In this paper we investigate problematic practices and consequences of large scale vision datasets (LSVDs). We examine broad issues such as the question of consent and justice as well as specific ...concerns such as the inclusion of verifiably pornographic images in datasets. Taking the ImageNet-ILSVRC-2012 dataset as an example, we perform a cross-sectional model-based quantitative census covering factors such as age, gender, NSFW content scoring, class- wise accuracy, human-cardinality-analysis, and the semanticity of the image class information in order to statistically investigate the extent and subtleties of ethical transgressions. We then use the census to help hand-curate a look-up-table of images in the ImageNet-ILSVRC-2012 dataset that fall into the categories of verifiably pornographic: shot in a non-consensual setting (up-skirt), beach voyeuristic, and exposed private parts. We survey the landscape of harm and threats both the society at large and individuals face due to uncritical and ill-considered dataset curation practices. We then propose possible courses of correction and critique their pros and cons. We have duly open-sourced all of the code and the census meta-datasets generated in this endeavor for the computer vision community to build on. By unveiling the severity of the threats, our hope is to motivate the constitution of mandatory Institutional Review Boards (IRB) for large scale dataset curation.
Purpose To compare standard diffusion-weighted (DW) imaging and diffusion kurtosis (DK) imaging for prostate cancer (PC) detection and characterization in a large patient cohort, with attention to ...the potential added value of DK imaging. Materials and Methods This retrospective institutional review board-approved study received a waiver of informed consent. Two hundred eighty-five patients with PC underwent 3.0-T phased-array coil prostate magnetic resonance (MR) imaging, including a DK imaging sequence (b values 0, 500, 1000, 1500, and 2000 sec/mm
) before prostatectomy. Maps of apparent diffusion coefficient (ADC) and diffusional kurtosis (K) were derived by using maximal b values of 1000 and 2000 sec/mm
, respectively. Mean ADC and K were obtained from volumes of interest (VOIs) placed on each patient's dominant tumor and benign prostate tissue. Metrics were compared between benign and malignant tissue, between Gleason score (GS) ≤ 3 + 3 and GS ≥ 3 + 4 tumors, and between GS ≤ 3 + 4 and GS ≥ 4 + 3 tumors by using paired t tests, analysis of variance, receiver operating characteristic (ROC) analysis, and exact tests. Results ADC and K showed significant differences for benign versus tumor tissues, GS ≤ 3 + 3 versus GS ≥ 3 + 4 tumors, and GS ≤ 3 + 4 versus GS ≥ 4 + 3 tumors (P < .001 for all). ADC and K were highly correlated (r = -0.82; P < .001). Area under the ROC curve was significantly higher (P = .002) for ADC (0.921) than for K (0.902) for benign versus malignant tissue but was similar for GS ≤ 3 + 3 versus GS ≥ 3 + 4 tumors (0.715-0.744) and GS ≤ 3 + 4 versus GS ≥ 4 + 3 tumors (0.694-0.720) (P > .15). ADC and K were concordant for these various outcomes in 80.0%-88.6% of patients; among patients with discordant results, ADC showed better performance than K for GS ≤ 3 + 4 versus GS ≥ 4 + 3 tumors (P = .016) and was similar to K for other outcomes (P > .136). Conclusion ADC and K were highly correlated, had similar diagnostic performance, and were concordant for the various outcomes in the large majority of cases. These observations did not show a clear added value of DK imaging compared with standard DW imaging for clinical PC evaluation.
RSNA, 2017 Online supplemental material is available for this article.
Objectives
To perform a systematic review on apparent diffusion coefficient (ADC) values of renal tumor subtypes and meta-analysis on the diagnostic performance of ADC for differentiation of ...localized clear cell renal cell carcinoma (ccRCC) from other renal tumor types.
Methods
Medline, Embase, and the Cochrane Library databases were searched for studies published until May 1, 2019, that reported ADC values of renal tumors. Methodological quality was evaluated. For the meta-analysis on diagnostic test accuracy of ADC for differentiation of ccRCC from other renal lesions, we applied a bivariate random-effects model and compared two subgroups of ADC measurement with vs. without cystic and necrotic areas.
