Text Labels Research Articles

A Review of Patient-Reported Outcome Labeling of FDA-Approved New Drugs (2016-2020): Counts, Categories, and Comprehensibility

ObjectivesA review of new drug approvals (NDAs) by the US Food and Drug Administration (FDA) for 2006 to 2015 showed that approximately 20% of new drugs had labeling based on patient-reported outcomes (PROs). The purpose of this study was to review labeling text based on PRO endpoints for NDAs from 2016 to 2020, with a special focus on the comprehensibility of such statements when included. MethodsWe reviewed drug approval reports on the Drugs@FDA web page of the FDA website to determine the number of NDAs from 2016 to 2020. For all identified NDAs, drug approval package and product labels were reviewed. NDAs from 2016 to 2020 were grouped by disease category as per International Classification of Diseases 10th Revision. Data were summarized for diseases that traditionally rely on PROs for evaluating treatment benefit (PRO dependent) and for diseases that traditionally do not rely on PROs (non-PRO dependent). Results were compared with NDAs from 2006 to 2010. ResultsNDAs amounting to 228 were identified from 2016 to 2020, 26.3% of which had labeling statements based on PRO endpoints. From 2006 to 2015 and from 2016 to 2020, PRO labeling statements were included in 46.5% (46 of 99) and 50.0% (47 of 94), respectively, of NDAs for PRO-dependent new molecular entities and in 6.0% (12 of 199) and 9.7% (13 of 199), respectively, of NDAs for non–PRO-dependent new molecular entities. Comprehensibility of labeling statements based on PRO endpoints was judged to be complex in 56.7% of product labels. ConclusionsThe increase in labeling text based on PRO endpoints in product labels is encouraging. However, there is room for improvement on the comprehensibility of labeling statements based on PRO endpoints.

EANDC: An explainable attention network based deep adaptive clustering model for mental health treatment

Internet-delivered Psychological Treatment (IDPT) has been shown to be an effective method for improving psychological disorders. Natural language processing (NLP) requires an appropriate set of linguistic features for word representation and emotion segmentation. For psychological applications, models must be trained on extensive and diverse datasets to achieve expert-level performance. Labeling psychological texts authorized by patients is challenging because emotional biases can lead to incorrect segmentation of emotions and labeling emotional data is time consuming. In this paper, we propose an assistance tool for psychologists to explore the emotional aspects of mentally ill individuals. We first use an NLP-based method to create emotional lexicon embeddings, and then apply attention-based deep clustering. The learned representation is then used to visualize the emotional aspect of the text authorized by patients. We expand the patient authored text using synonymous semantic expansion. A latent semantic representation based on context is clustered using EANDC, which is a Explainable Attention Network-based Deep adaptive Clustering model. We use similarity metrics to select a subset of the text and then improve the explainability of learning using a curriculum-based optimization method. The experimental results show that synonym expansion based on the emotion lexicon increases accuracy without affecting the results. The attention method with bidirectional LSTM architecture achieved 0.81 ROC in a blind test. The self-learning based embedding then visualizes the weighted attention words and helps the psychiatrist to improve his explanatory power of the qualitative match for clinical notes and the remedy. The method helps in labeling text and improves the recognition rate of symptoms of mental disorders.

Accessibility to digital technology: Virtual barriers, real opportunities

ABSTRACT The potential of digital technology to assist persons with disabilities has always been known. The capabilities of digital devices have been improving so impressively for so long, that the assumption that in parallel the same is happening with accessibility is common. Unfortunately, accessibility for persons with disabilities is neither certain nor constant, and in fact, a conscious and systemic effort is required to ensure that the potential of digital technologies for inclusion is realized. Digital accessibility is best understood as a chain of dependencies where training, hardware, software, content, and standards must work together harmoniously, and each of these elements must be understood as a dynamic process. For example, smartphones can be incompatible with hearing aids required by the deaf, touch screens too sensitive for those with motor impairments, and web pages often lack the text labels needed by screen reading software used by the blind. Even if each of these examples is fixed, the accessibility may be short lived if the production process behind that hardware or software was not corrected, as the digital world is constantly being updated. Training, hardware design, software development, content production, and standards definition processes, must be pursued taking accessibility and affordability into account.

QuickFigures: A toolkit and ImageJ PlugIn to quickly transform microscope images into scientific figures.

Publications involving fluorescent microscopy images generally contain many panels with split channels, merged images, scale bars and label text. Similar layouts of panels are used when displaying other microscopy images, electron micrographs, photographs, and other images. Assembling and editing these figures with even spacing, consistent font, text position, accurate scale bars, and other features can be tedious and time consuming. In order to save time, I have created a toolset and ImageJ Plugin called QuickFigures. QuickFigures includes many helpful features that streamline the process of creating, aligning, and editing scientific figures. Those features include tools that automatically create split channel figures from a region of interest (“Quick Figure” button and “Inset Tool”), layouts that make it easy to rearrange panels, multiple tools to align objects, and “Figure Format” menu options that help a user ensure that large numbers of figures have consistent appearance. QuickFigures was compared to previous tools by measuring the amount of time needed for a user to create a figure using each software (QuickFigures, OMERO.figure. EZFig, FigureJ and PowerPoint). QuickFigures significantly reduced the amount of time required to create a figure. The toolsets were also compared by checking each software against a list of features. QuickFigures had the most extensive set of features. Therefore, QuickFigures is an advantageous alternative to traditional methods of constructing scientific figures. After a user has saved time by creating their work in QuickFigures, the figures can be exported to a variety of formats including PowerPoint, PDF, SVG, PNG, TIFF and Adobe Illustrator. Export was successfully tested for each file format and object type. Exported objects and text are editable in their target software, making them suitable for sharing with collaborators. The software is free, open source and can be installed easily.

