Memory-efficient implementation of gravity inversion

This focus issue of Physiological Measurement follows the 38th Annual International Computing in Cardiology (CinC) Conference, hosted in Hangzhou, China in September 2011 by Zhejiang University. Each year, the NIH-sponsored PhysioNet resource (http://physionet.org/) runs an open competition lasting several months, aimed at encouraging the development of solutions to an unsolved or poorly solved problem in biomedicine, in most cases making use of relevant clinical and experimental data provided freely by PhysioNet. Participants in these annual challenges discuss their diverse approaches to the Challenge problems during dedicated scientific sessions at CinC. The topics of these PhysioNet/CinC Challenges range from physiologic signal processing and analysis to forecasting and modelling clinically important events and processes.

Event detection is an essential component of microseismic data analysis. This process is typically carried out using a short- and long-term-average-ratio (STA/LTA) method, which is simple and computationally efficient but often yields inconsistent results for noisy data sets. We have aimed to optimize the performance of the STA/LTA method by testing different input forms of 3C waveform data and different characteristic functions (CFs), including a proposed [Formula: see text]-mean CF. These tests are evaluated using receiver operating characteristic (ROC) analysis and are compared based on synthetic and field data examples. Our analysis indicates that the STA/LTA method using a [Formula: see text]-mean CF improves the detection sensitivity and yields more robust event detection on noisy data sets than some previous approaches. In addition, microseismic events are detected efficiently on field data examples using the same detection threshold obtained from the ROC analysis on synthetic data examples. We recommend the use of the Youden index based on ROC analysis using a training subset, extracted from the continuous data, to further improve the detection threshold for field microseismic data.

In the early stage of acoustic full-waveform inversion (AFWI), it is important to exploit the long-wavelength features of the gradient and suppress its short-wavelength features to update the background velocity. However, due to strong near-offset PP reflections and multiples within the pressure data, conventional AFWI sometimes primarily reconstructs the short-wavelength features of a given model and fails to converge on a reliable subsurface P-wave velocity model. Therefore, this study, we propose the usage of pseudo-horizontal particle acceleration (pseudo-ax) data for AFWI, rather than the original pressure data. Because the amplitudes of PP reflections are implicitly weighted according to the reflection angle in pseudo-ax data, the near-offset PP reflections and multiples of the pseudo-ax data are much weaker than those of the pressure data. As a result, AFWI using pseudo-ax data focuses on reconstructing the longer-wavelength features of a given model in the early iterations, and then gradually reconstructs short-wavelength features as the inversion process continues. Using synthetic and real data examples for the Volve oilfield in the North Sea, we examined the effects and feasibility of this strategy. In the synthetic data example, the proposed strategy rapidly reduced far-offset data misfits related to long-wavelength feature updates and achieved smaller model misfits than the conventional strategy. The real data example showed that the velocity model reconstructed using the proposed strategy flattened some of the reflectors at the angle-domain common-image gathers (ADCIGs) better than the velocity model obtained using the conventional strategy. These findings demonstrate that our strategy converges closer to the real P-wave velocity model than the conventional strategy by building a hierarchical velocity model from long- to short-wavelength structures.

The process of extracting meaningful rules from big and complex data is called data mining. Data mining has an increasing popularity in every field today. Data units are established in customer-oriented industries such as marketing, finance and telecommunication to work on the customer churn and acquisition, in particular. Among the data mining methods, classification algorithms are used in studies conducted for customer acquisition to predict the potential customers of the company in question in the related industry. In this study, bank marketing data set in UCI Machine Learning Data Set was used by creating models with the same classification algorithms in different data mining programs. Accuracy, precision and f- measure criteria were used to test performances of the classification models. When creating the classification models, the test and training data sets were randomly divided by the holdout method to evaluate the performance of the data set. The data set was divided into training and test data sets with the 60-40%, 75­25% and 80-20% separation ratios. Data mining programs used for these processes are the R, Knime, RapidMiner and WEKA. And, classification algorithms commonly used in these platforms are the k-nearest neighbor (k-nn), Naive Bayes, and C4.5 decision tree.

