Computer-Aided Diagnosis Scheme for Distinguishing Between Benign and Malignant Masses on Breast DCE-MRI Images Using Deep Convolutional Neural Network with Bayesian Optimization

Although magnetic resonance imaging (MRI) has a higher sensitivity of early breast cancer than mammography, the specificity is lower. The purpose of this study was to develop a computer-aided diagnosis (CAD) scheme for distinguishing between benign and malignant breast masses on dynamic contrast material-enhanced MRI (DCE-MRI) by using a deep convolutional neural network (DCNN) with Bayesian optimization. Our database consisted of 56 DCE-MRI examinations for 56 patients, each of which contained five sequential phase images. It included 26 benign and 30 malignant masses. In this study, we first determined a baseline DCNN model from well-known DCNN models in terms of classification performance. The optimum architecture of the DCNN model was determined by changing the hyperparameters of the baseline DCNN model such as the number of layers, the filter size, and the number of filters using Bayesian optimization. As the input of the proposed DCNN model, rectangular regions of interest which include an entire mass were selected from each of DCE-MRI images by an experienced radiologist. Three-fold cross validation method was used for training and testing of the proposed DCNN model. The classification accuracy, the sensitivity, the specificity, the positive predictive value, and the negative predictive value were 92.9% (52/56), 93.3% (28/30), 92.3% (24/26), 93.3% (28/30), and 92.3% (24/26), respectively. These results were substantially greater than those with the conventional method based on handcrafted features and a classifier. The proposed DCNN model achieved high classification performance and would be useful in differential diagnoses of masses in breast DCE-MRI images as a diagnostic aid.     

Bayesian evidence synthesis in case of multi-cohort datasets: An illustration by multi-informant differences in self-control

The trend toward large-scale collaborative studies gives rise to the challenge of combining data from different sources efficiently. Here, we demonstrate how Bayesian evidence synthesis can be used to quantify and compare support for competing hypotheses and to aggregate this support over studies. We applied this method to study the ordering of multi-informant scores on the ASEBA Self Control Scale (ASCS), employing a multi-cohort design with data from four Dutch cohorts. Self-control reports were collected from mothers, fathers, teachers and children themselves. The available set of reporters differed between cohorts, so in each cohort varying components of the overarching hypotheses were evaluated. We found consistent support for the partial hypothesis that parents reported more self-control problems than teachers. Furthermore, the aggregated results indicate most support for the combined hypothesis that children report most problem behaviors, followed by their mothers and fathers, and that teachers report the fewest problems. However, there was considerable inconsistency across cohorts regarding the rank order of children’s reports. This article illustrates Bayesian evidence synthesis as a method when some of the cohorts only have data to evaluate a partial hypothesis. With Bayesian evidence synthesis, these cohorts can still contribute to the aggregated results.     

The Strategic Use of Noise in Pragmatic Reasoning

We combine two recent probabilistic approaches to natural language understanding, exploring the formal pragmatics of communication on a noisy channel. We first extend a model of rational communication between a speaker and listener, to allow for the possibility that messages are corrupted by noise. In this model, common knowledge of a noisy channel leads to the use and correct understanding of sentence fragments. A further extension of the model, which allows the speaker to intentionally reduce the noise rate on a word, is used to model prosodic emphasis. We show that the model derives several well-known changes in meaning associated with prosodic emphasis. Our results show that nominal amounts of actual noise can be leveraged for communicative purposes.     

验前概率联合冠脉CT造影对于稳定型冠心病的诊断价值

目的：比较升级的Diamond-Forrester法(updated Diamond-Forrester method,UDFM)和Duke临床评分(Duke clinical score,DCS)对于冠心病的评估准确性,并进一步分析验前概率与冠脉CT造影(computed tomographic coronary angiography, CTCA)联合应用的诊断准确性。方法纳入2012年1月-2013年12月因稳定型心绞痛在解放军总医院心内科先后行CTCA和传统冠状动脉造影(conventional coronary angiography,CCA)的患者523例,分别用UDFM和DCS估算每例患者患冠心病的验前概率。以CCA结果为金标准,分析验前概率、CTCA及两者联合应用对冠心病的诊断准确性。理论验后概率根据贝叶斯公式进行计算。结果523例患者中有385例(74%)CCA结果为阳性。与UDFM相比,DCS将更多的CCA结果阳性患者分入高验前概率组(46%vs 23%,P＜0.0001)。DCS的ROC曲线下面积明显大于UDFM[0.77(0.73,0.82)vs 0.71(0.66,0.77),P=0.0009]。根据DCS估算结果划分的低、中和高3个验前概率亚组中,CTCA的敏感性、特异性、阳性预测值及阴性预测值分别是94%、98%和97%,94%、87%和55%,91%、94%和93%及96%、96%和77%。中验前概率亚组的理论验后概率十分接近实际验后概率(阳性：94.7%vs 93.6%,阴性：3.7%vs 4.0%)。结论对于稳定型心绞痛患者,DCS比UDFM更适用于冠心病验前概率的估算。将按DCS估算的验前概率与CTCA联合应用,能够有效提高CTCA的诊断准确性,并避免过度检查。     

