Cluster Centroids Research Articles

APPLICATION OF K-MEANS++ WITH DUNN INDEX VALIDATION OF GROUPING WEST KALIMANTAN REGION BASED ON CRIME VULNERABILITY

Crime is an unlawful behavior that will be given a punishment or sanctions based on Kitab Undang-Undang Hukum Pidana (KUHP) or other regulations in Indonesia. One of the provinces in Indonesia, namely West Kalimantan reported that criminal cases are increasing in 2021 and 2022. One of the solutions to minimize that case is grouping the district and city in West Kalimantan based on the level of vulnerability so the authority can be more responsive in solving these problems. The grouping can be done by cluster analysis. This analysis aims to group some objects based on the similarity of characteristics. K-Means++ is one of the methods of cluster analysis. K-Means++ is the development of K-Means, in which K-Means++ is smarter than K-Means in selecting the initial centroid because only one initial centroid is chosen randomly, and the initial centroids of the other clusters are done through calculations. This research uses secondary data from BPS of West Kalimantan, consisting of 10 variables. This research aims to form clusters to determine the level of vulnerability of each district and city in West Kalimantan. The selection of the optimal cluster is done by evaluating the cluster. One of these evaluations is the Dunn Index. Based on the analysis results, the optimum number of clusters is with a Dunn Index value of 0.55. The first cluster is categorized as non-vulnerable with ten members, the second cluster as vulnerable with three members, and the third cluster as very vulnerable with one member.

12500 Objective Evaluation Of Clustering Methods For Resolving The Heterogeneity Of PCOS

Abstract Disclosure: K. Brewer: None. R. Sisk: None. M. Dapas: None. C. Li: None. A. Dunaif: Consulting Fee; Self; AcaciaBio, Inc, Neurocine Biosciences, Inc. Speaker; Self; Quest Diagnostics. Other; Self; Co-Editor Endocrine Today, Healio, Slack Inc. Clustering methods have been used to resolve heterogeneity within complex diseases to elucidate underlying biologic mechanisms. We (Dapas et al. PLoS Med, 2020) and others have applied these methods to PCOS and identified discrete subtypes, including those that capture distinct reproductive or metabolic features. We objectively compared two widely used methods, hierarchical clustering (HC), which recursively merges subjects based on their similarity, and k-means (Km), which iteratively groups individuals to minimize their distance to cluster centroids. The number of clusters (k) was predefined in both approaches. We compared HC vs Km in 874 European ancestry PCOS cases diagnosed by NIH criteria using 8 traits (BMI, testosterone, SHBG, DHEAS, LH, FSH, fasting insulin, fasting glucose) and 9 traits with the addition of AMH. The ConsensusClusterPlus R package was used to evaluate cluster stability of both approaches with k=2, 3, or 4, using 8 or 9 traits. Genomewide association study (GWAS) meta-analysis was performed as reported with a discovery cohort (620 cases, 2951 controls) and a replication cohort (371 cases, 926 controls), except that genotypes were imputed to the newer TOPMED (r2) panel. Clustering using 8 traits and k=3 resulted in the best cluster stability for both HC and Km (pairwise consensus proportion: HC 0.98, Km 0.93) compared to k=2 with 8 traits (HC 0.88, Km 0.88) or 9 traits (HC 0.95, Km 0.92). No stability was observed for k=4 with 8 or 9 traits using either method (range: 0.35-0.66). Km showed slightly better cluster separation than HC (average silhouette width: Km 0.12 vs HC 0.09) with k=3 and 8 traits. The biologic relevance of the three PCOS subtypes, which we have designated as reproductive (cases with higher LH, FSH, SHBG), metabolic (higher BMI, insulin, glucose), and background, was assessed by GWAS, an orthogonal (i.e., uncorrelated) approach for subtype confirmation. The subtypes identified with HC had genomewide significant associations (metabolic subtype, c9orf3/FANCC, rs10761370, minor allele frequency [MAF] 0.47, p=1.21 x 10-8; background subtype, FSHB/ARL14EP, rs10835649, MAF 0.17, p=8.49 x 10-[1]0). There were no genomewide significant signals when the subtypes were identified with Km despite having a better cluster separation. In summary, only HC clusters appeared to capture biologically meaningful differences since two of the three subtypes thus identified were associated with genomewide significant loci. Neither the addition of AMH nor increasing the number of groups improved clustering metrics. Our results emphasize the importance of independent validation of clustering approaches using an orthogonal confirmation strategy such as GWAS. We conclude that HC clustering using 8 traits is superior to Km for resolving the genetic heterogeneity of PCOS. Presentation: 6/2/2024

