Train Random Forest Classifiers Research Articles

A fast H.266/QTMT intra coding scheme based on predictions of learned models

ABSTRACT H.266/VVC adopts a QuadTree-plus-MultiType tree (QTMT) coding-unit (CU) split structure to improve efficiency at the cost of high time complexity. Speeding up VVC coding while minimizing quality degradation is critical for practical applications. We propose predicting the coding depth and split type of an optimally coded 32 × 32 CU (CU32 × 32) to perform only a subset of exhaustive rate-distortion optimization (RDO) operations: (1) To predict the depth of an optimally coded CU32 × 32, we train a convolutional neural network (CNNdepth). CNNdepth outputs a label specifying a depth range subset by which the controller can execute EarlySkip or EarlyTerminate to reduce time complexity. (2) To predict the split type, we train random forest classifiers (RFCtype). The corresponding RDO operations can be omitted if the RFCtype classifies one CU32 × 32 as not of a specified split type. Experiments show that CNNdepth and RFCtype work seamlessly, reducing execution time by up to 69% and 39.16% on average, with a 0.7% increase in BDBR compared to the default VTM-7.0. Additionally, the proposed method yields the highest balanced time reduction rate of 61.5%.

Sampling and active learning methods for network reliability estimation using K-terminal spanning tree

Network reliability analysis remains a challenge due to the increasing size and complexity of networks. This paper presents a novel sampling method and an active learning method for efficient and accurate network reliability estimation under node failure and edge failure scenarios. The proposed sampling method adopts Monte Carlo technique to sample component lifetimes and the K-terminal spanning tree algorithm to accelerate structure function computation. Unlike existing methods that compute only one structure function value per sample, our method generates multiple component state vectors and corresponding structure function values from each sample. Network reliability is estimated based on survival signatures derived from these values. A transformation technique extends this method to handle both node failure and edge failure. To enhance efficiency of proposed sampling method and achieve adaptability to network topology changes, we introduce an active learning method utilizing a random forest (RF) classifier. This classifier directly predicts structure function values, integrates network behaviors across diverse topologies, and undergoes iterative refinement to enhance predictive accuracy. Importantly, the trained RF classifier can directly predict reliability for variant networks, a capability beyond the sampling method alone. Through investigating several network examples and two practical applications, the effectiveness of both proposed methods is demonstrated.

Unlocking the mechanism of action: a cost-effective flow cytometry approach for accelerating antimicrobial drug development.

Antimicrobial resistance is one of the greatest challenges to global health. While the development of new antimicrobials can combat resistance, low profitability reduces the number of new compounds brought to market. Elucidating the mechanism of action is crucial for developing new antimicrobials. This can become expensive as there are no universally applicable pipelines. Phenotypic heterogeneity of microbial populations resulting from antimicrobial treatment can be captured through flow cytometric fingerprinting. Since antimicrobials are classified into limited groups, the mechanism of action of known compounds can be used for predictive modeling. We demonstrate a cost-effective flow cytometry approach for determining the mechanism of action of new compounds. Cultures of Actinomyces viscosus and Fusobacterium nucleatum were treated with different antimicrobials and measured by flow cytometry. A Gaussian mixture mask was applied over the data to construct phenotypic fingerprints. Fingerprints were used to assess statistical differences between mechanism of action groups and to train random forest classifiers. Classifiers were then used to predict the mechanism of action of cephalothin. Statistical differences were found among the different mechanisms of action groups. Pairwise comparison showed statistical differences for 35 out of 45 pairs for A. viscosus and for 32 out of 45 pairs for F. nucleatum after 3.5 h of treatment. The best-performing random forest classifier yielded a Matthews correlation coefficient of 0.92 and the mechanism of action of cephalothin could be successfully predicted. These findings suggest that flow cytometry can be a cheap and fast alternative for determining the mechanism of action of new antimicrobials.IMPORTANCEIn the context of the emerging threat of antimicrobial resistance, the development of novel antimicrobials is a commonly employed strategy to combat resistance. Elucidating the mechanism of action of novel compounds is crucial in this development but can become expensive, as no universally applicable pipelines currently exist. We present a novel flow cytometry-based approach capable of determining the mechanism of action swiftly and cost-effectively. The workflow aims to accelerate drug discovery and could help facilitate a more targeted approach for antimicrobial treatment of patients.

