Breast mass segmentation in ultrasound with selective kernel U-Net convolutional neural network.

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.

ACR Ultrasound Liver Reporting and Data System: Multicenter Assessment of Clinical Performance at One Year.

PURPOSE: The aim of this study was to evaluate the clinical performance of the ACR's Ultrasound Liver Reporting and Data System (US LI-RADS™) for detecting hepatocellular carcinoma (HCC) in patients at high risk for HCC. METHODS: In this retrospective, multicenter study, 2,050 patients at high risk for HCC (1,078 men and 972 women; mean age, 57.7 years) at five sites in the United States had undergone screening liver ultrasound from January 2017 to February 2018, and US LI-RADS observation categories and visualization scores were assigned on a clinical basis. Ultrasound reports and patient records were retrospectively reviewed and follow-up imaging studies and/or pathologic reports recorded. Descriptive statistics were generated, and multivariate logistic regression analysis was used to analyze the relationship of clinical and reader-based predictors of limited visualization. Diagnostic performance data were calculated in the subset of patients with confirmatory testing. RESULTS: The most common indications for HCC screening were cirrhosis (n = 1,054 [51.4%]), noncirrhotic hepatitis B virus infection (n = 555 [27.1%]), and noncirrhotic hepatitis C virus infection (n = 234 [11.4%]). US LI-RADS observation categories assigned were US-1 (negative) in 90.4% (n = 1,854), US-2 (subthreshold) in 4.6% (n = 95), and US-3 (positive) in 4.9% (n = 101). Visualization scores were A (no or minimal limitations) in 76.8% (n = 1,575), B (moderate limitations) in 18.9% (n = 388), and C (severe limitations) in 4.2% (n = 87). Confirmatory tests, including multiphase contrast-enhanced CT or MRI (n = 331) or histopathology (n = 18), were available for 349 patients (17.0%). The sensitivity of US LI-RADS in this subset of patients was 82.4%, specificity was 74.2%, positive predictive value was 35.3%, and negative predictive value was 96.1%. CONCLUSIONS: Approximately 90% of US LI-RADS screening examinations were negative, 5% subthreshold, and 5% positive. Visualization scores were diagnostically acceptable in the vast majority (>95%) of examinations. US LI-RADS emphasized sensitivity and negative predictive value, which are key characteristics of a screening test.

Role of US LI-RADS in the LI-RADS Algorithm.

The US Liver Imaging Reporting and Data System (LI-RADS) was released in 2017 and is the newest of the four American College of Radiology (ACR) LI-RADS algorithms. US LI-RADS provides standardized terminology, technical recommendations, and a reporting framework for US examinations performed for screening or surveillance in patients at risk for developing hepatocellular carcinoma (HCC). The appropriate patient population for screening and surveillance includes individuals who are at risk for developing HCC but do not have known or suspected cancer. This includes patients with cirrhosis from any cause and subsets of patients with chronic hepatitis B virus infection in the absence of cirrhosis. In an HCC screening or surveillance study, US LI-RADS recommends assigning two scores that apply to the entire study: the US category, which determines follow-up, and a visualization score, which communicates the expected level of sensitivity of the examination but does not affect management. Three US categories are possible: US-1 negative, a study with no evidence of HCC; US-2 subthreshold, a study in which an observation less than 10 mm is depicted that is not definitely benign; and US-3 positive, a study in which an observation greater than or equal to 10 mm or a new thrombus in vein is identified, for which diagnostic contrast material-enhanced imaging is recommended. Three visualization scores are possible: A (no or minimal limitations), B (moderate limitations), and C (severe limitations). ©RSNA, 2019.

Reader agreement and accuracy of ultrasound features for hepatic steatosis.

