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

General General

Deep Learning of Warping Functions for Shape Analysis.

In Conference on Computer Vision and Pattern Recognition Workshops. IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Workshops

Rate-invariant or reparameterization-invariant matching between functions and shapes of curves, respectively, is an important problem in computer vision and medical imaging. Often, the computational cost of matching using approaches such as dynamic time warping or dynamic programming is prohibitive for large datasets. Here, we propose a deep neural-network-based approach for learning the warping functions from training data consisting of a large number of optimal matches, and use it to predict optimal diffeomorphic warping functions. Results show prediction performance on a synthetic dataset of bump functions and two-dimensional curves from the ETH-80 dataset as well as a significant reduction in computational cost.

Nunez Elvis, Joshi Shantanu H

2020-Jun

General General

Learning distinctive filters for COVID-19 detection from chest X-ray using shuffled residual CNN.

In Applied soft computing

COVID-19 is a deadly viral infection that has brought a significant threat to human lives. Automatic diagnosis of COVID-19 from medical imaging enables precise medication, helps to control community outbreak, and reinforces coronavirus testing methods in place. While there exist several challenges in manually inferring traces of this viral infection from X-ray, Convolutional Neural Network (CNN) can mine data patterns that capture subtle distinctions between infected and normal X-rays. To enable automated learning of such latent features, a custom CNN architecture has been proposed in this research. It learns unique convolutional filter patterns for each kind of pneumonia. This is achieved by restricting certain filters in a convolutional layer to maximally respond only to a particular class of pneumonia/COVID-19. The CNN architecture integrates different convolution types to aid better context for learning robust features and strengthen gradient flow between layers. The proposed work also visualizes regions of saliency on the X-ray that have had the most influence on CNN's prediction outcome. To the best of our knowledge, this is the first attempt in deep learning to learn custom filters within a single convolutional layer for identifying specific pneumonia classes. Experimental results demonstrate that the proposed work has significant potential in augmenting current testing methods for COVID-19. It achieves an F1-score of 97.20% and an accuracy of 99.80% on the COVID-19 X-ray set.

Karthik R, Menaka R, M Hariharan

2020-Sep-23

CNN, COVID-19, Chest X-ray, Deep learning, Pneumonia

Radiology Radiology

Distant metastasis prediction via a multi-feature fusion model in breast cancer.

In Aging ; h5-index 49.0

This study aimed to develop a model that fused multiple features (multi-feature fusion model) for predicting metachronous distant metastasis (DM) in breast cancer (BC) based on clinicopathological characteristics and magnetic resonance imaging (MRI). A nomogram based on clinicopathological features (clinicopathological-feature model) and a nomogram based on the multi-feature fusion model were constructed based on BC patients with DM (n=67) and matched patients (n=134) without DM. DM was diagnosed on average (17.31±13.12) months after diagnosis. The clinicopathological-feature model included seven features: reproductive history, lymph node metastasis, estrogen receptor status, progesterone receptor status, CA153, CEA, and endocrine therapy. The multi-feature fusion model included the same features and an additional three MRI features (multiple masses, fat-saturated T2WI signal, and mass size). The multi-feature fusion model was relatively better at predicting DM. The sensitivity, specificity, diagnostic accuracy and AUC of the multi-feature fusion model were 0.746 (95% CI: 0.623-0.841), 0.806 (0.727-0.867), 0.786 (0.723-0.841), and 0.854 (0.798-0.911), respectively. Both internal and external validations suggested good generalizability of the multi-feature fusion model to the clinic. The incorporation of MRI factors significantly improved the specificity and sensitivity of the nomogram. The constructed multi-feature fusion nomogram may guide DM screening and the implementation of prophylactic treatment for BC.

Ma Wenjuan, Wang Xin, Xu Guijun, Liu Zheng, Yin Zhuming, Xu Yao, Wu Haixiao, Baklaushev Vladimir P, Peltzer Karl, Sun Henian, Kharchenko Natalia V, Qi Lisha, Mao Min, Li Yanbo, Liu Peifang, Chekhonin Vladimir P, Zhang Chao

2020-Sep-28

artificial intelligence, breast neoplasms, early detection, neoplasm metastasis

General General

Near-hysteresis-free soft tactile electronic skins for wearables and reliable machine learning.

