Search

Classifying images using supervised machine learning (ML) relies on labeled training data—classes or text descriptions, for example, associated with each image. Data‐driven models are only as good as the data used for training, and this points to the importance of high‐quality labeled data for developing a ML model that has predictive skill. Labeling data is typically a time‐consuming, manual process. Here, we investigate the process of labeling data, with a specific focus on coastal aerial imagery captured in the wake of hurricanes that affected the Atlantic and Gulf Coasts of the United States. The imagery data set is a rich observational record of storm impacts and coastal change, but the imagery requires labeling to render that information accessible. We created an online interface that served labelers a stream of images and a fixed set of questions. A total of 1,600 images were labeled by at least two or as many as seven coastal scientists. We used the resulting data set to investigate interrater agreement: the extent to which labelers labeled each image similarly. Interrater agreement scores, assessed with percent agreement and Krippendorff's alpha, are higher when the questions posed to labelers are relatively simple, when the labelers are provided with a user manual, and when images are smaller. Experiments in interrater agreement point toward the benefit of multiple labelers for understanding the uncertainty in labeling data for machine learning research.

We use Genetic Programming (GP), a type of Machine Learning (ML) approach, to predict the total and infragravity swash excursion using previously published datasets that have been used extensively in swash prediction studies. Three previously published works with a range of new conditions are added to this dataset to extend the range of measured swash conditions. Using this newly compiled dataset we demonstrate that a ML approach can reduce the prediction errors compared to well-established parameterizations and therefore it contributes to the error in coastal hazards assessment (e.g. coastal inundation). Predictors obtained using GP can also be physically sound and replicate the functionality and dependencies of previous published formulas. Overall, we show that ML techniques are capable of both improving predictability (compared to classical regression approaches) and providing physical insight into coastal processes.

Despite decades of regulatory efforts in the United States to decrease vulnerability in developed coastal zones, exposure of residential assets to hurricane damage is increasing — even in places where hurricanes have struck before. Comparing plan-view footprints of individual residential buildings before and long after major hurricane strikes, we find a systematic pattern of ‘building back bigger’ among renovated and new properties. US coastlines that are exposed to hurricanes are subject to a myriad of regulations regarding building and rebuilding of structures, yet satellite imagery shows that the footprint of residential buildings increases after hurricane events for both new and renovated structures. Such an effect poses a challenge for vulnerable coastlines to build resilience in the face of growing hazards and houses.

Numerical models rely on the parameterization of processes that often lack a deterministic description. In this contribution we demonstrate the applicability of using machine learning, a class of optimization tools from the discipline of computer science, to develop parameterizations when extensive data sets exist. We develop a new predictor for near-bed suspended sediment reference concentration under unbroken waves using genetic programming, a machine learning technique. We demonstrate that this newly developed parameterization performs as well or better than existing empirical predictors, depending on the chosen error metric. We add this new predictor into an established model for inner-shelf sorted bedforms. Additionally we incorporate a previously reported machine-learning-derived predictor for oscillatory flow ripples into the sorted bedform model. This new "hybrid" sorted bedform model, whereby machine learning components are integrated into a numerical model, demonstrates a method of incorporating observational data (filtered through a machine learning algorithm) directly into a numerical model. Results suggest that the new hybrid model is able to capture dynamics previously absent from the model – specifically, two observed pattern modes of sorted bedforms. Lastly we discuss the challenge of integrating data-driven components into morphodynamic models and the future of hybrid modeling.