PURPOSE: The purpose of the study is to assess the reader agreement and accuracy of eight ultrasound imaging features for classifying hepatic steatosis in adults with known or suspected hepatic steatosis. METHODS: This was an IRB-approved, HIPAA-compliant prospective study of adult patients with known or suspected hepatic steatosis. All patients signed written informed consent. Ultrasound images (Siemens S3000, 6C1HD, and 4C1 transducers) were acquired by experienced sonographers following a standard protocol. Eight readers independently graded eight features and their overall impression of hepatic steatosis on ordinal scales using an electronic case report form. Duplicated images from the 6C1HD transducer were read twice to assess intra-reader agreement. Intra-reader, inter-transducer, and inter-reader agreement were assessed using intraclass correlation coefficients (ICC). Features with the highest intra-reader agreement were selected as predictors for dichotomized histological steatosis using Classification and Regression Tree (CART) analysis, and the accuracy of the decision rule was compared to the accuracy of the radiologists' overall impression. RESULTS: 45 patients (18 males, 27 females; mean age 56 ± 12 years) scanned from September 2015 to July 2016 were included. Mean intra-reader ICCs ranged from 0.430 to 0.777, inter-transducer ICCs ranged from 0.228 to 0.640, and inter-reader ICCs ranged from 0.014 to 0.561. The CART decision rule selected only large hepatic vein blurring and achieved similar accuracy to the overall impression (74% to 75% and 68% to 72%, respectively). CONCLUSIONS: Large hepatic vein blurring, liver-kidney contrast, and overall impression provided the highest reader agreement. Large hepatic vein blurring may provide the highest classification accuracy for dichotomized grading of hepatic steatosis.

Breast mass classification in sonography with transfer learning using a deep convolutional neural network and color conversion.

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.

Spectral Doppler signature waveforms in ultrasonography: a review of normal and abnormal waveforms.

Doppler ultrasound is routinely used in the clinical setting to evaluate blood flow in many major vessels of the body. Spectral Doppler is used to display the normal and abnormal signature waveforms that are unique to each vessel. It is important for the sonographer and the radiologist to recognize both what is normal and what is abnormal in a spectral Doppler display. In this review, we briefly explain the physics behind Doppler ultrasound and some of the most common mathematical equations applied in a routine clinical examination. We also describe and demonstrate normal versus abnormal spectral Doppler signature waveforms of vessels in the neck, abdomen, pelvis, and fetus.

An adolescent weight-loss program integrating family variables reduces energy intake.

Family variables such as cohesion and nurturance have been associated with adolescent weight-related health behaviors. Integrating family variables that improve family functioning into traditional weight-loss programs can provide health-related benefits. The current study evaluated a family-based psychoeducational and behavioral skill-building weight-loss program for adolescent girls that integrated Family Systems and Social Cognitive Theories. Forty-two overweight (> or = 95th percentile) female adolescent participants and parents participated in a 16-week randomized controlled trial comparing three groups: multifamily therapy plus psychoeducation (n=15), psychoeducation-only (n=16), or wait list (control; n=11) group. Body mass index, energy intake, and family measures were assessed at baseline and posttreatment. Adolescents in the psychoeducation-only group demonstrated a greater decrease in energy intake compared to the multifamily therapy plus psychoeducation and control groups (P<0.01). Positive changes in family nurturance were associated with lower levels of adolescent energy intake (P<0.05). No significant effects were found for body mass index. Results provide preliminary support for a psychoeducational program that integrates family variables to reduce energy intake in overweight adolescent girls. Results indicate that nurturance can be an important family variable to target in future adolescent weight-loss and dietary programs.

Prenatal diagnosis for detection of aneuploidy: the options.

The value of all noninvasive prenatal tests must be viewed with the perspective of the consequences of invasive testing. Regarding second trimester noninvasive testing, biochemical screening is more accurate in establishing risk than maternal age alone. One or more major ultrasound abnormalities, nuchal thickening, or a shortened humerus should raise concern for Down syndrome regardless of the patient's a priori risk based on age or biochemical markers. Isolated minor ultrasound markers should not be used in calculating risk in low-risk patients regarding Down syndrome unless the biochemical profile already places the patient at risk or in a borderline risk zone. If the ultrasound finding is hyperechoic bowel, problems other than aneuploidy may be the cause, including cystic fibrosis, infection, or hemorrhage, and these problems must be considered if hyperechoic bowel is an isolated finding. Improved risk adjustment seems to be applicable to a priori high-risk patients with completely normal sonograms. Genetic sonograms with specific risk adjustment schemata may be used to adjust a priori risk (either maternal age or biochemical screening results) at centers in which this has proven to be accurate, but whether this is statistically sound remains to be determined. The goal of second trimester ultrasound screening is to identify at-risk fetuses better and offer invasive testing to a more select group of patients. As the value of first trimester screening becomes more evident and practical, and if the risk of chorionic villus sampling becomes an acceptable norm, the patient population that reaches the second trimester of pregnancy will be select. Therefore, we can anticipate that second trimester screening and invasive testing may be needed only in a minority of cases, and the practice standards of prenatal testing and sonography (including minor ultrasound markers) will change entirely.