Correlation of benign incidental findings seen on whole-body PET-CT with knee MRI: patterns of 18F-FDG avidity, intra-articular pathology, and bone marrow edema lesions.

OBJECTIVES: To correlate patterns of 18F-FDG uptake on whole-body PET-CT with MR findings and compare the degree of FDG activity between symptomatic and asymptomatic knees. MATERIALS AND METHODS: Retrospective database query was performed using codes for knee MRI as well as whole-body PET-CT. Patients with malignant disease involving the knee or hardware were excluded. Patients who had both studies performed within 1 year between 2012 and 2017 were included for analysis. Knee joint osteoarthrosis, meniscal and ligamentous integrity, presence of joint effusion, and synovitis were assessed and recorded. Bone marrow edema lesions (BMELs) were identified, segmented, and analyzed using volumetric analysis. SUVmax was assessed over the suprapatellar joint space, intercondylar notch and Hoffa's fat pad. Symptomatic and asymptomatic knees were compared in patients with unilateral symptoms. RESULTS: Twenty-two cases (20 patients) with mean age 63.3 years (range, 36-91 years) were included. Two patients had bilateral pain. The most FDG avid regions in both symptomatic and asymptomatic knees were the intercondylar notch (SUVmax = 1.84 vs. 1.51), followed by suprapatellar pouch (SUVmax = 1.74 vs. 1.29) and Hoffa's fat pad (SUVmax = 1.01 vs. 0.87). SUVmax was significantly associated with cartilage loss (mean modified Outerbridge score) (r = 0.60, p = 0.003) and degree of synovitis (r = 0.48, p = 0023). Overall, mean SUVmax was significantly higher in the presence of a meniscal tear (1.83 ± 0.67 vs. 1.22 ± 0.40, p = 0.030). Nine patients had BMELs (volume: range = 0.6-27.8, mean = 7.79) however there was no significant association between BMEL volume and SUVmax. CONCLUSIONS: Higher FDG activity correlates with intra-articular derangement and the intercondylar notch represents the most metabolically active region of the knee.

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).

Male Breast Cancer Risk Assessment and Screening Recommendations in High-Risk Men Who Undergo Genetic Counseling and Multigene Panel Testing.

BACKGROUND: Emerging data suggest screening mammography may be effective in detecting breast cancer early in high-risk men. We evaluated current screening recommendations as a risk management strategy in men at elevated risk for breast cancer. PATIENTS AND METHODS: This institutional review board-approved, Health Insurance Portability and Accountability Act (HIPAA)-compliant study reviewed consecutive men who underwent genetic counseling and multigene panel testing (MGPT) for breast cancer risk assessment at our institution between 2012 and 2018. Patient risk factors, test indications, and MGPT outcomes were recorded, then correlated with screening recommendations by either clinical breast examination or screening mammography. Recommendation consistency among practitioners was evaluated. Patient adherence to screening mammography (defined as undergoing screening mammography as recommended) was assessed. Statistical analysis was performed at the 2-sided 5% significance level. RESULTS: A total of 414 asymptomatic men underwent both genetic counseling and MGPT (mean age, 47 years; range, 18-91 years) for breast cancer risk assessment. Of this group, 18 (4.3%) of 414 had a personal history of breast cancer, and 159 (38.4%) of 414 had a family history of breast cancer before MGPT. Among 112 men with positive MGPT results, BRCA1/2 mutations were the most common (56.3%, 63/112). Most BRCA mutation carriers (80.9%, 51/63) were recommended clinical breast examination only. Only 5.9% (2/34) BRCA2 and 10.3% (3/29) BRCA1 carriers were recommended screening mammograms (7.9%, 5/63 of all BRCA carriers). Among men with a personal history of breast cancer, only 9 (50%) of 18 were recommended screening mammograms. Overall adherence to screening mammogram in men was 71.4% (10/14), which ultimately yielded two cancers. Breast cancer screening recommendations varied widely among practitioners, with some recommending clinical breast examination only, and others also recommending mammography. CONCLUSION: Men who are found to be at an elevated risk for breast cancer after undergoing genetic counseling and testing currently receive relatively inconsistent screening recommendations.

Deep Neural Networks Improve Radiologists' Performance in Breast Cancer Screening.

We present a deep convolutional neural network for breast cancer screening exam classification, trained, and evaluated on over 200000 exams (over 1000000 images). Our network achieves an AUC of 0.895 in predicting the presence of cancer in the breast, when tested on the screening population. We attribute the high accuracy to a few technical advances. 1) Our network's novel two-stage architecture and training procedure, which allows us to use a high-capacity patch-level network to learn from pixel-level labels alongside a network learning from macroscopic breast-level labels. 2) A custom ResNet-based network used as a building block of our model, whose balance of depth and width is optimized for high-resolution medical images. 3) Pretraining the network on screening BI-RADS classification, a related task with more noisy labels. 4) Combining multiple input views in an optimal way among a number of possible choices. To validate our model, we conducted a reader study with 14 readers, each reading 720 screening mammogram exams, and show that our model is as accurate as experienced radiologists when presented with the same data. We also show that a hybrid model, averaging the probability of malignancy predicted by a radiologist with a prediction of our neural network, is more accurate than either of the two separately. To further understand our results, we conduct a thorough analysis of our network's performance on different subpopulations of the screening population, the model's design, training procedure, errors, and properties of its internal representations. Our best models are publicly available at https://github.com/nyukat/breast_cancer_classifier.

Characteristics of the Most Recently Awarded Magnetic Resonance Imaging Patents in the United States.

PURPOSE: To characterize recent magnetic resonance imaging (MRI) technical development and innovation based on data regarding MRI-related patents awarded in 2016. METHODS: The US Patent and Trademark Office website was searched for patents awarded in 2016 and an abstract containing "magnetic resonance." Patent characteristics were summarized. An MRI physicist classified patents' themes. RESULTS: A total of 423 MRI-related patents were awarded in 2016. Among these, 29% had 1 inventor, 24% had 2 inventors, and 47% had ≥2 inventors. Mean interval between patents being filed and awarded was 1389 ± 559days (range: 167-4029). Most common countries of patents' first assignee were USA (40%), Germany (24%), Netherlands (10%), and Japan (10%). In all, 3% included assignees with different countries (most common collaborators USA and Germany). Patents' first assignee had an industry affiliation in 76% vs an academic affiliation in 21% (4% indeterminate); and 3% had industry-academia collaboration. Patents' most common themes were coils (n = 77), sequence design (n = 65), and noncoil scanner hardware (n = 41). These top themes were similar for USA, international, and industry-based patents; however, for academic-based patents, the most common themes were sequence design, reconstruction, and exogenous agents. Less common themes included image analysis, postprocessing, spectroscopy, relaxometry, diffusion, motion correction, radiation therapy, implants, wireless devices, and positron emission tomography-MRI. CONCLUSION: Most MRI-related patents were by non-US inventors. A large majority had industry affiliation; minimal industry-academic collaborationwas observed. Patents from industry and academic inventors had distinct top focuses: hardware and software, respectively. Awarenessofthe most recent years' MRI patents may provide insights into forthcoming clinical translations and help guide ongoing research and entrepreneurism.