12500 Objective Evaluation Of Clustering Methods For Resolving The Heterogeneity Of PCOS

Abstract Disclosure: K. Brewer: None. R. Sisk: None. M. Dapas: None. C. Li: None. A. Dunaif: Consulting Fee; Self; AcaciaBio, Inc, Neurocine Biosciences, Inc. Speaker; Self; Quest Diagnostics. Other; Self; Co-Editor Endocrine Today, Healio, Slack Inc. Clustering methods have been used to resolve heterogeneity within complex diseases to elucidate underlying biologic mechanisms. We (Dapas et al. PLoS Med, 2020) and others have applied these methods to PCOS and identified discrete subtypes, including those that capture distinct reproductive or metabolic features. We objectively compared two widely used methods, hierarchical clustering (HC), which recursively merges subjects based on their similarity, and k-means (Km), which iteratively groups individuals to minimize their distance to cluster centroids. The number of clusters (k) was predefined in both approaches. We compared HC vs Km in 874 European ancestry PCOS cases diagnosed by NIH criteria using 8 traits (BMI, testosterone, SHBG, DHEAS, LH, FSH, fasting insulin, fasting glucose) and 9 traits with the addition of AMH. The ConsensusClusterPlus R package was used to evaluate cluster stability of both approaches with k=2, 3, or 4, using 8 or 9 traits. Genomewide association study (GWAS) meta-analysis was performed as reported with a discovery cohort (620 cases, 2951 controls) and a replication cohort (371 cases, 926 controls), except that genotypes were imputed to the newer TOPMED (r2) panel. Clustering using 8 traits and k=3 resulted in the best cluster stability for both HC and Km (pairwise consensus proportion: HC 0.98, Km 0.93) compared to k=2 with 8 traits (HC 0.88, Km 0.88) or 9 traits (HC 0.95, Km 0.92). No stability was observed for k=4 with 8 or 9 traits using either method (range: 0.35-0.66). Km showed slightly better cluster separation than HC (average silhouette width: Km 0.12 vs HC 0.09) with k=3 and 8 traits. The biologic relevance of the three PCOS subtypes, which we have designated as reproductive (cases with higher LH, FSH, SHBG), metabolic (higher BMI, insulin, glucose), and background, was assessed by GWAS, an orthogonal (i.e., uncorrelated) approach for subtype confirmation. The subtypes identified with HC had genomewide significant associations (metabolic subtype, c9orf3/FANCC, rs10761370, minor allele frequency [MAF] 0.47, p=1.21 x 10-8; background subtype, FSHB/ARL14EP, rs10835649, MAF 0.17, p=8.49 x 10-[1]0). There were no genomewide significant signals when the subtypes were identified with Km despite having a better cluster separation. In summary, only HC clusters appeared to capture biologically meaningful differences since two of the three subtypes thus identified were associated with genomewide significant loci. Neither the addition of AMH nor increasing the number of groups improved clustering metrics. Our results emphasize the importance of independent validation of clustering approaches using an orthogonal confirmation strategy such as GWAS. We conclude that HC clustering using 8 traits is superior to Km for resolving the genetic heterogeneity of PCOS. Presentation: 6/2/2024

A Review of Convex Clustering From Multiple Perspectives: Models, Optimizations, Statistical Properties, Applications, and Connections.

Traditional partition-based clustering is very sensitive to the initialized centroids, which are easily stuck in the local minimum due to their nonconvex objectives. To this end, convex clustering is proposed by relaxing K -means clustering or hierarchical clustering. As an emerging and excellent clustering technology, convex clustering can solve the instability problems of partition-based clustering methods. Generally, convex clustering objective consists of the fidelity and the shrinkage terms. The fidelity term encourages the cluster centroids to estimate the observations and the shrinkage term shrinks the cluster centroids matrix so that their observations share the same cluster centroid in the same category. Regularized by the lpn -norm ( pn ∈ {1,2,+∞} ), the convex objective guarantees the global optimal solution of the cluster centroids. This survey conducts a comprehensive review of convex clustering. It starts with the convex clustering as well as its nonconvex variants and then concentrates on the optimization algorithms and the hyperparameters setting. In particular, the statistical properties, the applications, and the connections of convex clustering with other methods are reviewed and discussed thoroughly for a better understanding the convex clustering. Finally, we briefly summarize the development of convex clustering and present some potential directions for future research.

