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

General General

Efficient Processing of Spatio-Temporal Data Streams With Spiking Neural Networks.

In Frontiers in neuroscience ; h5-index 72.0

Spiking neural networks (SNNs) are potentially highly efficient models for inference on fully parallel neuromorphic hardware, but existing training methods that convert conventional artificial neural networks (ANNs) into SNNs are unable to exploit these advantages. Although ANN-to-SNN conversion has achieved state-of-the-art accuracy for static image classification tasks, the following subtle but important difference in the way SNNs and ANNs integrate information over time makes the direct application of conversion techniques for sequence processing tasks challenging. Whereas all connections in SNNs have a certain propagation delay larger than zero, ANNs assign different roles to feed-forward connections, which immediately update all neurons within the same time step, and recurrent connections, which have to be rolled out in time and are typically assigned a delay of one time step. Here, we present a novel method to obtain highly accurate SNNs for sequence processing by modifying the ANN training before conversion, such that delays induced by ANN rollouts match the propagation delays in the targeted SNN implementation. Our method builds on the recently introduced framework of streaming rollouts, which aims for fully parallel model execution of ANNs and inherently allows for temporal integration by merging paths of different delays between input and output of the network. The resulting networks achieve state-of-the-art accuracy for multiple event-based benchmark datasets, including N-MNIST, CIFAR10-DVS, N-CARS, and DvsGesture, and through the use of spatio-temporal shortcut connections yield low-latency approximate network responses that improve over time as more of the input sequence is processed. In addition, our converted SNNs are consistently more energy-efficient than their corresponding ANNs.

Kugele Alexander, Pfeil Thomas, Pfeiffer Michael, Chicca Elisabetta


efficient inference, event-based vision, neuromorphic computing, sequence processing, spiking neural networks

General General

Identification of Patient Profiles with High Risk of Hospital Re-Admissions for Acute COPD Exacerbations (AECOPD) in France Using a Machine Learning Model.

In International journal of chronic obstructive pulmonary disease ; h5-index 50.0

Purpose : To characterise patients with chronic obstructive pulmonary disease (COPD) who are rehospitalised for an acute exacerbation, to estimate the cost of these hospitalisations, to characterise high risk patient sub groups and to identify factors potentially associated with the risk of rehospitalisation.

Patients and Methods : This was a retrospective study using the French National Hospital Discharge Database. All patients aged ≥40 years hospitalised for an acute exacerbation of COPD between 2015 and 2016 were identified and followed for six months. Patients with at least one rehospitalisation for acute exacerbation of COPD constituted the rehospitalisation analysis population. A machine learning model was built to study the factors associated with the risk of rehospitalisation using decision tree analysis. A direct cost analysis was performed from the perspective of national health insurance.

Results : A total of 143,006 eligible patients were hospitalised for an acute exacerbation of COPD (AECOPD) in 2015-2016 (mean age: 74 years; 62.1% men). 25,090 (18.8%) were rehospitalised for another exacerbation within six months. In this study, 8.5% of patients died during or immediately following the index hospitalisation and 10.5% died during or immediately after rehospitalisation (p <0.001). The specific cost of these rehospitalisations was € 5304. The overall total cost per patient of all AECOPD-related stays was € 9623, being significantly higher in patients who were rehospitalised (€ 16,275) compared to those who were not (€ 8208). In decision tree analysis, the most important driver of rehospitalisation was hospitalisation in the previous two years (contributing 85% of the information).

Conclusion : Rehospitalisations for acute exacerbations of COPD carry a high epidemiological and economic burden. Since hospitalisation for an acute exacerbation is the most important determinant of future rehospitalisations, management of COPD needs to focus on interventions aimed at decreasing the rehospitalisation risk of in order to lower the burden of disease.

Cavailles Arnaud, Melloni Boris, Motola Stéphane, Dayde Florent, Laurent Marie, Le Lay Katell, Caumette Didier, Luciani Laura, Lleu Pierre Louis, Berthon Geoffrey, Flament Thomas


comorbidity, cost, decision tree analysis, rehospitalisation

General General

Fusion of acoustic sensing and deep learning techniques for apple mealiness detection.

In Journal of food science and technology

Mealiness in apple fruit can occur during storage or because of harvesting in an inappropriate time; it degrades the quality of the fruit and has a considerable role in the fruit industry. In this paper, a novel non-destructive approach for detection of mealiness in Red Delicious apple using acoustic and deep learning techniques was proposed. A confined compression test was performed to assign labels of mealy and non-mealy to the apple samples. The criteria for the assignment were hardness and juiciness of the samples. For the acoustic measurements, a plastic ball pendulum was used as the impact device, and a microphone was installed near the sample to record the impact response. The recorded acoustic signals were converted to images. Two famous pre-trained convolutional neural networks, AlexNet and VGGNet were fine-tuned and employed as classifiers. According to the result obtained, the accuracy of AlexNet and VGGNet for classifying the apples to the two categories of mealy and non-mealy apples was 91.11% and 86.94%, respectively. In addition, the training and classification speed of AlexNet was higher. The results indicated that the suggested method provides an effective and promising tool for assessment of mealiness in apple fruit non-destructively and inexpensively.