Describing disease processes using a probabilistic logic of qualitative time

BackgroundClinical knowledge about progress of diseases is characterised by temporal information as well as uncertainty. However, precise timing information is often unavailable in medicine. In previous research this problem has been tackled using Allen's qualitative algebra of time, which, despite successful medical application, does not deal with the associated uncertainty.ObjectivesIt is investigated whether and how Allen's temporal algebra can be extended to handle uncertainty to better fit available knowledge and data of disease processes.MethodsTo bridge the gap between probability theory and qualitative time reasoning, methods from probabilistic logic are explored. The relation between the probabilistic logic representation and dynamic Bayesian networks is analysed. By studying a typical, and clinically relevant problem, the detection of exacerbations of chronic obstructive pulmonary disease (COPD), it is determined whether the developed probabilistic logic of qualitative time is medically useful.ResultsThe probabilistic logic extension of Allen's temporal algebra, called Qualitative Time CP-logic provides a tool to model disease processes at a natural level of abstraction and is sufficiently powerful to reason with imprecise, uncertain knowledge. The representation of the COPD disease process gives evidence that the framework can be applied functionally to a clinical problem.ConclusionThe combination of qualitative time and probabilistic logic offers a useful framework for modelling knowledge and data to describe disease processes in clinical medicine.     

基于LASSO变量选择联合贝叶斯网络构建恶性肿瘤相关急性肾损伤(AKI)风险预测模型

目的 利用套索（least absolute shrinkage and selection operator,LASSO）回归和贝叶斯网络分析方法,构建适合肿瘤患者急性肾损伤（acute kidney injury,AKI）发病风险的贝叶斯网络预测模型,为早期识别高危人群,制定AKI精准预防策略提供科学依据。方法 以2014年10月1日至2015年9月30日在复旦大学附属中山医院就诊的恶性肿瘤住院患者为研究对象。于医院数据管理平台收集患者年龄、性别、体重指数、既往病史、肿瘤类型/治疗、基础肝肾功能、生化和电解质指标等数据资料。通过LASSO回归筛选出与AKI发生显著相关的影响因素;借助贝叶斯网络分析进一步描述变量间相互作用并评价模型预测效能。结果 26 914名研究对象中,AKI发病率为12.4%（n=3 326）,其中肾癌（27.3%）,多发性骨髓瘤（24.1%）和急性粒细胞白血病（23.9%）患者的AKI发病率最高。LASSO回归筛选出22个与AKI发生相关性最显著的变量,包括年龄、性别、体重指数、糖尿病史、肿瘤类型/分期/治疗方式、肝功能、肾小球滤过率（estimated glomerular filtration rate,eGFR）/血清肌酐值/血尿酸、白蛋白、血红蛋白和白细胞计数、血钠/血钾等电解质。贝叶斯网络模型发现血红蛋白、eGFR、血氯和血磷与AKI的发生有直接联系;节点治疗方式通过影响血钠和白蛋白等中间节点间接影响AKI的发生;糖尿病和性别通过节点尿酸间接相连eGFR,而后者是AKI的父节点。模型推理在其他条件一致的情况下,贫血和eGFR ≤ 59 mL·min^-1·1.73 m^-2的患者发生AKI的概率最高（55.7%）;而上述指标均正常者AKI发病率最低（3.0%）。模型评价发现贝叶斯网络模型的分类准确率为88.8%,接受者操作特性曲线曲线下面积为0.806。结论 基于LASSO变量选择联合贝叶斯网络分析构建的模型在肿瘤相关AKI的影响因素分析中更符合实际理论,其在发病风险预测中有较好的临床应用价值。     

Craniofacial bone atrophy in Parry Romberg syndrome demonstrated using a Bayesian hierarchical model

