In this paper, we propose a novel method, called "Dynamic Cascade", for training an efficient face detector on massive data sets. There are three key contributions. The first is a new cascade algorithm called "Dynamic Cascade", which can train cascade classifiers on massive data sets and only requires a small number of training parameters. The second is the introduction of a new kind of weak classifier, called "Bayesian Stump", for training boost classifiers. It produces more stable boost classifiers with fewer features. Moreover, we propose a strategy for using our dynamic cascade algorithm with multiple sets of features to further improve the detection performance without significant increase in the detector's computational cost. Experimental results show that all the new techniques effectively improve the detection performance. Finally, we provide the first large standard data set for face detection, so that future researches on the topic can be compared on the same training and testing set.

BackgroundDyslipidaemia is an intermediary exacerbation factor for various diseases but the impact of obstructive sleep apnoea (OSA) on dyslipidaemia remains unclear.MethodsA total of 3582 subjects with suspected OSA consecutively admitted to our hospital sleep centre were screened and 2983 (2422 with OSA) were included in the Shanghai Sleep Health Study. OSA severity was quantified using the apnoea–hypopnea index (AHI), the oxygen desaturation index and the arousal index. Biochemical indicators and anthropometric data were also collected. The relationship between OSA severity and the risk of dyslipidaemia was evaluated via ordinal logistic regression, restricted cubic spline (RCS) analysis and multivariate linear regressions.ResultsThe RCS mapped a nonlinear dose–effect relationship between the risk of dyslipidaemia and OSA severity, and yielded knots of the AHI (9.4, 28.2, 54.4 and 80.2). After integrating the clinical definition and RCS-selected knots, all subjects were regrouped into four AHI severity stages. Following segmented multivariate linear modelling of each stage, distinguishable sets of OSA risk factors were quantified: low-density lipoprotein cholesterol (LDL-C), apolipoprotein E and high-density lipoprotein cholesterol (HDL-C); body mass index and/or waist to hip ratio; and HDL-C, LDL-C and triglycerides were specifically associated with stage I, stages II and III, and stages II–IV with different OSA indices.ConclusionsOur study revealed the multistage and non-monotonic relationships between OSA and dyslipidaemia and quantified the relationships between OSA severity indexes and distinct risk factors for specific OSA severity stages. Our study suggests that a new interpretive and predictive strategy for dynamic assessment of the risk progression over the clinical course of OSA should be adopted.

We demonstrated that patients with OSA had a higher percentage of dyslipidemia than subjects without OSA. Of the various components in serum lipid, only LDL-C was independently associated with OSA.