Machine learning and structure formation in modified gravity

ABSTRACT In general relativity, approximations based on the spherical collapse model such as Press–Schechter theory and its extensions are able to predict the number of objects of a certain mass in a given volume. In this paper, we use a machine learning algorithm to test whether such approximations hold in screened modified gravity theories. To this end, we train random forest classifiers on data from N-body simulations to study the formation of structures in lambda cold dark matter (ΛCDM) as well as screened modified gravity theories, in particular f(R) and nDGP gravity. The models are taught to distinguish structure membership in the final conditions from spherical aggregations of density field behaviour in the initial conditions. We examine the differences between machine learning models that have learned structure formation from each gravity, as well as the model that has learned from ΛCDM. We also test the generalizability of the ΛCDM model on data from f(R) and nDGP gravities of varying strengths, and therefore the generalizability of extended Press–Schechter spherical collapse to these types of modified gravity.

Key Signatures of Prominence Materials and Category of Cold Materials Identified by Random Forest Classifier

The origin of cold materials identified by different criteria is unclear. They are highly suspected to be erupted prominences. However, some cold materials defined by charge depletion exist in both quiet solar wind and interplanetary coronal mass ejections (ICMEs). Recently, solar observations show failed prominence eruption in coronal mass ejections (CMEs) that the prominence sometimes did not propagate into interplanetary space. This work uses Random Forest Classifier (RFC), which is an interpretable supervised machine-learning algorithm to study the distinct signatures of prominence cold materials (PCs) compared to quiet solar wind (QSW) and ICMEs excluding cold materials (ICMEEs). Twelve physical features measured by ACE at 1 au and the monthly averaged sunspot number are used in this study. The measurements from ACE are proton moments, magnetic field component B z , He/H, He/O, Fe/O, mean charge of oxygen and carbon, C6+/C5, C6+/C4+, and O7+/O6+. According to the returned weights from RFC that are checked by support vector machine classifier, the most important in situ signatures of PCs are obtained. Next, the trained RFC is used to check the category of the cold materials not related to CME observations. The results show that most segments of the cold materials are from prominences, but four of them are possibly from ICMEEs; another one segment is possibly from QSW. The most distinct signatures of PCs are lower (C6+/C5+)/(O7+/O6+), proton temperature, and He/O. Considering the obvious overlaps on key physical features between QSW, ICMEEs, and PCs, the multifeature classifier shows an advantage in identifying them than solid criteria.

Immune cell type signature discovery and random forest classification for analysis of single cell gene expression datasets.

Robust immune cell gene expression signatures are central to the analysis of single cell studies. Nearly all known sets of immune cell signatures have been derived by making use of only single gene expression datasets. Utilizing the power of multiple integrated datasets could lead to high-quality immune cell signatures which could be used as superior inputs to machine learning-based cell type classification approaches. We established a novel workflow for the discovery of immune cell type signatures based primarily on gene-versus-gene expression similarity. It leverages multiple datasets, here seven single cell expression datasets from six different cancer types and resulted in eleven immune cell type-specific gene expression signatures. We used these to train random forest classifiers for immune cell type assignment for single-cell RNA-seq datasets. We obtained similar or better prediction results compared to commonly used methods for cell type assignment in independent benchmarking datasets. Our gene signature set yields higher prediction scores than other published immune cell type gene sets in random forest-based cell type classification. We further demonstrate how our approach helps to avoid bias in downstream statistical analyses by re-analysis of a published IFN stimulation experiment. We demonstrated the quality of our immune cell signatures and their strong performance in a random forest-based cell typing approach. We argue that classifying cells based on our comparably slim sets of genes accompanied by a random forest-based approach not only matches or outperforms widely used published approaches. It also facilitates unbiased downstream statistical analyses of differential gene expression between cell types for significantly more genes compared to previous cell classification algorithms.

