Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

Unravelling the metabolism black-box in a dynamic wetland environment using a hybrid model framework: Storm driven changes in oxygen budgets.

In The Science of the total environment

Estimating gross primary production and ecosystem respiration from oxygen data is performed widely in aquatic systems, yet these estimates can be challenged by high advective fluxes of oxygen. In this study, we develop a hybrid framework linking data-driven and process-based modelling to examine the effect of storm events on oxygen budgets in a constructed wetland. After calibration against measured flow and water temperature data over a two-month period with three storm events, the model was successfully validated against high frequency dissolved oxygen (DO) data exhibiting large diurnal fluctuations. The results demonstrated that pulses of high-DO water injected into the wetland during storm events were able to dramatically change the wetland oxygen budget. A shift was observed in the dominant oxygen inputs, from benthic net production during non-storm periods, to inflows of oxygen during storm events, which served to dampen the classical diurnal oxygen signature. The model also demonstrated the changing balance of pelagic versus benthic production and hypoxia extent in response to storm events, which has implications for the nutrient attenuation performance of constructed wetlands. The study highlights the benefit of linking analysis of high-frequency oxygen data with process-based modelling tools to unravel the varied responses of components of the oxygen budget to storm events.

Liu Junjie, Wang Benya, Oldham Carolyn E, Hipsey Matthew R


Constructed wetland, Hydrodynamic model, Hypoxia, Machine learning, Photosynthesis, Urban drainage

General General

SDARE: A stacked denoising autoencoder method for game dynamics network structure reconstruction.

In Neural networks : the official journal of the International Neural Network Society

Complex network is a general model to represent the interactions within technological, social, information, and biological interaction. Often, the direct detection of the interaction relationship is costly. Thus, network structure reconstruction, the inverse problem in complex networked systems, is of utmost importance for understanding many complex systems with unknown interaction structures. In addition, the data collected from real network system is often contaminated by noise, which makes the network structure inference task much more challenging. In this paper, we develop a new framework for the game dynamics network structure reconstruction based on deep learning method. In contrast to the compressive sensing methods that employ computationally complex convex/greedy algorithms to solve the network reconstruction task, we introduce a deep learning framework that can learn a structured representation from nodes data and efficiently reconstruct the game dynamics network structure with few observation data. Specifically, we propose the denoising autoencoders (DAEs) as the unsupervised feature learner to capture statistical dependencies between different nodes. Compared to the compressive sensing based method, the proposed method is a global network structure inference method, which can not only get the state-of-art performance, but also obtain the structure of network directly. Besides, the proposed method is robust to noise in the observation data. Moreover, the proposed method is also effective for the network which is not exactly sparse. Accordingly, the proposed method can extend to a wide scope of network reconstruction task in practice.

Huang Keke, Li Shuo, Dai Penglin, Wang Zhen, Yu Zhaofei


Complex network, Compressive sensing, Deep learning, Network structure reconstruction, Stacked denoising autoencoder

General General

Assessing the biophysical and social drivers of burned area distribution at the local scale.

In Journal of environmental management

Understanding the characteristics of wildfire-affected communities and the importance of particular factors of different dimensions, is paramount to improve prevention and mitigation strategies, tailored to people's needs and abilities. In this study, we explored different combinations of biophysical and social factors to characterize wildfire-affected areas in Portugal. By means of machine-learning methods based on classification trees, we assessed the predictive ability of various models to discriminate different levels of wildfire incidence at the local scale. The model with the best performance included a reduced set of both biophysical and social variables and we found that, oveall, the exclusion of specific variables improved prediction rates of group classification. The most important variables were related to landcover; the civil parishes covered by more than 20% of shrublands were more fire-prone, whereas those parishes with at least 40% of agricultural land were less affected by wildfires. Regarding social variables, the most-affected parishes showed a lower proportion of foreign residents and lower purchasing power, conditions likely associated with the socioeconomic context of inland low-density rural areas, where rural abandonment, depopulation and ageing trends have been observed in the last decades. Further research is needed to investigate how other particular parameters representing the social context, and its evolution, can be integrated in wildfire occurrence modelling, and how these interact with the biophysical conditions over time.

Oliveira Sandra, Zêzere José Luís


Biophysical factors, Local communities, Portugal, Risk mitigation, Sociodemographic variables, Wildfire incidence

General General

A diffeomorphic unsupervised method for deformable soft tissue image registration.

In Computers in biology and medicine

BACKGROUND AND OBJECTIVES : The image registration methods for deformable soft tissues utilize nonlinear transformations to align a pair of images precisely. In some situations, when there is huge gray scale difference or large deformation between the images to be registered, the deformation field tends to fold at some local voxels, which will result in the breakdown of the one-to-one mapping between images and the reduction of invertibility of the deformation field. In order to address this issue, a novel registration approach based on unsupervised learning is presented for deformable soft tissue image registration.

