Quantifying and leveraging predictive uncertainty for medical image assessment

The interpretation of medical images is a challenging task, often complicated by the presence of artifacts, occlusions, limited contrast and more. Most notable is the case of chest radiography, where there is a high inter-rater variability in the detection and classification of abnormalities. This is largely due to inconclusive evidence in the data or subjective definitions of disease appearance. An additional example is the classification of anatomical views based on 2D Ultrasound images. Often, the anatomical context captured in a frame is not sufficient to recognize the underlying anatomy. Current machine learning solutions for these problems are typically limited to providing probabilistic predictions, relying on the capacity of underlying models to adapt to limited information and the high degree of label noise. In practice, however, this leads to overconfident systems with poor generalization on unseen data. To account for this, we propose a system that learns not only the probabilistic estimate for classification, but also an explicit uncertainty measure which captures the confidence of the system in the predicted output. We argue that this approach is essential to account for the inherent ambiguity characteristic of medical images from different radiologic exams including computed radiography, ultrasonography and magnetic resonance imaging. In our experiments we demonstrate that sample rejection based on the predicted uncertainty can significantly improve the ROC-AUC for various tasks, e.g., by 8% to 0.91 with an expected rejection rate of under 25% for the classification of different abnormalities in chest radiographs. In addition, we show that using uncertainty-driven bootstrapping to filter the training data, one can achieve a significant increase in robustness and accuracy. Finally, we present a multi-reader study showing that the predictive uncertainty is indicative of reader errors.     

Confounded association between proton pump inhibitor use and risk of biliary tract cancer: Result from three cohorts

Recent epidemiological studies suggested that proton pump inhibitor (PPI) use was associated with an increased risk of biliary tract cancer (BTC), however, confounders were not adequately controlled. Our study aimed to evaluate PPI use and subsequent risk of BTC and its subtypes in three well-established cohorts. We conducted a pooled analysis of the subjects free of cancers in UK Biobank (n = 463 643), Nurses' Health Study (NHS, n = 80 235) and NHS II (n = 95 869). Propensity score weighted Cox models were used to estimate marginal HRs of PPIs use on BTC risk, accounting for potential confounders. We documented 284 BTC cases in UK Biobank (median follow-up: 7.6 years), and 91 cases in NHS and NHS II cohorts (median follow-up: 15.8 years). In UK biobank, PPI users had a 96% higher risk of BTC compared to nonusers in crude model (HR 1.96, 95% CI 1.44-2.66), but the effect was attenuated to null after adjusting for potential confounders (HR 0.95, 95% CI 0.60-1.49). PPI use was not associated with risk of BTC in the pooled analysis of three cohorts (HR 0.93, 95% CI 0.60-1.43). We also observed no associations between PPI use with risk of intrahepatic (HR 1.00, 95% CI 0.49-2.04), extrahepatic bile duct (HR 1.09, 95% CI 0.52-2.27) and gallbladder cancers (HR 0.66, 95% CI 0.26-1.66) in UK Biobank. In summary, regular use of PPIs was not associated with the risk of BTC and its subtypes.     

Medical practitioners’ educational competence about oral and oropharyngeal carcinoma: a systematic review and meta-analysis