A data driven fault detection approach with an ensemble classifier based smart meter in modern distribution system

The operation of the distribution grids is constantly being threatened by occurrence of faults. These faults are unpredictable and dangerous. Moreover, the introduction of distributed generations (DGs) in modern distribution networks has made the detection of these faults more complicated. Thus, to maintain the continuity of the power supply, a fast and accurate fault detection strategy is required to isolate the fault. In view of this, the paper presents a data driven fault detection approach with an ensemble classifier based smart meter in modern distribution system. To achieve this, a random forest (RF) based fault detection algorithm is programmed within the smart meter. At first, the magnitude of maximum angular difference between positive and zero sequence component of the current at the DG bus is computed. The computed value is then fed to a trained RF classifier for detecting fault conditions. The greatest advantage of the proposed algorithm is the non-requisite of additional hardware or software kit for fault detection. The algorithm undergoes several test conditions to showcase its effectiveness. The algorithm depicts a high accuracy of 98.95%. The results signify that the algorithm is robust, reliable and accurate with an additional benefit of augmenting situational awareness.

Malware Detection System Using Mathematics of Random Forest Classifier

Most cyberattacks including data breaches, identity theft, fraud, and other issues, are known to be caused by malware. Some of the malware attacks are categories as adware, spyware, virus, worm, trojan, rootkit, backdoor, ransomware and command and control (C&C) bot, based on its purpose and behaviour. Malware detectors still utilise signature-based approaches to detect malicious software, which can only detect known malware. Attacks by malware pose a serious threat to people's and organizations' cybersecurity globally. These attacks are occurring more frequently and more frequently lately. Over eight billion malware attacks occurred in 2020, up 4% over the previous year, according to a Symantec report. It is crucial that computer users safeguard their computers with a malware detector like an antivirus, anti-spyware, etc. When creating a machine learning model to differentiate between malicious and benign files, it might be challenging to use domain-level expertise to extract the necessary attributes. This research aims to create a malware detector that uses a trained random forest classifier model to find malware and stop zero-day assaults. A dataset (including both harmful and benign software PE header information) was obtained from virusshare.com and used to train the random forest classifier in order to create this malware detector. The Random Forest Classifier generate greater accuracy when compared with other machine learning classifiers, such as KNN (K-Nearest Neighbors), Decision Tree, Logistic Regression etc., the random forest classifier gives a better accuracy of 99.4%. The Classifier model used here will be a better option to use in order to efficiently and effectively detect malware, it shows that the methodology can be utilized as the basis for an operational system for detecting an unknown malicious executable.

Mapping the distribution and extent of India's semi‐arid open natural ecosystems

AbstractAimNon‐forested open natural ecosystems (ONEs) support high biodiversity as well as livelihoods of millions of pastoralists, but are highly threatened due to land conversion. To improve the recognition of their value, we generated a high‐resolution (30 m) map of ONEs in the low elevation arid and semi‐arid regions of India.LocationIndian subcontinent.TaxonOpen natural ecosystems, consisting of a range of non‐forested habitats.MethodsWe used public datasets in the Google Earth Engine catalogue to identify and mask out pixels with non‐ONE land‐cover types, and biophysical attributes outside the environmental envelopes of ONEs. We used 181,812 ONE and 116,447 non‐ONE training points along with the input composite image to train random forest classifiers to estimate the probability of a pixel being an ONE. We developed a series of local classifiers for each of eight regions of interest and created a mask of pixels with ≥50% probability of being an ONE.ResultsThe overall classification accuracy for this map was 85%, with a Kappa statistic of 0.70. We find that ONEs cover approximately 320,000 km2 (10%) of India's land surface, but <5% of ONEs are covered under the existing protected area network of India. ONE areas vary greatly by region, from 6 km2 in New Delhi (0.4% area) to 115,069 km2 in Rajasthan (33% area).Main ConclusionsA large proportion of India's geographical area represents an ecosystem that is un‐recognized and under‐protected. This open data source can be used by policy makers and planners to exclude these habitats when considering the location of renewable energy, tree planting for carbon sequestration and other development projects that negatively impact ONEs. We encourage further refinement of ONE maps in India, incorporating vegetation and floristic characteristics, as well as further consideration of these vulnerable neglected ecosystems in other regions of the world.