METHODS : A novel unsupervised learning based registration approach, which consists of a registration network, a velocity field integration module and a grid sampling module, is presented for deformable soft tissue image registration. The main contributions are: (1) A novel encoder-decoder network is presented for the evaluation of stationary velocity field. (2) A Jacobian determinant based penalty term (Jacobian loss) is developed to reduce the folding voxels and to improve the invertibility of the deformation field.

RESULTS AND CONCLUSIONS : The experimental results show that a new pair of images can be accurately registered using the trained registration model. In comparison with the conventional state-of-the-art method, SyN, the invertibility of the deformation field, accuracy and speed are all improved. Compared with the deep learning based method, VoxelMorph, the proposed method improves the invertibility of the deformation field.

Zhang Shuo, Liu Peter Xiaoping, Zheng Minhua, Shi Wen


Deformable soft tissue image registration, Encoder–decoder network, Invertibility, Jacobian loss, Unsupervised learning

Radiology Radiology

Impact of hybrid supervision approaches on the performance of artificial intelligence for the classification of chest radiographs.

In Computers in biology and medicine

PURPOSE : To evaluate the impact of different supervision regimens on the training of artificial intelligence (AI) in the classification of chest radiographs as normal or abnormal in a moderately sized cohort of individuals more likely to be outpatients.

MATERIALS AND METHODS : In a retrospective study, 7000 consecutive two-view chest radiographs obtained from 2012 to 2015 were labeled as normal or abnormal based on clinical reports. A convolutional neural network (CNN) was trained on this dataset and then evaluated with an unseen subset of 500 radiographs. Five different training approaches were tested: (1) weak supervision and four hybrid approaches combining weak supervision and extra supervision with annotation in (2) an unbalanced set of normal and abnormal cases, (3) a set of only abnormal cases, (4) a set of only normal cases, and (5) a balanced set of normal and abnormal cases. Standard binary classification metrics were assessed.

RESULTS : The weakly supervised model achieved an accuracy of 82%, but yielded 75 false negative cases, at a sensitivity of 70.0% and a negative predictive value (NPV) of 75.5%. Extra supervision increased NPV at the expense of the false positive rate and overall accuracy. Extra supervision with training using a balance of abnormal and normal radiographs resulted in the greatest increase in NPV (87.2%), improved sensitivity (92.8%), and reduced the number of false negatives by more than fourfold (18 compared to 75 cases).

CONCLUSION : Extra supervision using a balance of annotated normal and abnormal cases applied to a weakly supervised model can minimize the number of false negative cases when classifying two-view chest radiographs. Further refinement of such hybrid training approaches for AI is warranted to refine models for practical clinical applications.

Ellis Ryan, Ellestad Erik, Elicker Brett, Hope Michael D, Tosun Duygu


Attention mining, Chest radiographs, Extra supervision, Hybrid supervision, Outpatient, Two-view radiographs, Weak supervision

General General

Machine learning for microbial identification and antimicrobial susceptibility testing on MALDI-TOF mass spectra: a systematic review.

In Clinical microbiology and infection : the official publication of the European Society of Clinical Microbiology and Infectious Diseases

BACKGROUND : The matrix assisted laser desorption/ionization and time-of-flight mass spectrometry (MALDI-TOF MS) technology has revolutionized the field of microbiology by facilitating precise and rapid species identification. Recently, machine learning techniques have been leveraged to maximally exploit the information contained in MALDI-TOF M with the ultimate goal to refine species identification and streamline antimicrobial resistance determination.

OBJECTIVES : To systematically review and evaluate studies employing machine learning for the analysis of MALDI-TOF mass spectra.

DATA SOURCES : Using PubMed/Medline, Scopus, and Web of Science, we searched the existing literature for machine learning supported applications of MALDI-TOF mass spectra for microbial species and antimicrobial susceptibility identification.

STUDY ELIGIBILITY CRITERIA : Original research studies using machine learning to exploit MALDI-TOF mass spectra for microbial species and antimicrobial susceptibility identification were included. Studies focusing on single proteins and peptides, case studies, and review articles were excluded.

METHODS : A systematic review according to the PRISMA guidelines was performed and a quality assessment of the machine learning models conducted.

RESULTS : From the 36 studies that met our inclusion criteria, 27 employed machine learning for species identification and 9 for antimicrobial susceptibility testing. Support Vector Machines, Genetic Algorithms, Artificial Neural Networks, and Quick Classifiers were the most frequently used machine learning algorithms. The quality of the studies ranged between poor and very good. While, the majority of the studies reported how to interpret the predictors (88.89%) and suggested possible clinical applications of the developed algorithm (100%), only four studies (11.11%) validated machine learning algorithms on external datasets.

CONCLUSIONS : A growing number of studies utilizes machine learning to optimize the analysis of MALDI-TOF mass spectra. This review, however, demonstrates that there are certain shortcomings of current machine learning supported approaches that have to be addressed to make them widely available and incorporated them in the clinical routine.

Cv Weis, Cr Jutzeler, K Borgwardt


MALDI-TOF MS, antimicrobial resistance, antimicrobial susceptibility testing, antimicrobial treatment, machine learning, microbial identification