PURPOSE: We propose a deep learning-based approach to breast mass classification in sonography and compare it with the assessment of four experienced radiologists employing breast imaging reporting and data system 4th edition lexicon and assessment protocol. METHODS: Several transfer learning techniques are employed to develop classifiers based on a set of 882 ultrasound images of breast masses. Additionally, we introduce the concept of a matching layer. The aim of this layer is to rescale pixel intensities of the grayscale ultrasound images and convert those images to red, green, blue (RGB) to more efficiently utilize the discriminative power of the convolutional neural network pretrained on the ImageNet dataset. We present how this conversion can be determined during fine-tuning using back-propagation. Next, we compare the performance of the transfer learning techniques with and without the color conversion. To show the usefulness of our approach, we additionally evaluate it using two publicly available datasets. RESULTS: Color conversion increased the areas under the receiver operating curve for each transfer learning method. For the better-performing approach utilizing the fine-tuning and the matching layer, the area under the curve was equal to 0.936 on a test set of 150 cases. The areas under the curves for the radiologists reading the same set of cases ranged from 0.806 to 0.882. In the case of the two separate datasets, utilizing the proposed approach we achieved areas under the curve of around 0.890. CONCLUSIONS: The concept of the matching layer is generalizable and can be used to improve the overall performance of the transfer learning techniques using deep convolutional neural networks. When fully developed as a clinical tool, the methods proposed in this paper have the potential to help radiologists with breast mass classification in ultrasound.

Background

Preoperative breast and lung markers have significant drawbacks, including migration, patient discomfort, and scheduling difficulties. SignalMark is a novel localizer device with a unique signal on Doppler ultrasound.

Objective

We aimed to evaluate intraoperative identification of SignalMark microspheres compared with HydroMARK^® clips. We also assessed the safety and efficacy of SignalMark in the lung.

Methods

Twelve breasts of lactating pigs were injected with SignalMark or HydroMARK^® by a breast radiologist, and subsequently identified using a standard ultrasound machine by three surgeons blinded to marker location. Time to identification of each marker was recorded, with a maximum allotted time of 300 s. To further demonstrate efficacy in lung parenchyma, a second cohort of pigs underwent lung injections.

Results

A total of eight SignalMark markers and four HydroMARK^® clips were placed in pig breasts. Overall, the surgeons correctly identified SignalMark 95.8% of the time (n = 23/24) and HydroMARK^® clips 41.7% of the time (n = 5/12) within 300 s (p < 0.001). The mean time to identification was significantly faster for SignalMark, at 80.8 ± 20.1 s, than for HydroMARK^®, at 209.4 ± 35.2 s (p < 0.002). For the lung injections, all 10 SignalMark markers were visible on Doppler ultrasound at the time of placement, and at the 7- and 21-day time points.

Conclusions

Surgeons identified SignalMark in significantly less time than HydroMARK^® clips in a simulated intraoperative setting, and SignalMark was easily viewed in the lung. These results suggest that SignalMark is a feasible option for efficient intraoperative localization of non-palpable breast and lung tumors using ultrasound guidance.

In this work, we propose a deep learning method for breast mass segmentation in ultrasound (US). Variations in breast mass size and image characteristics make the automatic segmentation difficult. To address this issue, we developed a selective kernel (SK) U-Net convolutional neural network. The aim of the SKs was to adjust network's receptive fields via an attention mechanism, and fuse feature maps extracted with dilated and conventional convolutions. The proposed method was developed and evaluated using US images collected from 882 breast masses. Moreover, we used three datasets of US images collected at different medical centers for testing (893 US images). On our test set of 150 US images, the SK-U-Net achieved mean Dice score of 0.826, and outperformed regular U-Net, Dice score of 0.778. When evaluated on three separate datasets, the proposed method yielded mean Dice scores ranging from 0.646 to 0.780. Additional fine-tuning of our better-performing model with data collected at different centers improved mean Dice scores by ~6%. SK-U-Net utilized both dilated and regular convolutions to process US images. We found strong correlation, Spearman's rank coefficient of 0.7, between the utilization of dilated convolutions and breast mass size in the case of network's expansion path. Our study shows the usefulness of deep learning methods for breast mass segmentation. SK-U-Net implementation and pre-trained weights can be found at github.com/mbyr/bus_seg.

Magnetic resonance imaging (MRI) is used for preoperative evaluation, high-risk screening, and other select indications for breast cancer. However, the interpretation of breast MR images in pregnant and lactating women is complicated by physiologic changes of the breast that may result in marked background enhancement. Breast MRI with contrast administration is contraindicated in pregnancy. Restriction spectrum imaging (RSI) is an advanced diffusion-weighted (DW)-MRI method that theoretically reflects signal from cells with high nuclear-to-cytoplasm ratio without gadolinium-based contrast. This report describes a case in which RSI notably increased tumor conspicuity in a lactating woman, compared to contrast-enhanced (CE)-MRI and conventional DW-MRI.