A range of computer science methods termed machine learning (ML) enables the extraction of insight and quantitative relationships from multidimensional datasets. Here, we review the use of ML on supervised regression tasks in studies of coastal morphodynamics and sediment transport. We examine aspects of ‘what’ and ‘why’, such as ‘what’ science problems ML tools have been used to address, ‘what’ was learned when using ML, and ‘why’ authors used ML methods. We find a variety of research questions have been addressed, ranging from small-scale predictions of sediment transport to larger-scale sand bar morphodynamics and coastal overwash on a developed island. We find various reasons justify the use of ML, including maximize predictability, emulation of model components, the need for smooth and continuous nonlinear regression, and explicit inclusion of uncertainty. The expanding use of ML has allowed for an expanding set of questions to be addressed. After reviewing the studies we outline a set of best practices for coastal researchers using machine learning methods. Finally we suggest possible areas for future research, including the use of novel machine learning techniques and exploring open data that is becoming increasingly available.

We use a machine learning approach to seek an accurate, physically sound predictor, to estimate the mean velocity for open-channel flow when submerged arrays of rigid cylinders (model vegetation) are present. A genetic programming routine is used to find a robust relationship between relevant properties of the model vegetation and flow parameters. We use published data from laboratory experiments covering a broad range of conditions to obtain an equation that matches the performance of other predictors from recent literature in terms of accuracy, while showing a less complex structure. We also investigate how different criteria for data selection, as well as the size of the data set used to train the algorithm, influences the accuracy of the resulting predictors. Our results show that a proper use of Machine-Learning techniques does not only provide empirical correlations, but can yield physically sound models as representative of the physical processes involved. We provide a clear, thorough example of the application of GP, its advantages and shortcomings, to encourage the use of data-driven techniques as part of the data analysis process, and to address common misconceptions of machine learning as simple correlation techniques or physically senseless statistical analysis.

We use a machine learning approach based on genetic programming to predict noncohesive particle settling velocity. The genetic programming routine is coupled to a novel selection algorithm that determines training data from a collected database of published experiments (985 measurements). While varying the training data set size and retaining an invariant validation set we perform multiple iterations of genetic programming to determine the least data needed to train the algorithm. This method retains a maximum quantity of data for testing against published predictors. The machine learning predictor for settling velocity performs better than two common predictors in the literature and indicates that particle settling velocity is a nonlinear function of all the provided independent variables: nominal diameter of the settling particle, kinematic viscosity of the fluid, and submerged specific gravity of the particle.

Are you looking to communicate science more broadly? Start by editing articles on the Internet's most popular general reference work.

Purpose Recently, the Accreditation Council for Graduate Medical Education (ACGME) has developed and enforced a complex set of regulations limiting resident duty hours (RDHs). One potential method to comply with these stringent regulations and better document resident work hours would be to use text message (TM)/short message service (SMS), allowing rapid, inexpensive, and interactive 2-way delivery of information. The purpose of this study was to document the successful implementation of TM to enhance compliance with the ACGME RDH regulations. Methods Our TM system uses TSHEETS (TS; Meridian, Ohio), which is an online time- and labor-management service. After our general surgical residents were registered with the system, the documentation of entry into (texting “t start” to a predetermined number) and exit out of (texting “t stop” to the same number) the hospital could easily be accomplished. The goals of this pilot study were to determine (1) the feasibility of implementing a program such as this and (2) if use of such a program was associated with high resident satisfaction and improved compliance. Results We registered 39 residents with TS. Within 3 days, compliance rose from 76.9% to 96.2%. After implementation of TM reminders, compliance increased to 100%. A time-tracking resident satisfaction survey was distributed after 8 weeks of data collection. Twenty six of 39 (67%) residents participated in the survey, with 25 (96%) being more satisfied with the current application as compared with either of the previous (manual or swipe card) systems, specifically regarding ease of use and overall ACGME RDH compliance. Self-reported resident scores of their ability to be compliant with ACGME RDH before versus after implementation increased from 47% to 75% (p Conclusions We were able to implement successfully a novel technique for ACGME RDH documentation and compliance in a general surgery residency program through the use of TM; this approach employed a state-of-the-art time-tracking method that was associated with high levels of resident work-hour compliance and overall satisfaction.