Results
We included 48 studies (2588 lesions) in the systematic review and 13 studies (1126 lesions) in the meta-analysis. There was no significant difference in ADC of renal parenchyma using
b
values of 0–800 vs. 0–1000 (
p
= 0.08). ADC measured on selected portions (sADC) excluding cystic and necrotic areas differed significantly from whole-lesion ADC (wADC) (
p
= 0.002). Compared to ccRCC, minimal-fat angiomyolipoma, papillary RCC, and chromophobe RCC showed significantly lower sADC while oncocytoma exhibited higher sADC. Summary estimates of sensitivity and specificity to differentiate ccRCC from other tumors were 80% (95% CI, 0.76–0.88) and 78% (95% CI, 0.64–0.89), respectively, for sADC and 77% (95% CI, 0.59–0.90) and 77% (95% CI, 0.69–0.86) for wADC. sADC offered a higher area under the receiver operating characteristic curve than wADC (0.852 vs. 0.785,
p
= 0.02).
Conclusions
ADC values of kidney tumors that exclude cystic or necrotic areas more accurately differentiate ccRCC from other renal tumor types than whole-lesion ADC values.
Key Points
• Selective ADC of renal tumors, excluding cystic and necrotic areas, provides better discriminatory ability than whole-lesion ADC to differentiate clear cell RCC from other renal lesions, with area under the receiver operating characteristic curve (AUC) of 0.852 vs. 0.785, respectively (p = 0.02).
• Selective ADC of renal masses provides moderate sensitivity and specificity of 80% and 78%, respectively, for differentiation of clear cell renal cell carcinoma (RCC) from papillary RCC, chromophobe RCC, oncocytoma, and minimal-fat angiomyolipoma.
• Selective ADC excluding cystic and necrotic areas are preferable to whole-lesion ADC as an additional tool to multiphasic MRI to differentiate clear cell RCC from other renal lesions whether the highest b value is 800 or 1000.
The purpose of this study is to evaluate the roles of self-directed learning and continual feedback in the learning curve for tumor detection by novice readers of prostate MRI.
A total of 124 ...prostate MRI examinations classified as positive (n = 52; single Prostate Imaging Reporting and Data System PI-RADS category 3 or higher lesion showing Gleason score ≥ 7 tumor at MRI-targeted biopsy) or negative (n = 72; PI-RADS category 2 or lower and negative biopsy) for detectable tumor were included. These were divided into four equal-sized batches, each with matching numbers of positive and negative examinations. Six second-year radiology residents reviewed examinations to localize tumors. Three of the six readers received feedback after each examination showing the preceding case's solution. The learning curve, plotting accuracy over time, was assessed by the Akaike information criterion (AIC). Logistic regression and mixed-model ANOVA were performed.
For readers with and without feedback, the learning curve exhibited an initial rapid improvement that slowed after 40 examinations (change in AIC > 0.2%). Accuracy improved from 58.1% (batch 1) to 71.0-75.3% (batches 2-4) without feedback and from 58.1% to 72.0-77.4% with feedback (p = 0.027-0.046), without a difference in the extent of improvement (p = 0.800). Specificity improved from 53.7% to 68.5-81.5% without feedback and from 55.6% to 74.1-81.5% with feedback (p = 0.006-0.010), without a difference in the extent of improvement (p = 0.891). Sensitivity improved from 59.0-61.5% (batches 1-2) to 71.8-76.9% (batches 3-4) with feedback (p = 0.052), though did not improve without feedback (p = 0.602). Sensitivity for transition zone tumors exhibited larger changes (p = 0.024) with feedback than without feedback. Sensitivity for peripheral zone tumors did not improve in either group (p > 0.3). Reader confidence increased only with feedback (p < 0.001).
The learning curve in prostate tumor detection largely reflected self-directed learning. Continual feedback had a lesser effect. Clinical prostate MRI interpretation by novice radiologists warrants caution.