The multi‐to hyperspectral evolution of satellite ocean color sensors is anticipated to enable satellite‐based identification of phytoplankton biodiversity, a key factor in aquatic ecosystem functioning and upper ocean biogeochemistry. In this work the bio‐optical Phytoplankton Detection with Optics (PHYDOTax) approach for deriving taxonomic (class‐level) phytoplankton community composition (PCC, e.g. diatoms, dinoflagellates) from hyperspectral information is evaluated in the Chesapeake Bay estuary on the U.S. East Coast. PHYDOTax is among relatively few optical‐based PCC differentiation approaches available for optically complex waters, but it has not yet been evaluated beyond the California coastal regime where it was developed. Study goals include: (a) testing the approach in a turbid estuary including novel incorporation of colored dissolved organic matter (CDOM) and non‐algal particles (NAP), and (b) performance assessment with both synthetic mixture and field data sets. Algorithm skill was robust on synthetic mixtures. Using field data, cryptophyte and/or cyanophyte phytoplankton groups were predicted, but diatom and dinoflagellate detection was not conclusive. For one field data set, small but significant improvements were observed in predicted PCC groups when tested with incorporation of CDOM and NAP into the algorithm, but not for the second field data set. Sensitivity to three hyperspectral‐relevant spectral resolutions (1, 5, 10 nm) was low for all field and synthetic data. PHYDOTax can identify some phytoplankton groups in the estuary using hyperspectral, field‐collected measurements, but validation‐quality data with broad temporospatial coverage are needed to determine whether the approach is robust enough for science applications.

We present a rank reduction algorithm that permits simultaneous reconstruction and random noise attenuation of seismic records. We based our technique on multichannel singular spectrum analysis (MSSA). The technique entails organizing spatial data at a given temporal frequency into a block Hankel matrix that in ideal conditions is a matrix of rank k, where k is the number of plane waves in the window of analysis. Additive noise and missing samples will increase the rank of the block Hankel matrix of the data. Consequently, rank reduction is proposed as a means to attenuate noise and recover missing traces. We present an iterative algorithm that resembles seismic data reconstruction with the method of projection onto convex sets. In addition, we propose to adopt a randomized singular value decomposition to accelerate the rank reduction stage of the algorithm. We apply MSSA reconstruction to synthetic examples and a field data set. Synthetic examples were used to assess the performance of the method in two reconstruction scenarios: a noise-free case and data contaminated with noise. In both cases, we found extremely low reconstructions errors that are indicative of an optimal recovery. The field data example consists of a 2D prestack volume that depends on common midpoint and offset. We use the MSSA reconstruction method to complete missing offsets and, at the same time, increase the signal-to-noise ratio of the seismic volume.

Agood seismic velocity model is required for many routine seismic imaging techniques. Velocity model building from seismic data is often labor intensive and time consuming. The process becomes more complicated by taking nonhyperbolic traveltime estimations into account. An alternative to the conventional time-domain imaging algorithms is to use techniques based on the local event slopes, which contain sufficient information about the traveltime moveout for velocity estimation and characterization of the subsurface geologic structures. Given the local slopes, there is no need for a prior knowledge of a velocity model. That is why the term “velocity independent” is commonly used for such techniques. We improved upon and simplified the previous versions of velocity-independent nonhyperbolic approximations for horizontally layered vertical transverse isotropy (VTI) media by removing one order of differentiation with respect to offset from the imaging kinematic attributes. These kinematic attributes are derived in terms of the local event slopes and zero-offset two-way traveltime (TWTT). We proposed the use of predictive painting, which keeps all the attributes curvature independent, to estimate the zero-offset TWTT. The theoretical contents and performance of the proposed approach were evaluated on synthetic and field data examples. We also studied the accuracy of moveout attributes for shifted hyperbola, rational, three-parameter, and acceleration approximations on a synthetic example. Our results show that regardless of the approximation types, NMO velocity estimate has higher accuracy than the nonhyperbolicity attribute. Computational time and accuracy of the inversion of kinematic attributes in VTI media using our approach were compared with routine/conventional multiparameter semblance inversion and with the previous velocity-independent inversion techniques.

Real-World Surveillance of FDA-Cleared Artificial Intelligence Models: Rationale and Logistics.

There are different methods to predict survival time, including traditional statistical regression, and modern neural network. Still, precise prediction of survival time remains as a challenge. In recent years algebraic topology was introduced as a novel method to infer information from high dimensional data set, where a mapper uses an algorithm to obtain geometric or topological features of the data set, and associate the features with the predicted variables. In this study, the data set includes gender, age, AFP, GGT, PPT, TB, ALT, tumor size, tumor number, BCLC stage, number of liver nodules, type of surgical approach etc., total 21 diagnostic variables. We use these variables to predicted OS and DFS, and the data set contains 329 records. In this study, the topology mapper is executed to acquire the topological features of all the data (including training and test sets), since this task is unsupervised, so the target values are not considered. Our method is to infer three new features of each data record, including the location of the record on the main branch of the topology, whether it is a branch node (i.e. has more than 2 branches), and whether it is an endpoint. A simple neural network with one hidden layer is designed and implemented with Tensorflow, and the three features will be added to the data set and provide valuable information for prediction of the target values. Without adding the new features from topology mapper, we have 72% accuracy (in predicting the survival time is greater or less than the mean value) on test set data when predicting OS, and we have 75% accuracy on test set data when predicting the DFS. With the new features from topology mapper, we have 83% accuracy on test set data when predicting OS, and 84% accuracy on test set data when predicting the DFS. The three new features inferred from the topology mapper have a significant effect on the accuracy of prediction and provide better accuracy. The topology based data analysis method in this study could be applied to other classification problems related to clinical trial data, and will need further verification. Citation Format: Cheng Huang, Hui-Chuan Sun, Junda Chen, Lin Shi, Fengqing Li, Yihui Lin, Hanyan Yang. A topology based data analysis method to predict survival time [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1634.

