Two-stage Clustering Technique Research Articles

Classification of anatomic structures in head and neck by CT-based radiomics

Background and purpose: Radiomics features are used to identify disease types and predict therapy outcomes. However, how the radiomics features are different among different anatomical structures has never been investigated. Hence, we analyzed the radiomics features of 22 anatomical structures in the head and neck area in CT images. Furthermore, we studied whether CT radiomics can classify anatomical structures of the head and neck using unsupervised machine-learning techniques. Materials and methods: We obtained IMRT/VMAT treatment planning data from 36 patients treated for head and neck cancers in a single institution. There were 1357 contours of more than 22 anatomical structures drawn on planning CTs. We calculated 174 radiomics features using the SIBEX program. First, we tested whether the radiomics features of anatomical structures were unique enough to classify all contours into 22 groups. We then developed a two-stage clustering technique to classify 22 anatomic structures into sub-groups with similar physiological or biological characteristics. Results: The heatmap of 174 radiomics features of 22 anatomical structures showed a distinct difference among tumors and other healthy structures. Radiomics features have allowed us to identify the eyes, lens, submandibular, pituitary glands, and thyroids with over 90% accuracy. The two-stage clustering of 22 structures resulted in six subgroups, which shared common characteristics such as fatty and bony tissues. Conclusions: We have shown that anatomical structures in head and neck tumors have distinguishable radiomics features. We could observe similarities of features among subgroups of the structures. The results suggest that CT radiomics can help distinguish the biological characteristics of head and neck lesions.

Identification of movement categories and associated velocity thresholds for elite Gaelic football and hurling referees

ABSTRACT The purpose of this study was to generate movement category velocity thresholds for elite Gaelic football (GF) and hurling referees using a two-stage unsupervised clustering technique. Activity data from 41 GF and 38 hurling referees was collected using global positioning system technology during 338 and 221 competitive games, respectively. The elbow method was used in stage one to identify the number of movement categories in the datasets. In stage two, the respective velocity thresholds for each category were identified using spectral clustering. The efficacy of these thresholds was examined using a regression analysis performed between the median of each of the velocity thresholds and the raw velocity data. Five velocity thresholds were identified for both GF and hurling referees (mean ± standard deviation: GF referees; 0.70±0.09, 1.66±0.19, 3.28±0.41, 4.87±0.61, 6.49±0.50 m·s−1; hurling referees; 0.69±0.11, 1.60±0.25, 3.09±0.52, 4.63±0.58, 6.35±0.43 m·s−1). With the exception of the lowest velocity threshold, all other thresholds were significantly higher for GF referees. The newly generated velocity thresholds were more strongly associated with the raw velocity data than traditional generic categories. The provision of unique velocity thresholds will allow applied practitioners to better quantify the activity profile of elite GF and hurling referees during training and competition.

Mapping future droughts under global warming across China: A combined multi-timescale meteorological drought index and SOM-Kmeans approach

Global warming will exacerbate the damages and losses caused by drought and significantly affect human society and economic development. Future drought outlook is of vital importance for policy-making to combat drought risk and water crises over the long term. Here, 3- and 12-month standardized precipitation evapotranspiration indices, which can aid short- and long-term drought monitoring, respectively, are coupled with the SOM-Kmeans two-stage clustering technique to map the future drought conditions in China for the future period (2021–2050) relative to the baseline period (1976–2005). The calculation is based on downscaled outputs of 10 models from the Coupled Model Intercomparison Project Phase 5 (CMIP5) under two representative concentration pathways (RCPs). The temperature increases by 0.5–2.0 °C and 1.0–2.5 °C under the selected RCP4.5 and RCP8.5 scenarios, respectively. The results show that the differences between precipitation and potential evapotranspiration are projected to increase throughout China. Most parts of China are projected to experience more severe, more frequent and longer duration droughts in the next three decades compared with the baseline period. Global warming is projected to have a serious impact on future droughts in the basins located in the northern part of China. The drought conditions in the northwest region of the “Hu Huanyong Line” in China are projected to be serious and complex in the future, while the opposite is true in the southeastern region. Drought conditions in northern China are more complex under high temperature changes (1.0–2.5 °C) than under low changes (0.5–2.0 °C). The results of this study provide insights on future drought planning and management in China and demonstrate that SOM-Kmeans can be used for the classification of drought conditions.