Detecting the Parkinson’s Disease Through the Simultaneous Analysis of Data From Wearable Sensors and Video

Machine Learning (ML) algorithms is an emerging tool helping automatically solve detection and classification tasks in myriads of applications. ML is extensively used in medical applications and, in particular, for detecting the Parkinson’s Disease (PD). However, the inference is typically made relying on a single data source. This work explores the results of combining more than one data source type for the diagnosis of PD. Data from the Commercial Off-the-Shelf (COTS) wearable sensors (accelerometer, gyroscope and magnetometer), along with video recordings from 83 patients completing a series of 15 tasks was analyzed with the use of ML methods. Statistical and frequency features were extracted and used to train Random Forest and XGBoost Classifiers. We investigate two use cases on classifying (i) healthy individuals and individuals with the PD, and (ii) PD and essential tremor. The experiment showed that using both the data from wearable sensors and video provided the increase of f1 score up to 18% for differentiating between healthy and PD classes and 21% for differentiating PD and essential tremor classes. At the same time, usage of COTS and ML opens wide vista for patient driven data acquisition and healthcare.

EDNAssay: A machine learning tool that accurately predicts qPCR cross-amplification.

Environmental DNA (eDNA) sampling is a highly sensitive and cost-effective technique for wildlife monitoring, notably through the use of qPCR assays. However, it can be difficult to ensure assay specificity when many closely related species co-occur. In theory, specificity may be assessed in silico by determining whether assay oligonucleotides have enough base-pair mismatches with nontarget sequences to preclude amplification. However, the mismatch qualities required are poorly understood, making in silico assessments difficult and often necessitating extensive in vitro testing-typically the greatest bottleneck in assay development. Increasing the accuracy of in silico assessments would therefore streamline the assay development process. In this study, we paired 10 qPCR assays with 82 synthetic gene fragments for 530 specificity tests using SYBR Green intercalating dye (n=262) and TaqMan hydrolysis probes (n=268). Test results were used to train random forest classifiers to predict amplification. The primer-only model (SYBR Green results) and full-assay model (TaqMan probe-based results) were 99.6% and 100% accurate, respectively, in cross-validation. We further assessed model performance using six independent assays not used in model training. In these tests the primer-only model was 92.4% accurate (n=119) and the full-assay model was 96.5% accurate (n=144). The high performance achieved by these models makes it possible for eDNA practitioners to more quickly and confidently develop assays specific to the intended target. Practitioners can access the full-assay model online via eDNAssay (https://NationalGenomicsCenter.shinyapps.io/eDNAssay), a user-friendly tool for predicting qPCR cross-amplification.

Automated detection and classification of tumor histotypes on dynamic PET imaging data through machine-learning driven voxel classification