PurposeParry Romberg syndrome (PRS) is a condition characterized by progressive hemifacial atrophy, predominantly affecting the soft tissues. Associated bone retraction is a common clinical feature of PRS but has never been assessed. Here we used 3D imaging and Bayesian statistics in order to demonstrate and quantify bone atrophy in PRS.Materials and methodsTen non-operated patients with PRS (4/10 males) and 12 age-matched controls (7/12 males) were included into the study. The average age at CT-scan was 9.67 ± 4.13 years for PRS patients and 12.5 ± 4.37 years for controls. Soft and hard tissue atrophy levels were quantified using computed tomography scans, based on the distances between surfaces of the affected side and the non-affected contralateral side, both for the skin and the bone. We used a hierarchical Bayesian model with clinical priors in order to assess the relationship between hard and soft tissue atrophies.ResultsPRS patients had significant hard tissue atrophy, and atrophy extents were similar for soft and hard tissues. There was a trend for a correlation between the extent of hard tissue retraction and the extent of soft tissue retraction, and we could not demonstrate that the relationship between hard and soft tissue retractions was different in PRS and controls.ConclusionOur results indicated that bone atrophy was most probably a primary process rather than a phenomenon secondary to soft tissue retraction. We have provided the first assessment of bone atrophy in PRS patients using Bayesian statistics.     

Phylogenetic reconstruction of Bantu kinship challenges Main Sequence Theory of human social evolution