Podcasting is a recent creation combining old and new technologies allowing rapid, inexpensive delivery of media content (primarily audio) to the end user, both via the desktop environment and personal media players. The authors' group (the Society of Critical Care Medicine) saw the educational and communication potential for the podcasting concept, and have successfully designed and implemented the first podcast of a national medical society. As of this writing, there are an average of (mean +/- SD) 664 +/- 290 total downloads per podcast, and their podcast feed has been hit over 68,000 times in its first seven months. In this manuscript, the authors provide documentation of their successful endeavor, as well as a structured framework for other organizations to create similar products.

Physics-based numerical models designed to study processes on the surface of the Earth are commonly built with conservation laws. Yet conservations laws are based on a treatable subset of all physical and biological processes operating in the system of interest, and empirical relationships are always required to fully describe systems (and provide mathematical “closure”). In contrast to conservation laws, empirical expressions used in model construction are inductive, and based on observations of phenomena. Any model that requires empirical expressions is also subject to revision as the empirical parameterization is refined: all empiricism is open to revision by corroborating or conflicting data. As more data becomes available, and more degrees of freedom are explored, it becomes harder to incorporate all available data into a single optimal empirical predictor. We argue in this contribution that empirical parameterizations for numerical models should be constructed using machine learning techniques because these techniques are built to operate on large, high dimensional datasets. Machine learning, in the context of this paper, defines a suite of algorithms used to develop predictive relationships (correlations) using a set of input data. Examples of commonly used machine learning techniques in the Earth sciences are artificial neural networks (e.g., Maier and Dandy, 2000; Pape et al., 2007; van Maanen et al., 2010), regression trees (e.g., Snelder et al., 2009; Oehler et al., 2012), Bayesian networks (e.g., Aguilera et al., 2011; Yates and Le Cozannet, 2012), and evolutionary algorithms (e.g., Knaapen and Hulscher, 2002; Ruessink, 2005; Goldstein et al., 2013). Machine learning techniques offer insight and are high-performance, reproducible, and scalable. The inclusion of powerful inductive techniques like those provided by machine learning offers the opportunity to enhance the predictions obtained from deductive approaches. Additionally the use of machine learning often leads to further insight. The use of empiricisms built from machine learning in a physics-based model results in a “hybrid” model. Combining the strengths of inductive (data-driven) and deductive (physics-based) approaches in a single hybrid model has been suggested as a valuable step forward in model development because of increases in accuracy (e.g., Babovic et al., 2001; Hall, 2004) and speed (Krasnopolsky and Fox-Rabinovitz, 2006). An additional benefit is the direct coupling of models to data

Neck masses, frequently encountered by physicians, comprise a vast range of diagnoses, with malignancy being the greatest concern. Calcifying fibrous pseudotumor (CFP) is a rare lesion with unknown pathogenesis, characterized pathologically by a predominance of abundant hyalinized collagenous tissue with focal lymphoplasmacytic infiltrate and psammomatous or dystrophic calcifications. We present the case of a 29-year-old woman who presented with a 4-cm left neck mass, accompanied by constitutional symptoms of vague weakness and lethargy. After the lesion failed to respond to a course of antibiotic therapy, fine-needle aspiration was performed, the pathology of which was indeterminate. The concern was that the lesion was a lymphoproliferative disorder–further workup was performed. CT of the chest, abdomen, and pelvis revealed no evidence of adenopathy or neoplasms. Subsequently, an incisional biopsy was performed, suggesting a diagnosis of CFP. Magnetic resonance imaging with contrast, performed to delineate the anatomy, revealed the lesion in the left neck, deep to the left clavicle, that extended superiorly into the supraclavicular fossa. Complete surgical removal of the lesion was successfully performed, with immunophenotyping confirming the initial diagnosis of CFP. We present a case report of cervical CFP, discuss the approach to neck masses, and review the recent literature on this rare, benign entity.