2-deoxy-2-fluorine-(18F)fluoro-d-glucose Positron Emission Tomography/Computed Tomography (18F-FDG-PET/CT) is widely used in oncology mainly for diagnosis and staging of various cancer types, including lung cancer, which is the most common cancer worldwide. Since histopathologic subtypes of lung cancer show different degree of 18F-FDG uptake, to date there are some diagnostic limits and uncertainties, hindering an 18F-FDG-PET-driven classification of histologic subtypes of lung cancers. On the other hand, since activated macrophages, neutrophils, fibroblasts and granulation tissues also show an increased 18F-FDG activity, infectious and/or inflammatory processes and post-surgical and post-radiation changes may cause false-positive results, especially for lymph-nodes assessment. Here we propose a model-free, machine-learning based algorithm for the automated classification of adenocarcinoma, the most common type of lung cancer, and other types of tumors. Input for the algorithm are dynamic acquisitions of PET data (dPET), providing for a spatially and temporally resolved characterization of the uptake kinetic. The algorithm consists in a trained Random Forest classifier which, relying contextually on several spatial and temporal features of 18F-FDG uptake, generates as an outcome probability maps allowing to distinguish adenocarcinoma from other lung histotype and to identify metastatic lymph-nodes, ultimately increasing the specificity of the technique. Its performance, evaluated on a dPET dataset of 19 patients affected by primary lung cancer, provides a probability 0.943 ± 0.090 for the detection of adenocarcinoma. The use of this algorithm will guarantee an automatic and more accurate localization and discrimination of tumors, also providing a powerful tool for detecting at which extent tumor has spread beyond a primary tumor into lymphatic system.

Investigation of Polarimetric Decomposition for Arctic Summer Sea Ice Classification Using Gaofen-3 Fully Polarimetric SAR Data

The aim of this article was to investigate the potential of polarimetric decomposition of Chinese Gaofen-3 (GF-3) C-band fully polarimetric synthetic aperture radar (PolSAR) data for Arctic sea ice classification during summer season. Five different polarimetric decomposition approaches, including the Cloude-Pottier decomposition (Cloude), the Freeman three-component decomposition (Freeman3), the Freeman three-component decomposition using the extended Bragg model (Freeman3X), the Yamaguchi three-component decomposition (Yamaguchi3), and the nonnegative eigenvalue decomposition (NNED) were analyzed using 35 scenes of GF-3 PolSAR data collected over the Fram Strait, Arctic from June 14–18, 2017. Polarimetric features extracted from these five methods were evaluated and utilized to train random forest classifiers to classify open water (calm water and rough water) and sea ice types (melted ice, unmelted ice, and deformed ice). The results show that NNED could ensure physically valid decomposed powers while the other three model-based decompositions had negative values. In terms of sea ice classification, NNED had the highest feature importance scores and achieved an overall accuracy and Kappa coefficient of about 86.18% and 0.82, respectively. Inclusion of radar incidence angle as a feature in the classifier could slightly improve the classification accuracy by about 3%. The influence of incidence angle on sea ice classification accuracy was also investigated and it was found that high incidence angles (39°–46°) were superior to low incidence angles (21°–27°) due to the overall higher accuracies.

PEPRF: Identification of Essential Proteins by Integrating Topological Features of PPI Network and Sequence-based Features via Random Forest

Background: Essential proteins play an important role in the process of life, which can be identified by experimental methods and computational approaches. Experimental approaches to identify essential proteins are of high accuracy but with the limitation of time and resource-consuming. Objective: Herein, we present a computational model (PEPRF) to identify essential proteins based on machine learning. Methods: Different features of proteins were extracted. Topological features of Protein-Protein Interaction (PPI) network-based are extracted. Based on the protein sequence, graph theory-based features, information- based features, composition and physichemical features, etc., were extracted. Finally, 282 features are constructed. In order to select the features that contributed most to the identification, ReliefF- based feature selection method was adopted to measure the weights of these features. Results: As a result, 212 features were curated to train random forest classifiers. Finally, PEPRF get the AUC of 0.71 and an accuracy of 0.742. Conclusion: Our results show that PEPRF may be applied as an efficient tool to identify essential proteins.