Clustering Anomalous Circulations and Their Implications on the Prediction of Minimum 2-m Air Temperature over Northeastern China during Winter

Abstract The ERA5 reanalysis during cold months (November–March) of 1979–2020 was used for determining four cluster centroids through the k-means for classifying regional anomalies of the daily geopotential height at 500 hPa (H500) over northeastern China. Empirical orthogonal function (EOF) was used to reduce dimensionality. Four clusters were linked to the EOF patterns with clear meteorological meanings, which are associated with the evolution of ridges and troughs over northeastern China. Those systems relate to warm and cold advections at 850 hPa. In each H500 cluster, the advection is the major contributor leading to temperature changes at 850 hPa, which significantly relates to the changes and anomalies of daily minimum air temperature at 2 m (T2min). Furthermore, the jet activities over Asia relate to more or less occurrence of specific H500 clusters in jet phases. This is because anomalous westerlies are generally in favor of positive anomalies of the vorticity tendency at 500 hPa. For the reforecasts during 2004–19 in the Chinese Meteorology Administration (CMA) S2S model, the hit rates above 50% for all the H500 clusters are within 9.5 days, which are in between those for the first two and the last two clusters. The correct prediction of H500 anomalies improves the T2min prediction up to 12 days, compared with 8 days for the incorrect one. The good prediction of the jet activities leads to a more accurate prediction of H500 anomalies. Therefore, improvement of the model prediction of jet activities and H500 anomalies will lead to better prediction of winter weather near the ground over northeastern China.

UAV clusters information geometry fusion positioning method with LEO satellite system

The existing Low-Earth-Orbit (LEO) positioning performance cannot meet the requirements of Unmanned Aerial Vehicle (UAV) clusters for high-precision real-time positioning in the Global Navigation Satellite System (GNSS) denial conditions. Therefore, this paper proposes a UAV Clusters Information Geometry Fusion Positioning (UC-IGFP) method using pseudo-ranges from the LEO satellites. A novel graph model for linking and computing between the UAV clusters and LEO satellites was established. By utilizing probability to describe the positional states of UAVs and sensor errors, the distributed multivariate Probability Fusion Cooperative Positioning (PF-CP) algorithm is proposed to achieve high-precision cooperative positioning and integration of the cluster. Criteria to select the centroid of the cluster were set. A new Kalman filter algorithm that is suitable for UAV clusters was designed based on the global benchmark and Riemann information geometry theory, which overcomes the discontinuity problem caused by the change of cluster centroids. Finally, the UC-IGFP method achieves the LEO continuous high-precision positioning of UAV clusters. The proposed method effectively addresses the positioning challenges caused by the strong direction of signal beams from LEO satellites and the insufficient constraint quantity of information sources at the edge nodes of the cluster. It significantly improves the accuracy and reliability of LEO-UAV cluster positioning. The results of comprehensive simulation experiments show that the proposed method has a 30.5% improvement in performance over the mainstream positioning methods, with a positioning error of 14.267 m.

Method for Extracting Inland Unmanned Vessel Routes Based on AIS Data

This study introduces a novel approach to enhance inland navigation safety and optimize unmanned vessel routes using AIS data. The trajectory partition algorithm categorizes Yangtze River trajectory data to formulate round-trip routes. A turning point identification algorithm contributes to recognizing crucial turning points, followed by clustering using the HDBSCAN algorithm. Cluster centroids extracted from these clusters serve as essential waypoints for route planning. The Akima interpolation polynomial is applied judiciously to interpolate waypoints, culminating in meticulous route generation. Validation employs a dataset of 5,480,049 dynamic trajectory points from the Yangtze River, demonstrating the method's efficacy. Results indicate mean squared errors of 0.77% and 6.21%, symmetrical mean absolute percentage errors of 5.3% and 7.3%, and correlation coefficients of 99.62% and 97.14% with actual routes, respectively. This method adeptly extracts inland vessel round-trip routes, aligning seamlessly with stringent criteria for accuracy, effectiveness, and applicability.