The application of data mining and RFM model in market segmentation of a veterinary hospital

This study adopts a two-stage clustering technique, the combination of self-organizing maps and K-means method, and RFM (recency, frequency, and monetary) model to categorize customers in a veterinary hospital in Taichung City, Taiwan in order to make effective marketing strategies in this competitive market. Based on 1,784 customers with the focus solely on cats in 2014, twelve clusters are formed. Specifically, six out of twelve clusters are classified into the best and loyal customers. Three clusters are uncertain and lost customers. One cluster is viewed as the best but lost customers. Finally, two clusters that have relatively higher recency values than the average value belong to uncertain but new customers based on RFM model and based on RFM model.

Functional parcellation of the hippocampus from resting-state dynamic functional connectivity

The hippocampus consists of functionally and structurally heterogeneous regions that are involved in multiple functions such as learning and memory. Previous studies on connectivity-based functional subdivisions of the hippocampus, however, overlooked the dynamic nature of resting-state functional connectivity (FC). In this study, we selected 50 subjects with the lowest head motion from the Human Connectome Project dataset and performed a two-stage spectral clustering technique to windowed FC correlations for identifying the potential covariant structures during the spontaneous fluctuation of hippocampal-cortical FC. The obtained covariant structures were believed to be functionally homogeneous by coupling with whole-brain regions in all transient connectivity states and consequently subdivided the left and right hippocampus into six and five functional subregions, respectively. Further, we demonstrated that this dynamic-FC-derived hippocampal parcellation exhibited significantly improved reproducibility of segmented subregions across subjects compared with static FC analysis. The findings extend our understanding to the functional organization within the hippocampus and provide a more comprehensive view of the functional flexibility of the hippocampus over time.

EcoAnthromes of Alberta: An example of disturbance-informed ecological regionalization using remote sensing

Humans influence ecosystems on magnitudes that often exceed that of natural forces such as climate and geology; however, frameworks rarely include anthropogenic disturbance when delineating unique ecological regions. A critical step toward understanding, managing and monitoring human-altered ecosystems is to incorporate disturbance into ecological regionalizations. Furthermore, quantitative regionalization approaches are desirable to provide cost-effective, repeatable and statistically sound stratification for environmental monitoring. We applied a two-stage multivariate clustering technique to identify ‘EcoAnthromes’ across a large area – the province of Alberta, Canada – at 30 m spatial resolution, and using primarily remotely sensed inputs. The EcoAnthrome clusters represent regions with unique ecological characteristics based on a combination of natural ecological potential (e.g., climatic and edaphic factors) and disturbance, both natural and anthropogenic. Compared to existing expert-derived Natural Subregions in Alberta, the model-based EcoAnthromes showed greater class separation and explained more variance for an assortment of variables related to land cover, disturbance and species intactness. The EcoAnthromes successfully separated important ecological regions that are defined by complex assemblages of topography, climate and disturbance, such as gravel-bed river valleys, boreal forests, grasslands, post-fire recovery areas and highly disturbed agricultural, industrial and urban landscapes. In addition to presenting a flexible method for EcoAnthrome regionalization, we group and describe the EcoAnthromes created for Alberta and discuss how they can complement expert-derived regionalizations to aid in environmental management efforts, such as species recovery planning and monitoring for threatened species.