Lashgari Majid, Imanmehr Abdullah, Tavakoli Hamed


Apple mealiness assessment, Classification, Convolutional neural networks, Impact response, Red Delicious

General General

Analysis of rice root bacterial microbiota of Nipponbare and IR24.

In Yi chuan = Hereditas

The root-associated bacterial microbiota is closely related to life activities of land plants, and its composition is affected by geographic locations and plant genotypes. However, the influence of plant genotypes on root microbiota in rice grown in northern China remains to be explained. In this study, we performed 16S rRNA gene amplicon sequencing to generate bacterial community profiles of two representative rice cultivars, Nipponbare and IR24. They are planted in Changping and Shangzhuang farms in Beijing and have reached the reproductive stage. We compared their root microbiota in details by Random Forest machine learning algorithm and network analysis. We found that the diversity of rice root microbiota was significantly affected by geographic locations and rice genotypes. Nipponbare and IR24 showed distinct taxonomic composition of the root microbiota and the interactions between different bacteria. Moreover, the root bacteria could be used as biomarkers to distinguish Nipponbare from IR24 across regions. Our study provides a theoretical basis for the in-depth understanding of rice root microbiota in Northern China and the improvement of rice breeding from the perspective of the interaction between root microorganisms and plants.

Hu Ya Li, Dai Rui, Liu Yong Xin, Zhang Jing Ying, Hu Bin, Chu Cheng Cai, Yuan Huai Bo, Bai Yang


diversity analysis, machine learning, network analysis, rice root microbiota, taxonomic composition

Radiology Radiology

Concentration and effective T2 relaxation times of macromolecules at 3T.

In Magnetic resonance in medicine ; h5-index 66.0

PURPOSE : We aimed to investigate the concentration and effective T2 relaxation time of macromolecules assessed with an ultra-short TE sLASER sequence in 2 brain regions, the occipital and frontal cortex, in both genders at 3T.

METHODS : An optimized sLASER sequence was used in conjunction with a double-inversion preparation module to null the metabolites. Eight equally spaced TEs were chosen from 20.1 to 62.1 ms, and the macromolecules were modeled by 10 line broadened singlets. The amplitude of each of the macromolecule signals was extracted at each TE and fit to a monoexponential function to extract the respective effective T2 values. Absolute quantification of the macromolecule resonances was performed using water signal as a reference. A total of 10 young healthy adult subjects (5 females) were scanned, with spectra being obtained from both the frontal and occipital cortex. Differences in the effective T2 relaxation times and concentrations were investigated between both regions and genders.

RESULTS : A wide disparity was observed between the effective T2 values of the individual resonances; however, no significant differences between gender or region for any of the measured macromolecule concentration or effective T2 values were found.

CONCLUSION : The effective T2 relaxation times and concentration of 10 different macromolecule resonances were measured and found to be well represented by the monoexponential model. These results will be useful for absolute quantification of macromolecules in future studies, or in the generation of synthetic basis sets for optimization or machine learning.

Landheer Karl, Gajdošík Martin, Treacy Michael, Juchem Christoph


3T, T2, macromolecules, magnetic resonance spectroscopy, sLASER, short TE

General General

Stability Assessment of Intracranial Aneurysms Using Machine Learning Based on Clinical and Morphological Features.

In Translational stroke research ; h5-index 39.0

Machine learning (ML) as a novel approach could help clinicians address the challenge of accurate stability assessment of unruptured intracranial aneurysms (IAs). We developed multiple ML models for IA stability assessment and compare their performances. We enrolled 1897 consecutive patients with unstable (n = 528) and stable (n = 1539) IAs. Thirteen patient-specific clinical features and eighteen aneurysm morphological features were extracted to generate support vector machine (SVM), random forest (RF), and feed-forward artificial neural network (ANN) models. The discriminatory performances of the models were compared with statistical logistic regression (LR) model and the PHASES score in IA stability assessment. Based on the receiver operating characteristic (ROC) curve and area under the curve (AUC) values for each model in the test set, the AUC values for RF, SVM, and ANN were 0.850 (95% CI 0.806-0.893), 0.858 (95 %CI 0.816-0.900), and 0.867 (95% CI 0.828-0.906), demonstrating good discriminatory ability. All ML models exhibited superior performance compared with the statistical LR and the PHASES score (the AUC values were 0.830 and 0.589, respectively; RF versus PHASES, P < 0.001; RF versus LR, P = 0.038). Important features contributing to the stability discrimination included three clinical features (location, sidewall/bifurcation type, and presence of symptoms) and three morphological features (undulation index, height-width ratio, and irregularity). These findings demonstrate the potential of ML to augment the clinical decision-making process for IA stability assessment, which may enable more optimal management for patients with IAs in the future.

Zhu Wei, Li Wenqiang, Tian Zhongbin, Zhang Yisen, Wang Kun, Zhang Ying, Liu Jian, Yang Xinjian


Artificial intelligence, Intracranial aneurysms, Machine learning, Risk evaluation, Unstable aneurysm