Graph Neural Network-Based Modeling with Subcategory Exploration for Drug Repositioning

Drug repositioning is a cost-effective approach to identifying new indications for existing drugs by predicting their associations with new diseases or symptoms. Recently, deep learning-based models have become the mainstream for drug repositioning. Existing methods typically regard the drug-repositioning task as a binary classification problem to find the new drug–disease associations. However, drug–disease associations may encompass some potential subcategories that can be used to enhance the classification performance. In this paper, we propose a prototype-based subcategory exploration (PSCE) model to guide the model learned with the information of a potential subcategory for drug repositioning. To achieve this, we first propose a prototype-based feature-enhancement mechanism (PFEM) that uses clustering centroids as the attention to enhance the drug–disease features by introducing subcategory information to improve the association prediction. Second, we introduce the drug–disease dual-task classification head (D3TC) of the model, which consists of a traditional binary classification head and a subcategory-classification head to learn with subcategory exploration. It leverages finer-grained pseudo-labels of subcategories to introduce additional knowledge for precise drug–disease association classification. In this study, we conducted experiments on four public datasets to compare the proposed PSCE with existing state-of-the-art approaches and our PSCE achieved a better performance than the existing ones. Finally, the effectiveness of the PFEM and D3TC was demonstrated using ablation studies.

A Toolpath Generator Based on Signed Distance Fields and Clustering Algorithms for Optimized Additive Manufacturing

Additive manufacturing (AM) methods have been gaining momentum because they provide vast design and fabrication possibilities, increasing the accessibility of state-of-the-art hardware through recent developments in user-friendly computer-aided drawing/engineering/manufacturing (CAD/CAE/CAM) tools. However, in comparison to the conventional manufacturing methods, AM processes have some disadvantages, including the machining precision and fabrication process times. The first issue has been mostly resolved through the recent advances in manufacturing hardware, sensors, and controller systems. However, the latter has been widely investigated by researchers with different toolpath planning perspectives. As a contribution to these investigations, the present study proposes a toolpath planning method for AM, which aims to provide highly continuous yet distance-optimized solutions. The approach is based on the utilization of the signed distance field (SDF), clustering, and minimization of toolpath distances among cluster centroids. The method was tested on various geometries with simple closed curves to complex geometries with holes, which provides effective toolpaths, e.g., with relative distance reduction percentages up to 16.5% in comparison to conventional rectilinear infill patterns.

B - 57 Examining Symptom Patterns in Anomalous Health Incidents: a Spectral Co-Clustering Analysis Using the Post-Concussion Symptom Scale

Abstract Objective The ImPACT Post-Concussion Symptom Scale (PCSS) is widely used to document post-concussion symptoms. This study investigated the structural patterns of PCSS using spectral co-clustering on data from U.S. diplomats and their families who served in Cuba and China, who reported experiencing Anomalous Health Incidents (AHIs). Method Clinicians at the University of Pennsylvania evaluated 65 individuals with potential AHI exposure (Median age-group = 35–45 years; 73.84% in Cuba; 49.23% females). PCSS feature scores were discretized into 4-level symptom severity expressions: none, mild, moderate, and severe. The preprocessed data were fit with spectral co-clustering model to extract optimal number of biclusters, by measuring scree plot with principal component analysis and Calinski-Harabasz (CH) score. Final model was also evaluated by CH score and a random label t-test to compare the significance of the clustering results. Results Overall, spectral co-clustering yielded two biclusters with CH score of 20.29 and random label t-test of p < 0.001, suggesting reasonable clustering quality. Biclustering identified two sets of interpretable PCSS symptom characteristics. Cluster centroids X and Y (IDs retracted) were estimated using Manhattan distance, summarizing symptom characteristics of each cluster. More precisely, centroid X represented mild to moderate symptom profiles (e.g., headache, dizziness, drowsiness, and nervousness), while centroidY represented a relative absence of symptoms, except for mild nervousness and emotionality. Conclusion The spectral co-clustering of PCSS reveals patterns of elevated symptoms in one cluster and minimal symptom burden in the other, similar to outcomes observed after mild TBI. These patterns can guide future research directions and ultimately improve patient outcomes in various medical contexts.