Two-stage Clustering Technique Based on the Neighboring Union Histogram for Hyperspectral Remote Sensing Images

This paper presents a novel unsupervised classification approach suited to hyperspectral remote sensing image data sets that uses $K$ -means clustering combined with the neighboring union histogram (NUH). The approach is implemented in five main steps as follows: 1) dividing the hyperspectral image data intt uncorrelated groups based on the correlation coefficient matrix; 2) extracting the first few principal component analysis (PCA) components from each group; 3) computing the NUHs of every group; 4) obtaining several relatively rough clustering results by employing the first-stage $K$ -means procedure to classify every group’s NUHs; and 5) using the second-stage $K$ -means procedure to refine the rough clustering results for the final clustering map. The NUH indicates the regional statistical features of a point, thereby making the proposed approach insensitive to noise and abnormal data. The two-stage clustering technique improves the recognition rate of similar land cover classes. The proposed approach is compared with five other unsupervised methods. The experimental results on two different types of real-world hyperspectral remote sensing images validate the high accuracy of the proposed approach.

Customer-oriented benefit segmentation: an integrated approach

Segmentation is a common and important task for most marketing departments. Besides, other marketing decisions are influenced by market segmentation results. Benefit segmentation is one of the best approaches for market segmentation among others. In this paper, we proposed a novel hybrid benefit segmentation approach which applied two-stage clustering, conjoint analysis, and Delphi technique to segment customers according to their benefits. More precisely, we used two-stage clustering based on main benefits derived from conjoint analysis to classify customers into different segments. Conjoint analysis was used to assess and balance dissimilar aspects of consumers' needs. Delphi method was used for the selection of the most important criteria and their levels which conjoint analysis required all of them for evaluation. This approach did not only help managers to segment customers but also to provide a way to analyse consumer behaviour, determine the marking strategy and extract the importance of each attribute in each segment.

Clustering the clusters - knowledge enhancing tool for diagnosing elderly falling risk

Falls which affect the musculoskeletal system are the leading cause of injury in people over 65 years. To address the growing problem of falls in an ageing society and to support and improve the healthcare service provided, a diagnostic tool is required. This study proposes a new approach to analyse and diagnose the risks associated with elderly falling by applying K-means clustering to cluster and assess the fall risks data of elderly Thai people, captured using motion capture technology. These clusters are mapped into two-dimensional space using self-organising map (SOM). The resulting 95.45% accuracy suggests that the two-stage clustering technique is applicable and useful in managing fall risks which can then be included in decision support system to assist physiotherapists, in recommending a customised rehabilitation programme.

군집분석과 상세기상모델을 통한 포항지역 계절별 바람장 특성

The typical characteristics of seasonal winds were studied around the Pohang using two-stage (average linkage then k-means) clustering technique based on u- and v-component wind at 850 hpa from 2004 to 2006 (obtained the Pohang station) and a high-resolution (0.5 km grid for the finest domain) WRF-UCM model along with an up-to-date detailed land use data during the most predominant pattern in each season. The clustering analysis identified statistically distinct wind patterns (7, 4, 5, and 3 clusters) representing each spring, summer, fall, and winter. During the spring, the prevailed pattern (80 days) showed weak upper northwesterly flow and late sea-breeze. Especially at night, land-breeze developed along the shoreline was converged around Yeongil Bay. The representative pattern (92 days) in summer was weak upper southerly flow and intensified sea-breeze combined with sea surface wind. In addition, convergence zone between the large scale background flow and well-developed land-breeze was transported around inland (industrial and residential areas). The predominant wind distribution (94 days) in fall was similar to that of spring showing weak upper-level flow and distinct sea-land breeze circulation. On the other hand, the wind pattern (117 days) of high frequency in winter showed upper northwesterly and surface westerly flows, which was no change in daily wind direction.

Re-engaging with the physical environment: a health-related environmental classification of the UK

This paper presents a health-related area-level multiple environmental classification of the UK and examines ecological associations with health. This classification, akin to a geodemographic profile of the environment, classifies small areas across the UK into seven environment types ranging from ‘Industrial’ to ‘Sunny, clean and green’. The data for the classification were gathered from a range of agencies, rendered to Census Area Statistic Wards (n = 10 654) and processed through a two-stage clustering technique to create a Multiple Environmental Deprivation Classification, or MEDClass. In order to explore the utility of MEDClass, this paper presents an empirical investigation into the extent to which the type of physical environment one lives in can influence self-reported health and mortality rates. The findings suggest that while physical environment ‘type’ makes a modest contribution towards our understanding of health inequalities, socio-economic deprivation remains the most important challenge for those seeking to address these inequalities. In conclusion we suggest that human geographers should embrace a broader conceptualisation of the environment and in particular, re-engage with traditional aspects of the physical environment.