Medical practitioners’ (MP) role is pivotal in primary prevention, early diagnosis, prompt referral and effective management of oral and oropharyngeal carcinomas (OC/OPC), which raises the importance of their effective OC/OPC education at all levels of medical education. The purpose of this systematic review was to summarise the available scientific evidence about their educational competence in dealing with OC/OPC. We made a systematic search of papers in the English language in MEDLINE, Scopus, Cochrane Library CENTRAL and CINAHL databases from their inception until December 2018. Overall, 23 cross-sectional and three interventional studies have been selected for the systematic review and 18 of these were included in the meta-analyses. Excluding tobacco use (synthesised estimate of 95% of respondents identified tobacco as an OC/OPC risk factor, 95% CI of synthesised estimate 92% to 97%) and alcohol consumption (65%, 95%CI 52% to 77%), less than half of MP (approximately) were knowledgeable about important OC/OPC risk factors including human papilloma virus (42%, 95% CI 30% to 54%), poor diet (34%, 95% CI 17% to 54%), and advancing age (45%, 95% CI 21% to 70%). There was a low to moderate level of awareness among MP regarding common precancerous oral lesions involving leukoplakia (56%, 95% CI 32% to 79%), erythroplakia (30%, 95% CI 8% to 58%), and oral lichen planus (13%, 95% CI 0 to 41%). Moderate knowledge was also recorded about frequent sites of OC development involving the tongue (48%, 95% CI 33% to 64%) and floor of the mouth (37%, 95% CI 19% to 57%). Most MP enquired about tobacco use (86%, 95% CI 74% to 96%), and alcohol consumption (73%, 95% CI 47% to 94%) during history taking, and expressed willingness to be given supplementary OC/OPC education (78%, 95% CI 54% to 96%), as well. With regard to the incidence of intraoral screening, 27% of MP (95% CI 12% to 46%) make an intraoral examination as a routine. Interestingly, studies from each continent yielded significantly different outcomes to some research questions in the review. From the MP’s perspective, clinical time restrictions and deficiencies in organised training were recognised as the main barriers towards their OC/OPC educational competence. The findings of this systematic review indicated the existence of deficiencies in knowledge and misconceptions, neglected preventive responsibilities, and associated barriers towards OC/OPC. A need for improved OC/OPC training at all levels of medical education is required to increase competence worldwide.     

Skipjack tuna (Katsuwonus pelamis) eyeball oil exerts an anti-inflammatory effect by inhibiting NF-κB and MAPK activation in LPS-induced RAW 264.7 cells and croton oil-treated mice

The effect of tuna eyeball oil (TEO) on lipopolysaccharide (LPS)-induced inflammation in macrophage cells was investigated. TEO had no cytotoxicity in cell viability as compared to the control in LPS induced RAW 264.7 cells. TEO reduced the levels of NO and pro-inflammatory cytokines by up to 50% in a dose-dependent manner. The expression of NF-κB and MAPKs as well as iNOS and COX-2 proteins was reduced by TEO, which suggests that its anti-inflammatory activity is related to the suppression of the NF-κB and MAPK signaling pathways. The rate of formation of ear edema was reduced compared to that in the control at the highest dose tested. In an acute toxicity test, no mice were killed by TEO doses of up to 5000 mg/kg body weight during the two week observation period. These results suggested that TEO may have a significant effect on inflammatory factors and be a potential anti-inflammatory therapeutic.     

Association of weight-adjusted-waist index with incident hypertension: The Rural Chinese Cohort Study

AimsTo explore the association between WWI and the incidence of HTN in the Rural Chinese Cohort Study.Methods and ResultsWe examined data for 10,338 non-hypertensive participants (39.49% men) aged ≥ 18 years from the Rural Chinese Cohort Study who completed a baseline examination during 2007–2008 and follow-up during 2013–2014. WWI was calculated as waist circumference (cm) divided by the square root of weight (kg). Multiple logistic regression models were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for the probability of HTN across four WWI categories. Restricted cubic splines analysis was used to model the dose–response association of WWI and HTN. A total of 2078 participants had HTN during a median follow-up of 6 years. After adjusting for potential confounders, as compared with the lowest WWI category (<9.94 cm/√kg), with WWI 9.94 to 10.42, 10.42 to 10.91 and ≥ 10.91 cm/√kg, the ORs (95% CIs) for HTN were 1.12 (0.93–1.35), 1.40 (1.17–1.69) and 1.50 (1.24–1.82), respectively. Results of the sensitivity analyses were robust. The ORs were generally consistent on subgroup analysis by sex, smoking status, systolic blood pressure and diastolic blood pressure. Multiple logistic regression models with restricted cubic splines showed a non-linear positive association between WWI and HTN (P_nonlinearity < 0.001).ConclusionThe highest WWI category was significantly associated with increased risk of HTN. Our findings may facilitate the development and promotion of obesity prevention strategies aimed at reducing the risk of HTN and provide evidence for healthcare policy in rural China.