The internet has become an integral component of daily life, with its content broadly grouped into social media and non-social online content. The use of social media, comprising interactive ...information sharing and networking tools, has proliferated in radiology, with as many as 85% of radiologists utilizing social media and adoption by both private and academic practices. In radiology, social media has been used to increase patient and public awareness of the specialty, establish a professional brand, share scholarly activity, aid in professional development and recruitment, improve communication, and increase engagement during scientific meetings. Organizations like the Society of Abdominal Radiology have increased the visibility of abdominal and pelvic radiology through their activity on social media; however, individual radiologists are integral to this function as well. In this article, we outline a stepwise approach to the individual use of social media, with practical tips for radiologists interested in effectively engaging the plethora of social media and non-social online content available. We also briefly discuss an approach to performing research using publicly available online media.
Coronavirus disease of 2019 (COVID-19) is associated with increased risk of stroke and intracranial hemorrhage. This first report of fulminant panvascular arteriovenous thrombosis with subarachnoid ...hemorrhage (SAH) in a post-COVID-19 infection is attributed to extensive arteriovenous inflammation leading to arterial rupture following vasculitis.
We report a rare case of extensive extra- and intra-cranial cerebral arteriovenous thrombosis following COVID-19 infection, presenting as fatal non-aneurysmal subarachnoid hemorrhage. The clinical course, biochemical and radiological evaluation is discussed. The other possible etiological differentials which were analysed and ruled out during case management are also detailed.
A high degree of suspicion for COVID-19 induced coagulopathy leading to extensive non- aneurysmal, non-hemispheric SAH and malignant intracranial hypertension should be entertained. Our experience and previous reports on non-aneurysmal SAH in such patients show a poor prognosis.
Abstract Background Urinary incontinence is a common short-term complication of radical prostatectomy (RP). Little is known about the long-term impact of RP on continence. Objective To elucidate the ...long-term progression of continence after RP. Design, setting, and participants From October 2000 through September 2012, 1788 men undergoing open RP for clinically localized prostate cancer by a single surgeon at an urban tertiary care center prospectively signed consent to be followed before RP and at 3, 6, 12, 24, 96, and 120 mo after RP. A consecutive sampling method was used and all men were included in this study. Intervention Men underwent open RP. Outcome measurements and statistical analysis Regression models controlled for preoperative University of California, Los Angeles–Prostate Cancer Index urinary function score (UCLA-PCI-UFS), age, prostate-specific antigen level, Gleason score, stage, nerve-sparing status, race, and marital status were used to evaluate the association of time since RP with two dependent variables: UCLA-PCI-UFS and continence status. Results and limitation The mean UCLA-PCI-UFS declined between 2 yr and 8 yr (83.8 vs 81.8; p = 0.007) and marginally between 8 yr and 10 yr (81.8 vs 79.6; p = 0.036) after RP, whereas continence rate did not significantly change during these intervals. Men ≥60 yr old experienced a decline in mean UCLA-PCI-UFS between 2 yr and 8 yr ( p = 0.002) and a marginal decline in continence rate between 2 yr and 10 yr ( p = 0.047), whereas these variables did not change significantly in men <60 yr old. These outcomes are for an experienced surgeon, so caution should be exercised in generalizing these results. Conclusions Between 2 yr and 10 yr after RP, there were slight decreases in mean UCLA-PCI-UFS and continence rates in this study. Men aged <60 yr had better long-term outcomes. These results provide realistic long-term continence expectations for men undergoing RP.
The aim of this study was to evaluate use of the microblogging social network Twitter by academic radiology departments (ARDs) in the United States.
Twitter was searched to identify all accounts ...corresponding with United States ARDs. All original tweets from identified accounts over a recent 3-month period (August to October 2014) were archived. Measures of account activity, as well as tweet and link content, were summarized.