We have developed an inversion approach based on particle swarm optimization (PSO) for 2D direct current resistivity data. Once the PSO algorithm is implemented, cost-function changes on constraints and misfit criteria can be easily performed and an approximate invariance of the Lagrange multipliers is obtained if each term of the cost function is properly normalized. More importantly, the interpreter can divide the mesh into partitions and apply a different constraint or combination of constraints in each partition. We explore several restrictions on the log-resistivity variation, either in single or in joint versions, including spatial continuity constraints in the [Formula: see text]-and [Formula: see text]-norms, total variation, sparsity constraints using the discrete cosine transform and Daubechies bases, and minimum moment of inertia constraints to impose concentration of resistive or conductive materials along target axes. In the latter case, the earth surface might be used also as a target axis. In addition, the final state of the PSO algorithm, once a stopping condition has been reached, contains not only the best solution but also a cluster of good suboptimal solutions that can be used for uncertainty analysis. As a result, the interpreter has the flexibility to perform an interpretation process based on a feedback trial-and-error inversion approach, in a similar manner that he/she has when using a friendly forward modeling software. The interpreter can then drive the tentative solutions obtained along the inversion process to incorporate their conceptions about the geologic environment, besides appraising misfit and stability of the solutions. We evaluate synthetic and field data examples. In the first case, we determine how the interpreter can drive the inversion process in a karst environment to image dissolution features. In the second case, we explore a similar situation aiming to optimize borehole locations in fracture zones in crystalline rocks.

An implementation of differential evolution algorithm for inversion of geoelectrical data

Hydraulic fracturing (HF) plays an important role when it comes to the extraction of resources in unconventional reservoirs. The microseismic activity arising during HF operations needs to be monitored to improve productivity and to make decisions about mitigation measures. Recently, deep-learning methods have been investigated to localize earthquakes given field-data waveforms as input. For optimal results, these methods require large field data sets that cover the entire region of interest. In practice, such data sets often are scarce. To overcome this shortcoming, we have initially used a (large) synthetic data set with full waveforms to train a U-Net that reconstructs the source location as a 3D Gaussian distribution. As a field data set for our study, we use data recorded during HF operations in Texas. Synthetic waveforms are modeled using a velocity model from the site that is also used for a conventional diffraction-stacking (DS) approach. To increase the U-Nets’ ability to localize seismic events, we augment the synthetic data with different techniques, including the addition of field noise. We select the best performing U-Net using 22 events that have previously been identified to be confidently localized by DS and apply that U-Net to all 1245 events. We compare our predicted locations to DS and the DS locations refined by a relative location (DSRL) method. The U-Net-based locations are better constrained in depth compared to DS and the mean hypocenter difference with respect to DSRL locations is 163 m. This indicates potential for the use of synthetic data to complement or replace field data for training. Furthermore, after training, the method returns the source locations in near real time given the full waveforms, alleviating the need to pick arrival times.

The role of pertinently diversified and balanced training as well as testing data sets in achieving the true performance of classifiers in predicting the antifreeze proteins

In mineral exploration, geophysical inversion is a common mathematical tool to obtain reliable information on subsurface density properties based on gravity measurements. Many inversion algorithms were developed to obtain the density distribution in the Earth’s subsurface. Recovered density values are usually lower/higher than the actual density as a consequence of inversion algorithm. This paper presents the use of a fuzzy cross-update inversion (FCUI) procedure to improve the subsurface density model based on a triangular grid. The algorithm is written in MATLAB and uses fuzzy c means clustering to improve the density modeling result per iteration. Two additional input parameters are added, namely the number of geologic units in the model (i.e., number of clusters) and the cluster center values of the geologic units (mean density value of each geologic unit). Inversion results from the FCUI are presented and compared with conventional inversion. The effectiveness of the developed technique is tested for the interpretation of synthetic data and two sets of field data. The FCUI approach shows improvement over conventional inversion approaches in differentiating geologic units. Further, FCUI was performed to reduce ambiguity of interpretation for the delineation of chromite and uranium deposits as the first and second case studies, respectively. We integrated favorable information and show the efficacy of FCUI over conventional inversion for the field datasets.

Seismic fracture characterization in tight sand reservoirs: A case study of the Xujiahe Formation, Sichuan Basin, China