Kinship provides the fundamental structure of human society: descent determines the inheritance pattern between generations, whereas residence rules govern the location a couple moves to after they marry. In turn, descent and residence patterns determine other key relationships such as alliance, trade, and marriage partners. Hunter-gatherer kinship patterns are viewed as flexible, whereas agricultural societies are thought to have developed much more stable kinship patterns as they expanded during the Holocene. Among the Bantu farmers of sub-Saharan Africa, the ancestral kinship patterns present at the beginning of the expansion are hotly contested, with some arguing for matrilineal and matrilocal patterns, whereas others maintain that any kind of lineality or sex-biased dispersal only emerged much later. Here, we use Bayesian phylogenetic methods to uncover the history of Bantu kinship patterns and trace the interplay between descent and residence systems. The results suggest a number of switches in both descent and residence patterns as Bantu farming spread, but that the first Bantu populations were patrilocal with patrilineal descent. Across the phylogeny, a change in descent triggered a switch away from patrifocal kinship, whereas a change in residence triggered a switch back from matrifocal kinship. These results challenge “Main Sequence Theory,” which maintains that changes in residence rules precede change in other social structures. We also indicate the trajectory of kinship change, shedding new light on how this fundamental structure of society developed as farming spread across the globe during the Neolithic.Kinship is the key structure underlying human society: descent determines how wealth, land, and position are inherited across generations, whereas residence describes the rules governing where a couple should move to once they are married (1). In turn, descent and residence patterns determine other key relationships within society such as alliance, trade, and marriage partners (2). Contemporary hunter-gatherer societies, which are often considered a model for preagricultural human societies, are predominantly bilateral, tracing descent through both lines, and multilocal, with each couple choosing where to live (2–4, but see ref. 5), allowing for flexibility in those societies. In the last 10,000 years, however, a number of groups developed agriculture, which led to the expansion of food production techniques, cultures, and in some cases their populations too (6). The emergence of farming is thought to have coincided with more sex-biased dispersal and unilineal kinship (7).From their ancestral homeland in the Benue valley in Eastern Nigeria 3,000–5,000 BP (8, 9), possibly using a grassland corridor that opened up through the Cameroon rainforest (10, 11), the Bantu undertook one of the great farming expansions of the Neolithic (6). The history of their kinship is therefore key to debates about the fundamental processes that drove the evolution of human society during the Holocene. However, there are disagreements about the ancestral pattern of Bantu kinship and how this evolved through time.Vansina (12) suggests that proto-Bantu society had a bilateral descent and bilocal residence system (definitions in SI Appendix, Table S1) that was adaptive for expanding populations. Based on linguistic reconstruction, such as the proto-Bantu word for “house” being sex neutral, he argues that residence was flexible (13). Hunting required cooperation and mobility, and would be best served by males having a choice about their residence rather than being constrained by unilocality (12). Vansina suggests that only in the 18th or 19th century did unilineal descent and residence patterns begin to emerge, due to increased wealth and the disorder faced by some Bantu-speaking people (12).The second theory proposes a unilineal descent and unilocal residence system in the ancestral Bantu population (14). Hage and Marck argue, based on the linguistic reconstruction of kin terms, that the early Bantu speakers were, in fact, matrilocal and matrilineal (see also ref. 3), and dispute the sex neutrality of the proto-word for “house.” They also suggest that matriliny is consistent with a people that face an external threat experienced as populations expand and colonize new territory, arguing that absent males would trust their sisters, but not their wives, to look after their common lineage interests (15). Crucially, Marck and Bostoen (7) argue that matrifocal Bantu cultures dissolved with residence changing first from matrilocal to patrilocal, thus supporting Divale’s (16) proposal that changes in residence drive changes in inheritance patterns for a migrating population. This suggestion provides support for “Main Sequence Theory,” which proposes that worldwide there is a pattern of change with residence rules driving change in other social structures (2). In particular, Murdock argues “when any social system undergoes change, such change regularly begins with a modification in the rule of residence” (ref. 2, p. 221). Changes in descent follow and are always consistent with the change in residence. Changes in kinship terminology are affected by changes in both residence and descent and may follow some considerable time afterward (2).There is therefore no agreement among researchers about the ancestral states or patterns of change in kinship traits among Bantu societies based on historical linguistic methods. There are a number of problems with relying solely on reconstructions of ancestral vocabulary to infer social organization in the past. First, although linguists use systematic methods of reconstructing proto-forms based on regular sound changes (13, 14), there is still an element of subjectivity in this approach, which can lead different researchers to suggest contradictory results based on the same words (7, 13, 14). Furthermore, the inference of social organization in a past society is a further step removed from this process and assumes a direct relationship between particular words and particular forms of organization. Although it may be possible to reconstruct the sound of a particular lexical item, the meaning of such a word is less clear because of the possibility of semantic shifts (17).Phylogenetic comparative methods, adapted from evolutionary biology, offer an alternative way of reconstructing the evolutionary history of cultural traits such as kinship structures. These techniques map the traits of interest (in this case descent and residence) onto a phylogenetic tree, which represents the way societies are related to each other historically. The likely forms of these traits in past societies can then be inferred by using an explicit statistical model of trait evolution. Importantly, these reconstructions are probabilistic, meaning it is possible to assess how much confidence to place in any particular reconstruction. The likelihood of alternative hypotheses for the pattern of evolution of traits over the tree can then be estimated. Furthermore, performing analyses over a sample of phylogenetic trees can explicitly incorporate uncertainty about the historical relationships between societies. These methods can be used to estimate the cultural history of a language family, even where historical records or archaeological evidence are absent, and have been used to examine the history of residence patterns in both the Indo-European and Austronesian language families (18, 19).The Bantu speaking people of sub-Saharan Africa represent one of the major early farming expansions; the advantage of agricultural food production allowed a single language group to displace and expand into the lands of previous hunter-gatherer populations (6). This process has allowed for language phylogenies of Bantu societies to be generated (9, 20–22), which can be combined with comprehensive data on extant kinship patterns across a large number of Bantu cultures (23, 24) to infer historical patterns of cultural evolutionary change. A previous attempt to infer the ancestral state of descent for the original Bantu population (25, 26) was inconclusive, possibly because of the small sample size, the single phylogeny (20), or the maximum likelihood methods used.Here, we advance previous attempts to reconstruct the kinship traits across the Bantu language family by applying Bayesian phylogenetic comparative methods (27–29) to a dataset of 122 Bantu ethnolinguistic groups to infer the ancestral state and evolutionary trajectory of Bantu kinship patterns. Analysis of basic vocabulary items (30) has enabled the phylogenetic relationships among more than 500 Bantu languages to be inferred (9), which has increased the scope for the number of different Bantu cultures that can be analyzed and enabled the incorporation of groups that were underrepresented in previous analyses. In particular, the current sample includes a number of additional societies from the “Northwest” and “Forest West” regions, which may be particularly important in inferring trait states at the earliest nodes in the trees. Furthermore, Bayesian phylogenetic methods, which can incorporate uncertainty about the phylogenetic relationships between cultures, provide a more accurate picture of the inferences that can be drawn from comparative data. In this way, it is possible to infer the ancestral states of kinship traits and, hence, reconstruct the cultural history of the Bantu expansion, which means it is possible to estimate the likely order of trait change and test whether changes in residence rules precede changes in inheritance patterns as proposed by Main Sequence Theory (2, 7, 16).