Machine Learning Assisted Approach for Finding Novel High Activity Agonists of Human Ectopic Olfactory Receptors

Olfactory receptors (ORs) constitute the largest superfamily of G protein-coupled receptors (GPCRs). ORs are involved in sensing odorants as well as in other ectopic roles in non-nasal tissues. Matching of an enormous number of the olfactory stimulation repertoire to its counterpart OR through machine learning (ML) will enable understanding of olfactory system, receptor characterization, and exploitation of their therapeutic potential. In the current study, we have selected two broadly tuned ectopic human OR proteins, OR1A1 and OR2W1, for expanding their known chemical space by using molecular descriptors. We present a scheme for selecting the optimal features required to train an ML-based model, based on which we selected the random forest (RF) as the best performer. High activity agonist prediction involved screening five databases comprising ~23 M compounds, using the trained RF classifier. To evaluate the effectiveness of the machine learning based virtual screening and check receptor binding site compatibility, we used docking of the top target ligands to carefully develop receptor model structures. Finally, experimental validation of selected compounds with significant docking scores through in vitro assays revealed two high activity novel agonists for OR1A1 and one for OR2W1.

Deep learning-based model for diagnosing Alzheimer's disease and tauopathies.

AimsThis study aimed to develop a deep learning‐based model for differentiating tauopathies, including Alzheimer's disease (AD), progressive supranuclear palsy (PSP), corticobasal degeneration (CBD) and Pick's disease (PiD), based on tau‐immunostained digital slide images.MethodsWe trained the YOLOv3 object detection algorithm to detect five tau lesion types: neuronal inclusions, neuritic plaques, tufted astrocytes, astrocytic plaques and coiled bodies. We used 2522 digital slide images of CP13‐immunostained slides of the motor cortex from 10 cases each of AD, PSP and CBD for training. Data augmentation was performed to increase the size of the training dataset. We next constructed random forest classifiers using the quantitative burdens of each tau lesion from motor cortex, caudate nucleus and superior frontal gyrus, ascertained from the object detection model. We split 120 cases (32 AD, 36 PSP, 31 CBD and 21 PiD) into training (90 cases) and test (30 cases) sets to train random forest classifiers.ResultsThe resultant random forest classifier achieved an average test score of 0.97, indicating that 29 out of 30 cases were correctly diagnosed. A validation study using hold‐out datasets of CP13‐ and AT8‐stained slides from 50 cases (10 AD, 17 PSP, 13 CBD and 10 PiD) showed >92% (without data augmentation) and >95% (with data augmentation) diagnostic accuracy in both CP13‐ and AT8‐stained slides.ConclusionOur diagnostic model trained with CP13 also works for AT8; therefore, our diagnostic tool can be potentially used by other investigators and may assist medical decision‐making in neuropathological diagnoses of tauopathies.

Machine learning identifies stroke features between species.

Identification and localization of ischemic stroke (IS) lesions is routinely performed to confirm diagnosis, assess stroke severity, predict disability and plan rehabilitation strategies using magnetic resonance imaging (MRI). In basic research, stroke lesion segmentation is necessary to study complex peri-infarction tissue changes. Moreover, final stroke volume is a critical outcome evaluated in clinical and preclinical experiments to determine therapy or intervention success. Manual segmentations are performed but they require a specialized skill set, are prone to inter-observer variation, are not entirely objective and are often not supported by histology. The task is even more challenging when dealing with large multi-center datasets, multiple experimenters or large animal cohorts. On the other hand, current automatized segmentation approaches often lack histological validation, are not entirely user independent, are often based on single parameters, or in the case of complex machine learning methods, require vast training datasets and are prone to a lack of model interpretation.Methods: We induced IS using the middle cerebral artery occlusion model on two rat cohorts. We acquired apparent diffusion coefficient (ADC) and T2-weighted (T2W) images at 24 h and 1-week after IS induction. Subsets of the animals at 24 h and 1-week post IS were evaluated using histology and immunohistochemistry. Using a Gaussian mixture model, we segmented voxel-wise interactions between ADC and T2W parameters at 24 h using one of the rat cohorts. We then used these segmentation results to train a random forest classifier, which we applied to the second rat cohort. The algorithms' stroke segmentations were compared to manual stroke delineations, T2W and ADC thresholding methods and the final stroke segmentation at 1-week. Volume correlations to histology were also performed for every segmentation method. Metrics of success were calculated with respect to the final stroke volume. Finally, the trained random forest classifier was tested on a human dataset with a similar temporal stroke on-set. Manual segmentations, ADC and T2W thresholds were again used to evaluate and perform comparisons with the proposed algorithms' output.Results: In preclinical rat data our framework significantly outperformed commonly applied automatized thresholding approaches and segmented stroke regions similarly to manual delineation. The framework predicted the localization of final stroke regions in 1-week post-stroke MRI with a median Dice similarity coefficient of 0.86, Matthew's correlation coefficient of 0.80 and false positive rate of 0.04. The predicted stroke volumes also strongly correlated with final histological stroke regions (Pearson correlation = 0.88, P < 0.0001). Lastly, the stroke region characteristics identified by our framework in rats also identified stroke lesions in human brains, largely outperforming thresholding approaches in stroke volume prediction (P<0.01).Conclusion: Our findings reveal that the segmentation produced by our proposed framework using 24 h MRI rat data strongly correlated with the final stroke volume, denoting a predictive effect. In addition, we show for the first time that the stroke imaging features can be directly translated between species, allowing identification of acute stroke in humans using the model trained on animal data. This discovery reduces the gap between the clinical and preclinical fields, unveiling a novel approach to directly co-analyze clinical and preclinical data. Such methods can provide further biological insights into human stroke and highlight the differences between species in order to help improve the experimental setups and animal models of the disease.