Rationale and objectives

Breast cancer is the leading type of cancer among women. Visualization and characterization of breast lesions based on vascularity kinetics was evaluated using three-dimensional (3D) contrast-enhanced ultrasound imaging in a clinical study.

Materials and methods

Breast lesions (n = 219) were imaged using power Doppler imaging (PDI), 3D contrast-enhanced harmonic imaging (HI), and 3D contrast-enhanced subharmonic imaging (SHI) with a modified Logiq 9 ultrasound scanner using a 4D10L transducer. Quantitative metrics of vascularity derived from 3D parametric volumes (based on contrast perfusion; PER and area under the curve; AUC) were generated by off-line processing of contrast wash-in and wash-out. Diagnostic accuracy of these quantitative vascular parameters was assessed with biopsy results as the reference standard.

Results

Vascularity was observed with PDI in 93 lesions (69 benign and 24 malignant), 3D HI in 8 lesions (5 benign and 3 malignant), and 3D SHI in 83 lesions (58 benign and 25 malignant). Diagnostic accuracy for vascular heterogeneity, PER, and AUC ranged from 0.52 to 0.75, while the best logistical regression model (vascular heterogeneity ratio, central PER, and central AUC) reached 0.90.

Conclusion

3D SHI successfully detects contrast agent flow in breast lesions and characterization of these lesions based on quantitative measures of vascular heterogeneity and 3D parametric volumes is promising.

Purpose: While mammography has excellent sensitivity for the detection of breast lesions, its specificity is limited. Adjunct screening with ultrasound may partially alleviate this issue but also increases false positives, resulting in unnecessary biopsies. Our study investigated the use of Google AutoML Vision (Mountain View, California), a commercially available machine learning service, to both identify and characterize indeterminate breast lesions on ultrasound. Approach: B-mode images from 253 independent cases of indeterminate breast lesions scheduled for core biopsy were used for model creation and validation. The performances of two sub-models from AutoML Vision, the image classification model and object detection model, were evaluated, while also investigating training strategies to enhance model performances. Pathology from the patients biopsy was used as a reference standard. Results: The image classification models trained under different conditions demonstrated areas under the precision-recall curve (AUC) ranging from 0.85 to 0.96 during internal validation. Once deployed, the model with highest internal performance demonstrated a sensitivity of 100% [95% confidence interval (CI) of 73.5% to 100%], specificity of 83.3% (CI=51.6% to 97.9%), positive predictive value (PPV) of 85.7% (CI=62.9% to 95.5%), and negative predictive value (NPV) of 100% (CI non-evaluable) in an independent dataset. The object detection model demonstrated lower performance internally during development (AUC=0.67) and during prediction in the independent dataset [sensitivity=75% (CI=42.8 to 94.5), specificity=80% (CI=51.9 to 95.7), PPV=75% (CI=50.8 to 90.0), and NPV=80% (CI=59.3% to 91.7%)], but was able to demonstrate the location of the lesion within the image. Conclusions: Two models appear to be useful tools for identifying and classifying suspicious areas on B-mode images of indeterminate breast lesions.

OBJECTIVE: Breast cancer is the most frequent type of cancer among women. This multi-center study assessed the ability of 3D contrast-enhanced ultrasound to characterize suspicious breast lesions using clinical assessments and quantitative parameters. METHODS: Women with suspicious breast lesions scheduled for biopsy were enrolled in this prospective, study. Following 2D grayscale ultrasound and power Doppler imaging (PDI), a contrast agent (Definity; Lantheus) was administrated. Contrast-enhanced 3D harmonic imaging (HI; transmitting/receiving at 5.0/10.0 MHz), as well as 3D subharmonic imaging (SHI; transmitting/receiving at 5.8/2.9 MHz), were performed using a modified Logiq 9 scanner (GE Healthcare). Five radiologists independently scored the imaging modes (including standard-of-care imaging) using a 7-point BIRADS scale as well as lesion vascularity and diagnostic confidence. Parametric volumes were constructed from time-intensity curves for vascular heterogeneity, perfusion, and area under the curve. Diagnostic accuracy was determined relative to pathology using receiver operating characteristic (ROC) and reverse, step-wise logistical regression analyses. The κ-statistic was calculated for inter-reader agreement. RESULTS: Data were successfully acquired in 219 cases and biopsies indicated 164 (75%) benign and 55 (25%) malignant lesions. SHI depicted more anastomoses and vascularity than HI (P < .021), but there were no differences by pathology (P > .27). Ultrasound achieved accuracies of 82 to 85%, which was significantly better than standard-of-care imaging (72%; P < .03). SHI increased diagnostic confidence by 3 to 6% (P < .05), but inter-reader agreements were medium to low (κ < 0.52). The best regression model achieved 97% accuracy by combining clinical reads and parametric SHI. CONCLUSIONS: Combining quantitative 3D SHI parameters and clinical assessments improves the characterization of suspicious breast lesions.