Name Clustering on the Basis of Parental Preferences

AbstractParents do not choose first names for their children at random. Using two large datasets, for the UK and the Netherlands, covering the names of children born in the same family over a period of two decades, this paper seeks to identify clusters of names entirely inferred from common parental naming preferences. These name groups can be considered as coherent sets of names that have a high probability to be found in the same family. Operational measures for the statistical association between names and clusters are developed, as well as a two-stage clustering technique. The name groups are subsequently merged into a limited set of grand clusters. The results show that clusters emerge with cultural, linguistic, or ethnic parental backgrounds, but also along characteristics inherent in names, such as clusters of names after flowers and gems for girls, abbreviated names for boys, or names ending in –y or -ie.

The synoptic-scale surface wind climate regimes of the Mediterranean Sea according to the cluster analysis of ERA-40 wind fields

In 2008, Burlando et al. proposed a two-stage clustering technique for the classification of mesoscale wind regimes. The first stage of this classification scheme is based on a hierarchical cluster analysis, according to Ward’s minimum variance technique applied to an Euclidean distance, for a first-guess subdivision of the events into groups. In the second stage, a partitional k-means clustering for the optimal reassignment of the events among clusters is performed. Following this methodology, in the present paper, the synoptic-scale wind fields over the Mediterranean Sea have been analysed in order to check the suitability of this technique to a higher-dimensional phase space. The study is based on a 30-year-long data set of wind speed and direction at 10 m above sea level obtained from the reanalysis of the European Centre for Medium-range Weather Forecasts. The cluster analysis has been performed on wind speeds only, while wind directions have been used to show the existence, in the spatial structure of the wind climate regimes, of particular regions which correspond to the main topographic and/or thermal forcing of the Mediterranean. These regions are the peaks of the probability density function in the climatological phase space of wind speed patterns recognised by the clustering algorithm. The final classification has been able to identify, for instance, the surface circulation patterns corresponding to Mistral events in the western Mediterranean sub-basin and Etesian winds in the eastern sub-basin.

Meteorological pattern classification and application for forecasting air pollution episode potential in a mountain-valley area

In this study, we developed an automated scheme to classify the synoptic meteorological conditions governing over the northern Thailand during the winter period (November–January) into six-synoptic patterns representing different development stages of the regional pressure system. The meteorological classification scheme is hybrid in nature combining the air mass and the flow pattern based approaches. The classification is done though the principal component analysis of data from six regional surface meteorological stations and one upper-air station at 01:00 Thailand Local Standard Time daily for 5-year winter periods, which is followed by a two-stage clustering technique. Examination of the SO2 pollution levels in the complicated terrain of the Mae Moh valley in the northern Thailand, where a coal-fired power plant is located, shows that the scheme successfully identifies the patterns that are conducive to high pollutant and those that are conducive to low pollutant build up in the valley. Levels and spatial distributions of the daily highest 1-h SO2 concentrations in the valley exhibited similar characteristics within each pattern, and distinct differences between the six-synoptic patterns. The scheme is recommended for the early warning of the episode potentials in the valley. The stepwise regression was performed to develop prediction models for the two most polluted synoptic patterns, which have high frequencies of exceeding the National Ambient Air Quality Standards. The regression equations have R2 values above 60% and produce the daily highest 1-h SO2 concentrations with average relative errors of around 33%.

Feature-based recommendations for one-to-one marketing.