In silico Comparison of Left Atrial Ablation Techniques That Target the Anatomical, Structural, and Electrical Substrates of Atrial Fibrillation.

Catheter ablation therapy for persistent atrial fibrillation (AF) typically includes pulmonary vein isolation (PVI) and may include additional ablation lesions that target patient-specific anatomical, electrical, or structural features. Clinical centers employ different ablation strategies, which use imaging data together with electroanatomic mapping data, depending on data availability. The aim of this study was to compare ablation techniques across a virtual cohort of AF patients. We constructed 20 paroxysmal and 30 persistent AF patient-specific left atrial (LA) bilayer models incorporating fibrotic remodeling from late-gadolinium enhancement (LGE) MRI scans. AF was simulated and post-processed using phase mapping to determine electrical driver locations over 15 s. Six different ablation approaches were tested: (i) PVI alone, modeled as wide-area encirclement of the pulmonary veins; PVI together with: (ii) roof and inferior lines to model posterior wall box isolation; (iii) isolating the largest fibrotic area (identified by LGE-MRI); (iv) isolating all fibrotic areas; (v) isolating the largest driver hotspot region [identified as high simulated phase singularity (PS) density]; and (vi) isolating all driver hotspot regions. Ablation efficacy was assessed to predict optimal ablation therapies for individual patients. We subsequently trained a random forest classifier to predict ablation response using (a) imaging metrics alone, (b) imaging and electrical metrics, or (c) imaging, electrical, and ablation lesion metrics. The optimal ablation approach resulting in termination, or if not possible atrial tachycardia (AT), varied among the virtual patient cohort: (i) 20% PVI alone, (ii) 6% box ablation, (iii) 2% largest fibrosis area, (iv) 4% all fibrosis areas, (v) 2% largest driver hotspot, and (vi) 46% all driver hotspots. Around 20% of cases remained in AF for all ablation strategies. The addition of patient-specific and ablation pattern specific lesion metrics to the trained random forest classifier improved predictive capability from an accuracy of 0.73 to 0.83. The trained classifier results demonstrate that the surface areas of pre-ablation driver regions and of fibrotic tissue not isolated by the proposed ablation strategy are both important for predicting ablation outcome. Overall, our study demonstrates the need to select the optimal ablation strategy for each patient. It suggests that both patient-specific fibrosis properties and driver locations are important for planning ablation approaches, and the distribution of lesions is important for predicting an acute response.

