to mammography and similar performance to contrast-enhanced breast MRI. As radiology groups incorporate CEM into clinical practice they must first select the indications for which CEM will be used. Many practices initially use CEM as an MRI alternative or in cases recommended for biopsy. Practices should then define the CEM clinical workflow and patient selection to include ordering, scheduling, contrast safety screening, and managing imaging on the day of the exam. The main equipment requirements for performing CEM include CEM-capable mammography equipment, a power injector for contrast administration, and imaging-viewing capability. The main staffing requirements include personnel to place the intravenous line, perform the CEM exam, and interpret the CEM. To safely and appropriately perform CEM, staff must be trained in their respective roles and to manage potential contrast-related events. Lastly, informing referring colleagues and patients of CEM through marketing campaigns is helpful for successful implementation.

Semi-Automated 3D Volumetric Renal Measurements in Polycystic Kidney Disease Using b0-Images-A Feasibility Study.

Autosomal dominant polycystic kidney disease (ADPKD) eventually leads to end stage renal disease (ESRD) with an increase in size and number of cysts over time. Progression to ESRD has previously been shown to correlate with total kidney volume (TKV). An accurate and relatively simple method to perform measurement of TKV has been difficult to develop. We propose a semi-automated approach of calculating TKV inclusive of all cysts in ADPKD patients based on b0 images relatively quickly without requiring any calculations or additional MRI time. Our purpose is to evaluate the reliability and reproducibility of our method by raters of various training levels within the environment of an advanced 3D viewer. Thirty patients were retrospectively identified who had DWI performed as part of 1.5T MRI renal examination. Right and left TKVs were calculated by five radiologists of various training levels. Interrater reliability (IRR) was estimated by computing the intraclass correlation (ICC) for all raters. ICC values calculated for TKV measurements between the five raters were 0.989 (95% CI = (0.981, 0.994), p < 0.01) for the right and 0.961 (95% CI = (0.936, 0.979), p < 0.01) for the left. Our method shows excellent intraclass correlation between raters, allowing for excellent interrater reliability.