Most recommendation systems face challenges from products that change with time, such as popular or seasonal products, since traditional market basket analysis or collaborative filtering analysis are unable to recommend new products to customers due to the fact that the products are not yet purchased by customers. Although the recommendation systems can find customer groups that have similar interests as target customers, brand new products often lack ratings and comments. Similarly, products that are less often purchased, such as furniture and home appliances, have fewer records of ratings; therefore, the chances of being recommended are often lower. This research attempts to analyze customers' purchasing behaviors based on product features from transaction records and product feature databases. Customers' preferences toward particular features of products are analyzed and then rules of customer interest profiles are thus drawn in order to recommend customers products that have potential attraction with customers. The advantage of this research is its ability of recommending to customers brand new products or rarely purchased products as long as they fit customer interest profiles; a deduction which traditional market basket analysis and collaborative filtering methods are unable to do. This research uses a two-stage clustering technique to find customers that have similar interests as target customers and recommend products to fit customers' potential requirements. Customers' interest profiles can explain recommendation results and the interests on particular features of products can be referenced for product development, while a one-to-one marketing strategy can improve profitability for companies.

An Upper-Air Synoptic Climatology of the Western United States

Abstract An automated, year-round synoptic climatology is developed for the western United States from rawinsonde observations from 1979 to 1988. The classification uses thermal, moisture, and flow parameters to characterize seasonal and interannual synoptic-scale variations in hydrodynamic and thermodynamic conditions. Based on twice daily observations for a network of 21 stations from the Pacific coast to the Rockies, a synoptic climatology is developed using air temperature, dewpoint temperature, geopotential height, and the east-west and north-south components of the wind vector at 800-, 700-, 500-, and 250-mb constant pressure surfaces. The 798 variable by 3620 day matrix is reduced to six orthogonal principal components, and the resulting component scores are grouped using a two-stage clustering technique. The 13 synoptic situations represent days experiencing homogeneous weather conditions. These synoptic situations exhibit marked seasonality and interannual variability and depict features observed...

Development of an automated spatial synoptic climatological classification

AbstractA spatial synoptic climatological index is developed that both characterizes the dynamic and thermodynamic surface weather regime and is capable of assessing and analysing the impact of synoptic‐scale variations on numerous environmental parameters. Based upon surface weather data for the contiguous USA for 1984, daily maps are produced that show those areas under the influence of similar weather conditions. The index is developed by applying principal components analysis to a daily matrix of 141 stations by 24 weather elements (chosen to represent the thermal‐moisture and surface flow characteristics). The resulting component scores are grouped using a two‐stage clustering technique. This yields a map showing lobes of air having relatively homogeneous weather characteristics. These lobes (clusters) are spatially coherent and correspond closely to surface weather map features.To allow for daily tracking of the clusters and identification days or weeks after disappearance, the means of each of the daily clusters are grouped across days. The resulting 90‐cluster categorization describes all of the major synoptic situations that occurred in 1984. Some of the common categories include maritime and continental polar air masses, frontal passages, and continental high dominance. These groups demonstrate seasonality, regionality, and high within‐category similarity. The resulting maps are useful since they are surface weather maps yet add clarity for environmental interpretation.

USING A SPATIAL SYNOPTIC CLIMATOLOGICAL CLASSIFICATION TO ASSESS CHANGES IN ATMOSPHERIC POLLUTION CONCENTRATIONS

A spatial synoptic climatological classification is developed for assessing spatial and temporal changes in atmospheric pollution concentrations. Based upon surface weather data for the conterminous United States for 1984, daily maps are produced depicting areas influenced by similar weather conditions. The classification is developed by applying principal components analysis to a daily matrix of 141 stations by 24 weather elements, chosen to represent both thermal/moisture conditions and surface air flow. The principal component scores are grouped using a two-stage clustering technique. The result is a map showing lobes of air which have relatively homogeneous meteorological characteristics and correspond to surface weather map features. The classification is then related to atmospheric concentrations of sulfur dioxide and nitrogen dioxide in nine cities. High concentrations are found in stable air masses and during strong anticyclonic control, whereas low concentrations occur in moist, unstable air mass...