HYBRID ACQUISITION OF HIGH QUALITY TRAINING DATA FOR SEMANTIC SEGMENTATION OF 3D POINT CLOUDS USING CROWD-BASED ACTIVE LEARNING

Abstract. Automated semantic interpretation of 3D point clouds is crucial for many tasks in the domain of geospatial data analysis. For this purpose, labeled training data is required, which has often to be provided manually by experts. One approach to minimize effort in terms of costs of human interaction is Active Learning (AL). The aim is to process only the subset of an unlabeled dataset that is particularly helpful with respect to class separation. Here a machine identifies informative instances which are then labeled by humans, thereby increasing the performance of the machine. In order to completely avoid involvement of an expert, this time-consuming annotation can be resolved via crowdsourcing. Therefore, we propose an approach combining AL with paid crowdsourcing. Although incorporating human interaction, our method can run fully automatically, so that only an unlabeled dataset and a fixed financial budget for the payment of the crowdworkers need to be provided. We conduct multiple iteration steps of the AL process on the ISPRS Vaihingen 3D Semantic Labeling benchmark dataset (V3D) and especially evaluate the performance of the crowd when labeling 3D points. We prove our concept by using labels derived from our crowd-based AL method for classifying the test dataset. The analysis outlines that by labeling only 0:4% of the training dataset by the crowd and spending less than 145 $, both our trained Random Forest and sparse 3D CNN classifier differ in Overall Accuracy by less than 3 percentage points compared to the same classifiers trained on the complete V3D training set.

Comparison of machine learning techniques to predict unplanned readmission following total shoulder arthroplasty

Machine learning (ML) techniques have been shown to successfully predict postoperative complications for high-volume orthopedic procedures such as hip and knee arthroplasty and to stratify patients for risk-adjusted bundled payments. The latter has not been done for more heterogeneous, lower-volume procedures such as total shoulder arthroplasty (TSA) with equally limited discussion around strategies to optimize the predictive ability of ML algorithms. The purpose of this study was to (1) assess which of 5 ML algorithms best predicts 30-day readmission, (2) test select ML strategies to optimize the algorithms, and (3) report on which patient variables contribute most to risk prediction in TSA across algorithms. We identified 9043 patients in the American College of Surgeons National Surgical Quality Improvement Databasewho underwent primary TSA between 2011 and 2015. Predictors included demographics, comorbidities, laboratory data, and intraoperative variables. The outcome of interest was 30-day unplanned readmission. Five ML algorithms-support-vector machine (SVM), logistic regression, random forest (RF), an adaptive boosting algorithm, and neural network-were trained on the derivation cohort (2011-2014 TSA patients) to predict 30-day unplanned readmission rates. After training, weights for each respective model were fixed and the classifiers were evaluated on the 2015 TSA cohort to simulate a prospective evaluation. C-statistic and f1 scores were used to assess the performance of each classifier. After evaluation, features were removed independently to assess which features most affected classifier performance. The derivation and validation cohorts comprised 5857 and 3186 primary TSA patients, respectively, with similar demographics, comorbidities, and 30-day unplanned readmission rates (2.9% vs. 2.7%). Of the ML algorithms, SVM performed the worst with a c-statistic of 0.54 and an f1-score of 0.07, whereas the random-forest classifier performed the best with the highest c-statistic of 0.74 andan f1-score of 0.18. In addition, SVM was most sensitive to loss of single features, whereas the performance of RF did not dramatically decrease after loss of single features. Within the trained RF classifier, 5 variables achieved weights >0.5 in descending order: high bilirubin (>1.9 mg/dL), age >65, race, chronic obstructive pulmonary disease, and American Society of Anesthesiologists' scores ≥3. In our validation cohort, we observed a 2.7% readmission rate. From this cohort, using the RF classifier we were then able to identify 436 high-risk patients with a predicted risk score >0.6, of whom 36 were readmitted (readmission rate of 8.2%). Predictive analytics algorithms can achieve acceptable prediction of unplanned readmission for TSA with the RF classifier outperforming other common algorithms.