Sectoral electricity micro-spatial load forecasting based on partitional clustering technique

<p>Load demand forecasting is crucial in energy supply planning due to economic progress and territorial expansion, where land utilization transforms dynamically. An accurate sectoral load prediction can preclude the loss of beneficial opportunities arising from excessive load demand or excessive investment at a low-growth juncture. However, the particular area in this sectoral approach is still relatively large, rendering it incapable of precisely projecting load at minor points (micro-spatial). This study has proposed a micro-spatial load prediction strategy that categorizes identified areas into smaller grids or districts. This procedure includes clustering similar sites together for improved accuracy. K-Means is one of the partitional clustering approaches, a clustering algorithm utilizing object-based centroid-based partitioning approaches. The algorithm determines a cluster's centroid or centre as the average point for the cluster. This technique is advantageous as it can process extensive data efficiently and is appropriate for circular data. This technique can divide the data into multiple partitions, ensuring that each object belongs to precisely one cluster. Subsequently, mathematical modelling is used to predict the load of each cluster, which can then be utilized to more accurately evaluate the positions and sizes of prospective substations, transmission, and distribution facilities.</p>

Energy efficient routing for improving lifetime in MWSN: A clustering approach

A Mobile wireless sensor network (MWSN) consists of mobile sensor nodes, which can be deployed in any specific environment, and due to its’ mobility it can perform with rapid topological transformations of a network. The sensor nodes having limited battery power are used to collect specific data and this raw data is sent to a static sink node of the network. Under such a scenario, to avoid frequent disconnections due to topological change in the network and can avail more reliable data transmissions in energy awareness perspective, an energy efficient routing protocol for MWSNs to improve its lifetime is proposed in this paper by utilizing a clustering approach. A MWSN with random number of sensor nodes are initially considered and then, clustering algorithm K-means is used to find a predefined number of clusters with their initial cluster heads (CHs) and the centroid location of these clusters is also determined. The role of these CHs is to elect our DDBLACH (distance to sink and cluster centroid with battery level aware cluster head) nodes from each cluster, by sending and receiving intra-cluster messages among other member sensor nodes. A DDBLACH node is determined by using three factors, such as minimum distance from cluster centroid location, minimum distance from sink and the maximum battery level of the node from each cluster. These DDBLACH nodes are used to collect data from intra-cluster sensors and thereafter, send those towards sink node for further processing using tree based hierarchical routes. Finally, an energy efficient routing technique for MWSNs is proposed for data transmission from these DDBLACH nodes of clusters to a sink of the network. Here simulation results indicate the supremacy of our proposed scheme over other existing methods in various aspects, such as improving the presence of alive nodes and average network lifetime.

Partial label learning via weighted centroid clustering disambiguation

Partial Label Learning (PLL) is a weakly supervised learning problem that induces a multi-class classifier from data with candidate labels, among which only one is the ground-truth label. The crucial challenge in PLL is to disambiguate the false-positive labels from the candidate labels. However, most existing PLL methods fail to simultaneously consider both the instance-level similarity and the class-level information during label disambiguation. In this paper, we propose a novel two-stage method based on weighted centroid clustering, which efficiently utilizes both the local similarity among instances and the per-class global information. In the first stage, we initialize the center of each class using label propagation based on the instance-level similarity, and obtain the disambiguated instances via weighted centroid clustering derived from the per-class global information. In the second stage, the disambiguated instances are used to train a multi-class classifier. Extensive experiments on both controlled UCI datasets and real-world datasets show the superiority of the proposed method in classification accuracy.

Empirical study of outlier impact in classification context

In the field of data mining, outlier detection is an important and challenging task. This paper focuses on studying the impacts of outliers on the performance of models learned from large-scale datasets while it is unknown whether a data point is outlier or not. In the proposed approach, a fuzzy c-means clustering algorithm is applied and outliers are defined as those that have greater membership values in the cluster and are located further away from the cluster centroid. Ideally, the sample with a higher membership value should be located closer to the cluster centroid. In this context, we calculate the weight of each sample using the AdaBoost algorithm, where a weight determines the representativeness of each sample within the data distribution. Additionally, in this study, the impact of weighted loss functions in different situations are discussed in detail. At last, our method is evaluated on 12 UCI datasets and the accuracy of our method is greater than 95% on some datasets, such as banknote 99.99%, biodeg 99.01%, optdigits 99.19%, and letters 97.37%. The experimental results show the efficiency and effectiveness of the proposed approach.