Although gallstone disease is highly prevalent, cholelithiasis causing gallstone ileus is uncommon. Consideration has been given for nonoperative strategies to resolve obstruction due to the significant age and comorbidities afflicting this population. A 94-year-old man presented with a 5-day history of abdominal distension and tenderness. CT scan revealed multiple large gallstones within the gallbladder, pneumobilia, and two ectopic gallstones (antrum of the stomach and distal ileum). The patient was taken to the operating room where an enterolithotomy and gastrotomy was performed with removal of gallstones and subsequent relief of obstruction. During the postoperative course, the patient developed symptoms of gastric outlet obstruction and underwent gastrointestinal endoscopy for diagnosis and treatment. Two large gallstones, present in the duodenum, were retracted into the stomach using a Roth net but could not be retrieved beyond the upper esophageal sphincter. A holmium: yttrium-aluminum-garnet (Holmium: YAG) laser was used for fragmentation of the stones, with subsequent successful removal. This is the first documented successful use of the holmium: YAG laser for the treatment of recurrent gallstone ileus. Physicians should remember that in a small but important subgroup of patients, endoscopy accompanied by laser lithotripsy may prove beneficial.

The management of gastric lymphoma is a rapidly changing field. The classification and staging of lymphomas have been revised in the past two decades, reflecting diagnostic advances that include the use of immunohistochemical stains and cell-surface markers. Furthermore, the use of CT scanning and endoscopic ultrasound has revolutionized the non-operative diagnostic modalities available. Despite these advances, the future of gastric lymphoma research lies in the development of therapeutic regimens.

Checklists have been recently promulgated as a method to enhance patient safety and improve outcomes for critically ill patients. Specifically, recent work performed by researchers from the Johns Hopkins Medical Institutions has demonstrated that the addition of checklists to usual care in the intensive care unit is associated with a decrease in the incidence of catheter-related bloodstream infections. Initially evaluated at the institutional level, this effort has been successfully expanded to the state level as part of the Michigan Keystone Project. Although this work has recently received significant positive attention in the lay press, the Office for Human Research Protections--as they felt that this was a research project requiring Institutional Review Board approval and informed consent--put the data collection on hold for lack of approval by the Institutional Review Board at the participating hospitals in Michigan as well as for not having obtained informed consent from each patient and clinician involved in the project. This article documents the recent events surrounding the Keystone Project and the response to the actions taken by the Office for Human Research Protections in the lay press and the new media (Internet and blogs), articulates how a determination can be made if a project is quality-improvement, human-subjects research, or both, and proposes some solutions to create a structured approach to this kind of research in the future.

2 F 2-day-old, full-term neonate diagnosed with a cystic lesion of the left lung during antenatal ultrasound ith subsequent magnetic resonance imaging for charcterization was referred for surgical consultation. The renatal magnetic resonance image scan was remarkable or abnormal increased T2 signal of the left upper lobe arrow, Fig 1) as compared with the remainder of the ungs. The differential diagnosis included bronchopulonary sequestration versus cystic adenomatoid malforation. The neonate underwent a 16-slice multi-detector omputed tomographic angiogram with volume renered imaging (Fig 2) that revealed a hypervascular mass 3.7 2.6 2.1 cm) in the anterior aspect of the left lung ase with the appearance of normal lung parenchyma. oth the arterial supply, arising from the left subclavian rtery immediately proximal to the origin of a dilated left nternal mammary artery, and venous drainage exiting he left innominate vein were prominent vessels running arallel to each other (arrow, Fig 2). Findings in the left ung base were compatible with an extralobar sequestraion. The patient underwent a muscle-sparing left thoraotomy. Intraoperative findings included the aforemenioned sequestered lung segment, adjacent to the inferior ingular segment, which was attached by a solitary vasular pedicle to the chest wall. Division of this attachent allowed for subsequent excision. Pulmonary sequestration is an uncommon mass comrising 6% of all congenital pulmonary malformations.