OBJECTIVE: To determine the early impact of the COVID-19 pandemic on breast imaging centers in California and Texas and compare regional differences. METHODS: An 11-item survey was emailed to American College of Radiology accredited breast imaging facilities in California and Texas in August 2020. A question subset addressed March-April government restrictions on elective services (during the shutdown and after reopening). Comparisons were made between states with chi-square and Fishers tests, and timeframes with McNemars and paired t-tests. RESULTS: There were 54 respondents (54/240, 23%, 26 California, 28 Texas). Imaging volumes fell during the shutdown and remained below pre-pandemic levels after reopening, with reduction in screening greatest (ultrasound 12% of baseline, mammography 13%, MRI 23%), followed by diagnostic MRI (43%), procedures (44%), and diagnostics (45%). California reported higher volumes during the shutdown (procedures, MRI) and after reopening (diagnostics, procedures, MRI) versus Texas (P = 0.001-0.02). Most screened patients (52/54, 96% symptoms and 42/54, 78% temperatures), and 100% (53/53) modified check-in and check-out. Reading rooms or physician work were altered for social distancing (31/54, 57%). Physician mask (45/48, 94%), gown (15/48, 31%), eyewear (22/48, 46%), and face shield (22/48, 46%) use during procedures increased after reopening versus pre-pandemic (P < 0.001-0.03). Physician (47/54, 87%) and staff (45/53, 85%) financial impacts were common, but none reported terminations. CONCLUSION: Breast imaging volumes during the early pandemic fell more severely in Texas than in California. Safety measures and financial impacts on physicians and staff were similar in both states.

Objective

DNA sequencing tests are enabling physicians to interrogate the molecular profiles of patients' tumors, but most oncologists have not been trained in advanced genomics. We initiated a molecular tumor board to provide expert multidisciplinary input for these patients.

Materials and methods

A team that included clinicians, basic scientists, geneticists, and bioinformatics/pathway scientists with expertise in various cancer types attended. Molecular tests were performed in a Clinical Laboratory Improvement Amendments environment.

Results

Patients (n = 34, since December 2012) had received a median of three prior therapies. The median time from physician order to receipt of molecular diagnostic test results was 27 days (range: 14-77 days). Patients had a median of 4 molecular abnormalities (range: 1-14 abnormalities) found by next-generation sequencing (182- or 236-gene panels). Seventy-four genes were involved, with 123 distinct abnormalities. Importantly, no two patients had the same aberrations, and 107 distinct abnormalities were seen only once. Among the 11 evaluable patients whose treatment had been informed by molecular diagnostics, 3 achieved partial responses (progression-free survival of 3.4 months, ≥6.5 months, and 7.6 months). The most common reasons for being unable to act on the molecular diagnostic results were that patients were ineligible for or could not travel to an appropriately targeted clinical trial and/or that insurance would not cover the cognate agents.

Conclusion

Genomic sequencing is revealing complex molecular profiles that differ by patient. Multidisciplinary molecular tumor boards may help optimize management. Barriers to personalized therapy include access to appropriately targeted drugs.

129 Background: We developed IRB-approved secure research registries for our Breast Imaging and Breast Care Clinics at Moores UC San Diego Cancer Center, as part of the UC statewide Athena Breast Health Network. Our clinical care registries securely store patient-reported intake data, which are summarized and uploaded into the medical record for patient care. Following consent, data are securely maintained in a separate research registry. In addition to facilitating quality assurance data collection, these registries aim to 1) maintain a data-rich research registry, 2) offer patients research opportunities, and 3) facilitate participant screening and recruitment into research studies.Breast Imaging or Breast Care clinic patients complete an online clinical intake form prior to their appointment, either at home or using an iPad in clinic, and are given the opportunity to be involved in research. Patients are asked for site-specific consent to keep personally identifiable intake data in a research registry, for consent to be approached about providing a biospecimen sample, and for consent to be contacted for future research opportunities. Data from consented participants are pulled into secure databases available to study personnel.Participants to date include 4,480 patients, of whom 3,246 consented to use of data for research (72%), and 2,627 have agreed to be contacted for future research opportunities (59%). In a pilot biospecimen collection protocol, 46% of patients agreed to be approached and we have collected over 360 blood or saliva and 51 tissue samples. Additionally, with the use of future contact consent, we have facilitated recruitment of more than 370 participants to multiple lifestyle and survey-based clinical studies.Patients are willing to participate in research, especially in a research registry that requires little additional time on their part. The use of research registries allows collection of a variety of data elements useful for prescreening participants for research studies, including body mass index, age, menopausal status and breast cancer diagnosis. Data- and participant-rich research registries facilitate efficient screening and recruitment for other research studies.