Fifteen ARDs (8.2%) had Twitter accounts. Ten (5.5%) had "active" accounts, with ≥1 tweet over the 3-month period. Active accounts averaged 711 ± 925 followers (maximum, 2,885) and 61 ± 93 tweets (maximum, 260) during the period. Among 612 tweets from active accounts, content most commonly related to radiology-related education (138), dissemination of departmental research (102), general departmental or hospital promotional material (62), departmental awards or accomplishments (60), upcoming departmental lectures (59), other hospital-related news (55), medical advice or information for patients (38), local community events or news (29), social media and medicine (27), and new departmental or hospital hires or expansion (19). Eighty percent of tweets (490 of 612) included 315 unique external links. Most frequent categories of link sources were picture-, video-, and music-sharing websites (89); the ARD's website or blog (83); peer-reviewed journal articles (40); the hospital's or university's website (34), the lay press (28), and Facebook (14).
Twitter provides ARDs the opportunity to engage their own staff members, the radiology community, the department's hospital, and patients, through a broad array of content. ARDs frequently used Twitter for promotional and educational purposes. Because only a small fraction of ARDs actively use Twitter, more departments are encouraged to take advantage of this emerging communication tool.
The purpose of this study was to evaluate themes related to patients' experience in undergoing mammography, as expressed on Twitter.
A total of 464 tweets from July to December 2015 containing the ...hashtag #mammogram and relating to a patient's experience in undergoing mammography were reviewed.
Of the tweets, 45.5% occurred before the mammogram compared to 49.6% that occurred afterward (remainder of tweets indeterminate). However, in patients undergoing their first mammogram, 32.8% occurred before the examination, whereas in those undergoing follow-up mammogram, 53.0% occurred before the examination. Identified themes included breast compression (24.4%), advising other patients to undergo screening (23.9%), recognition of the health importance of the examination (18.8%), the act of waiting (10.1%), relief regarding results (9.7%), reflection that the examination was not that bad (9.1%), generalized apprehension regarding the examination (8.2%), interactions with staff (8.0%), the gown (5.0%), examination costs or access (3.4%), offering or reaching out for online support from other patients (3.2%), perception of screening as a sign of aging (2.4%), and the waiting room or waiting room amenities (1.3%). Of the tweets, 31.9% contained humor, of which 56.1% related to compression. Themes that were more common in patients undergoing their first, rather than follow-up, mammogram included breast compression (16.4% vs 9.1%, respectively) and that the test was not that bad (26.2% vs 7.6%, respectively).
Online social media provides a platform for women to share their experiences and reactions in undergoing mammography, including humor, positive reflections, and encouragement of others to undergo the examination. Social media thus warrants further evaluation as a potential tool to help foster greater adherence to screening guidelines.
Radiology social media research has focused on Twitter, Facebook, and YouTube, with minimal attention to Instagram, which may be a natural platform for engagement. We evaluated features of public ...radiology-related Instagram accounts.
Instagram accounts were searched using the term "radiology." Two independent raters used a coding system developed through initial content review to categorize account user types and post aims over the study month (October 2019). Multivariate linear regressions were performed.
49 Instagram accounts and their 459 posts were retrieved. Users had median 117 total posts (interquartile range 43-203), 9 posts during the study month (1-16), 3079 followers (695-19,600), and 106 followed users (50-351). Most users were radiologists (41%, most commonly academic), radiology practices (12%, most commonly academic) and 10% radiology professional societies. The 459 posts had median 178 likes (interquartile range 51-353) and 2 comments (0-5); single videos had median 4693 views (2183-6080). 75% of posts were radiology images or videos (of these, 29% MRI, 28% CT). 65% of posts were clinical imaging cases (most commonly neuroradiology). Seven posts violated HIPAA regulations, all originating outside the U.S. When controlling for number of account followers, posts by radiology society (β = -130.6, p = 0.017) and practice accounts (β = -90.6, p = 0.049) had fewer post likes. An artwork post was the only significant predictor of greater post likes (β = +866.2, p < 0.001) and comments (β = +14.3, p < 0.001).
Instagram is commonly used by academic radiologists to present clinical imaging cases. Radiologists should consider opportunities for greater application of this unique image-based social media platform.