In Proceedings of the National Academy of Sciences of the United States of America

Electronic skins are essential for real-time health monitoring and tactile perception in robots. Although the use of soft elastomers and microstructures have improved the sensitivity and pressure-sensing range of tactile sensors, the intrinsic viscoelasticity of soft polymeric materials remains a long-standing challenge resulting in cyclic hysteresis. This causes sensor data variations between contact events that negatively impact the accuracy and reliability. Here, we introduce the Tactile Resistive Annularly Cracked E-Skin (TRACE) sensor to address the inherent trade-off between sensitivity and hysteresis in tactile sensors when using soft materials. We discovered that piezoresistive sensors made using an array of three-dimensional (3D) metallic annular cracks on polymeric microstructures possess high sensitivities (> 107 Ω ⋅ kPa-1), low hysteresis (2.99 ± 1.37%) over a wide pressure range (0-20 kPa), and fast response (400 Hz). We demonstrate that TRACE sensors can accurately detect and measure the pulse wave velocity (PWV) when skin mounted. Moreover, we show that these tactile sensors when arrayed enabled fast reliable one-touch surface texture classification with neuromorphic encoding and deep learning algorithms.

Yao Haicheng, Yang Weidong, Cheng Wen, Tan Yu Jun, See Hian Hian, Li Si, Ali Hashina Parveen Anwar, Lim Brian Z H, Liu Zhuangjian, Tee Benjamin C K

2020-Sep-28

electronic skin, machine learning, robotics, sensor, wearable

General General

Placing language in an integrated understanding system: Next steps toward human-level performance in neural language models.

In Proceedings of the National Academy of Sciences of the United States of America

Language is crucial for human intelligence, but what exactly is its role? We take language to be a part of a system for understanding and communicating about situations. In humans, these abilities emerge gradually from experience and depend on domain-general principles of biological neural networks: connection-based learning, distributed representation, and context-sensitive, mutual constraint satisfaction-based processing. Current artificial language processing systems rely on the same domain general principles, embodied in artificial neural networks. Indeed, recent progress in this field depends on query-based attention, which extends the ability of these systems to exploit context and has contributed to remarkable breakthroughs. Nevertheless, most current models focus exclusively on language-internal tasks, limiting their ability to perform tasks that depend on understanding situations. These systems also lack memory for the contents of prior situations outside of a fixed contextual span. We describe the organization of the brain's distributed understanding system, which includes a fast learning system that addresses the memory problem. We sketch a framework for future models of understanding drawing equally on cognitive neuroscience and artificial intelligence and exploiting query-based attention. We highlight relevant current directions and consider further developments needed to fully capture human-level language understanding in a computational system.

McClelland James L, Hill Felix, Rudolph Maja, Baldridge Jason, Schütze Hinrich

2020-Sep-28

artificial intelligence, cognitive neuroscience, deep learning, natural language understanding, situation models

General General

Pre-trained language model augmented adversarial training network for Chinese clinical event detection.

In Mathematical biosciences and engineering : MBE

Clinical event detection (CED) is a hot topic and essential task in medical artificial intelligence, which has attracted the attention from academia and industry over the recent years. However, most studies focus on English clinical narratives. Owing to the limitation of annotated Chinese medical corpus, there is a lack of relevant research about Chinese clinical narratives. The existing methods ignore the importance of contextual information in semantic understanding. Therefore, it is urgent to research multilingual clinical event detection. In this paper, we present a novel encoder-decoder structure based on pre-trained language model for Chinese CED task, which integrates contextual representations into Chinese character embeddings to assist model in semantic understanding. Compared with existing methods, our proposed strategy can help model harvest a language inferential skill. Besides, we introduce the punitive weight to adjust the proportion of loss on each category for coping with class imbalance problem. To evaluate the effectiveness of our proposed model, we conduct a range of experiments on test set of our manually annotated corpus. We compare overall performance of our proposed model with baseline models on our manually annotated corpus. Experimental results demonstrate that our proposed model achieves the best precision of 83.73%, recall of 86.56% and F1-score of 85.12%. Moreover, we also evaluate the performance of our proposed model with baseline models on minority category samples. We discover that our proposed model obtains a significant increase on minority category samples.

Zhang Zhi Chang, Zhang Min Yu, Zhou Tong, Qiu Yan Long

2020-Mar-24

** Chinese clinical event detection , Chinese clinical narratives , adversarial training network , class imbalance problem , medical artificial intelligence , pre-trained language model , semantic understanding , transfer learning **