A momentum quantification method for tennis matches based on t-SNE and k means clustering

Momentum is a crucial factor influencing the outcomes of sports competitions. This study proposes a method for visualising and quantifying momentum using various modern analytical techniques. These techniques include feature engineering, dimensionality reduction with the t-SNE algorithm, and k-means clustering. By analysing data from the past five years of Wimbledon, the study extracted various performance features such as serve success rate and scoring rate. By calculating the similarities between features and conducting cluster analysis, the study reveals changes in momentum during matches. This research employs data visualisation and clustering results to not only display the conditions of matches under different momentum states but also quantifies momentum changes by calculating the Euclidean distance between data points and cluster centroids and standardising these distances using the Z-score method. This approach provides a visual perspective that reveals the dynamic changes that occur within matches and quantifies momentum. It demonstrates that these changes in momentum can indeed reflect the progression and outcomes of matches.

Dynamic Cluster Head Selection in WSN

A Wireless Sensor Network (WSN) comprises an ad-hoc network of nodes laden with sensors that are used to monitor a region mostly in the outdoors and often not easily accessible. Despite exceptions, several deployments of WSN continue to grapple with the limitation of finite energy derived through batteries. Thus, it is imperative that the energy of a WSN be conserved and its life prolonged. An important direction of work to this end is towards the transmission of data between nodes in a manner that minimum energy is expended. One approach to doing this is cluster-based routing, wherein nodes in a WSN are organised into clusters, and transmission of data from the node is through a representative node called a cluster-head. Forming optimal clusters and choosing an optimal cluster-head is an NP-Hard problem. Significant work is done towards devising mechanisms to form clusters and choosing cluster heads to reduce the transmission overhead to a minimum. In this article, an approach is proposed to create clusters and identify cluster heads that are near optimal. The approach involves two-stage clustering, with the clustering algorithm for each stage chosen through an exhaustive search. Furthermore, unlike existing approaches that choose a cluster-head on the basis of the residual energy of nodes, the proposed approach utilises three factors in addition to the residual energy, namely the distance of a node from the cluster centroid, the distance of a node from the final destination (base-station), and the connectivity of the node. The approach is shown to be effective and economical through extensive validation via simulations and through a real-world prototypical implementation.

Protein Retrieval via Integrative Molecular Ensembles (PRIME) through Extended Similarity Indices.

Molecular dynamics (MD) simulations are ideally suited to describe conformational ensembles of biomolecules such as proteins and nucleic acids. Microsecond-long simulations are now routine, facilitated by the emergence of graphical processing units. Clustering, which groups objects based on structural similarity, is typically used to process ensembles, leading to different states, their populations, and the identification of representative structures. A popular pipeline combines hierarchical clustering for clustering and selecting the cluster centroid as representative of the cluster. Here, we propose to improve on this approach, by developing a module-Protein Retrieval via Integrative Molecular Ensembles (PRIME), that consists of tools to improve the prediction of the representative in the most populated cluster using extended continuous similarity. PRIME is integrated with our Molecular Dynamics Analysis with N-ary Clustering Ensembles (MDANCE) package and can be used as a postprocessing tool for arbitrary clustering algorithms, compatible with several MD suites. PRIME predictions produced structures that when aligned to the experimental structure were better superposed (lower RMSD). A further benefit of PRIME is its linear scaling─rather than the traditional O(N2) traditionally associated with comparisons of elements in a set.

O-082 From words to wellness: a retrospective study leveraging large language models on conversational data to uncover patient needs

Abstract Study question Can large language models be used to understand patient needs from conversational data? Summary answer Large language models can provide significant assistance for identifying and summarizing patients' queries. What is known already Traditionally, clinics have relied on techniques such as patient surveys, reviews and complaints procedures in order to understand shortcomings in the patient experience. As many clinics adopt digital communication platforms with patients, they have collected a wealth of conversational data that may shed light on the patient experience. However, the volume of data in clinic chat apps is often so great that analyzing them becomes challenging. Recently, advances in large language models (LLMs) have enabled the automated analysis of text at almost human level performance. This study therefore investigates whether LLMs can be used to extract insights from conversational data. Study design, size, duration This study is a retrospective analysis of 132,596 messages sent by patients to fertility advisors representing 40,853 questions asked. These conversations took place on a single centre’s patient communication app from 01/01/2021 to 09/06/2023. All patient types at all treatment stages were included. A private instance of the open-source Mistral-7B-Instruct-v0.1 LLM running on a single NVIDIA Titan X GPU was used for text analysis. Participants/materials, setting, methods Conversations were broken down into sentences and then categorized as either questions or non-questions by the LLM. Next the LLM categorized the individual questions, returning a category, a subcategory and question summary. These summaries were then embedded and clustered using the K-means algorithm (with k chosen by the elbow method). The LLM was then used to summarize the content of each cluster as five questions. Thematic analysis was then conducted by a patient experience expert. Main results and the role of chance In the initial phase of the study, the LLM, classified 34,222 questions into 6,177 categories and 13,533 subcategories. These were subsequently consolidated into 145 distinct clusters. Each cluster, on average, comprised 215±108 (M+SD) inquiries (excluding a notably larger outlier cluster that seemed to functionally encompass 3,300 inquiries that didn’t fit neatly into any other cluster). Examination of cluster centroids identified seven predominant themes: legal/financial (N = 1,218), general fertility advice (N = 1,430), patient administration (N = 8,734), medical tests (N = 4,820), medical procedures (N = 1,575), appointment scheduling (N = 4,320), and specific treatment/medication information (N = 9,125).Of the 145 clusters, 64 clusters comprising 12,313 inquiries were identified by the expert to highlight points for improvement in the patient experience. These broadly encompassed operational bottlenecks/weak points, issues with patient guidance and issues with communication. For instance, “Are there any late afternoon or evening appointment options available?” or “How long do eggs remain viable after they have been frozen?”. Overall, our use of LLMs enabled the analysis of a large number of queries that would previously have proven prohibitively expensive, time-consuming and labor-intensive. Limitations, reasons for caution Splitting conversations into sentences meant that context could not be taken into account and multi-message questions were hard to identify. Additionally, the interpretation of messages by LLMs and humans may not be aligned. Finally, the technical expertise required to execute this style of analysis may prove a barrier for clinics. Wider implications of the findings We demonstrate that LLMs can be used to draw insights from the wealth of digital communications held by modern IVF clinics. LLMs may thus enable clinics to collect data on the patient experience in a faster, more reliable way than traditional approaches such as patient surveys and complaints. Trial registration number Not applicable

Jacobian-scaled K-means clustering for physics-informed segmentation of reacting flows

This work introduces Jacobian-scaled K-means (JSK-means) clustering, which is a physics-informed clustering strategy centered on the K-means framework. The method allows for the injection of underlying physical knowledge into the clustering procedure through a distance function modification: instead of leveraging conventional Euclidean distance vectors, the JSK-means procedure operates on distance vectors scaled by matrices obtained from dynamical system Jacobians evaluated at the cluster centroids. The goal of this work is to show how the JSK-means algorithm – without modifying the input dataset – produces clusters that capture regions of dynamical similarity, in that the clusters are redistributed towards high-sensitivity regions in phase space and are described by similarity in the source terms of samples instead of the samples themselves. The algorithm is demonstrated on a complex reacting flow simulation dataset (a channel detonation configuration), where the dynamics in the thermochemical composition space are known through the highly nonlinear and stiff Arrhenius-based chemical source terms. Interpretations of cluster partitions in both physical space and composition space reveal how JSK-means shifts clusters produced by standard K-means towards regions of high chemical sensitivity (e.g., towards regions of peak heat release rate near the detonation reaction zone). The findings presented here illustrate the benefits of utilizing Jacobian-scaled distances in clustering techniques, and the JSK-means method in particular displays promising potential for improving former partition-based modeling strategies in reacting flow (and other multi-physics) applications.

Accurate Registration of Cross-Modality Geometry via Consistent Clustering.

The registration of unitary-modality geometric data has been successfully explored over past decades. However, existing approaches typically struggle to handle cross-modality data due to the intrinsic difference between different models. To address this problem, in this article, we formulate the cross-modality registration problem as a consistent clustering process. First, we study the structure similarity between different modalities based on an adaptive fuzzy shape clustering, from which a coarse alignment is successfully operated. Then, we optimize the result using fuzzy clustering consistently, in which the source and target models are formulated as clustering memberships and centroids, respectively. This optimization casts new insight into point set registration, and substantially improves the robustness against outliers. Additionally, we investigate the effect of fuzzier in fuzzy clustering on the cross-modality registration problem, from which we theoretically prove that the classical Iterative Closest Point (ICP) algorithm is a special case of our newly defined objective function. Comprehensive experiments and analysis are conducted on both synthetic and real-world cross-modality datasets. Qualitative and quantitative results demonstrate that our method outperforms state-of-the-art approaches with higher accuracy and robustness. Our code is publicly available at https://github.com/zikai1/CrossModReg.