A Partially-Sorted Concentric Layout for Efficient Label Localization in Augmented Reality.

A common approach for Augmented Reality labeling is to display the label text on a flag planted into the real world element at a 3D anchor point. When there are more than just a few labels, the efficiency of the interface decreases as the user has to search for a given label sequentially. The search can be accelerated by sorting the labels alphabetically, but sorting all labels results in long and intersecting leader lines from the anchor points to the labels. This paper proposes a partially-sorted concentric label layout that leverages the search efficiency of sorting while avoiding the label display problems of long or intersecting leader lines. The labels are partitioned into a small number of sorted sequences displayed on circles of increasing radii. Since the labels on a circle are sorted, the user can quickly search each circle. A tight upper bound derived from circular permutation theory limits the number of circles and thereby the complexity of the label layout. For example, 12 labels require at most three circles. When the application allows it, the labels are presorted to further reduce the number of circles in the layout. The layout was tested in a user study where it significantly reduced the label searching time compared to a conventional single-circle layout.

Mid-air haptics for shape recognition of virtual objects

This paper presents an experiment in which participants had to discriminate three mid-air haptic shapes (circle, square, point) by reporting whether the haptic stimulus (e.g., circle on the palm of the hand) was compatible with an image (e.g., a circle) or a word (e.g., “circle”) displayed on a screen. Results indicate that only the “point” stimulus was appreciably recognised and discriminated in terms of accuracy and time needed for the identification. Accuracy increased with repetition, and response time decreased, suggesting a learning effect. The comparison between visual and textual labels shows that for the haptic point stimulus there is no significant difference but a tendency to have greater accuracy with images than with texts, while the opposite result is found for the circle stimulus. This outcome suggests the need for new experiments focused of the effect of visual/textual labels to make the recognition/discrimination tasks of haptic stimuli easier. Practitioner Summary: Three haptic shapes were presented with images or texts, matching or not the stimuli. The point was easy to recognise, while the circle and the square were difficult to discriminate against each other. Visual/textual labels bring contradictory results for different shapes.

Disabilities and the Digital Divide: Assessing Web Accessibility, Readability, and Mobility of Popular Health Websites

With recognition of (1) unmet health information needs of persons with disabilities; and (2) anticipated growth of the persons with disabilities population, we seek to better understand the relative general accessibility of health information on popular, highly ranked health websites. That is, instead of examining the efficacy and impact of specific types of health message appeals, the present investigation assesses the functional performance of specific content platforms of health websites. We examine the components of accessing and understanding health information through the lens of web and language accessibility, readability, and mobility. A total of 139 health-focused websites were randomly sampled from the Alexa® “Top Sites” global ranking system which uses web traffic data to rank the most popular websites. The sample was processed using computer aided software (e.g., WEB AIM, Web FX & Google Test My Site.) We found 91.3% of sampled home pages had detectable accessibility failures. The most common accessibility failures were low contrast, empty links, missing ALT text, empty buttons, and missing form labels. Most websites offered neither an accessibility policy statement nor an option for automated language translation service, likely impacting those with limited English proficiency (LEP). The implications for these findings are reviewed with discussion and limitations provided.

An ensemble model for classifying idioms and literal texts using BERT and RoBERTa

An idiom is a common phrase that means something other than its literal meaning. Detecting idioms automatically is a serious challenge in natural language processing (NLP) domain applications like information retrieval (IR), machine translation and chatbot. Automatic detection of Idioms plays an important role in all these applications. A fundamental NLP task is text classification, which categorizes text into structured categories known as text labeling or categorization. This paper deals with idiom identification as a text classification task. Pre-trained deep learning models have been used for several text classification tasks; though models like BERT and RoBERTa have not been exclusively used for idiom and literal classification. We propose a predictive ensemble model to classify idioms and literals using BERT and RoBERTa, fine-tuned with the TroFi dataset. The model is tested with a newly created in house dataset of idioms and literal expressions, numbering 1470 in all, and annotated by domain experts. Our model outperforms the baseline models in terms of the metrics considered, such as F-score and accuracy, with a 2% improvement in accuracy.

Off the Record

Off the Record

Sequential alignment attention model for scene text recognition

Scene text recognition has been a hot research topic in computer vision due to its various applications. The state-of-the-art solutions usually depend on the attention-based encoder-decoder framework that learns the mapping between input images and output sequences in a purely data-driven way. Unfortunately, there often exists severe misalignment between feature areas and text labels in real-world scenarios. To address this problem, this paper proposes a sequential alignment attention model to enhance the alignment between input images and output character sequences. In this model, an attention gated recurrent unit (AGRU) is first devised to distinguish the text and background regions, and further extract the localized features focusing on sequential text regions. Furthermore, CTC guided decoding strategy is integrated into the popular attention-based decoder, which not only helps to boost the convergence of the training but also enhances the well-aligned sequence recognition. Extensive experiments on various benchmarks, including the IIIT5k, SVT, and ICDAR datasets, show that our method substantially outperforms the state-of-the-art methods.

Creating Tactile Educational Materials for the Visually Impaired and Blind Students Using AI Cloud Computing

There are 24.5 million visually impaired and blind (VIB) students who have limited access to educational materials due to cost or availability. Although advancement in technology is prevalent, providing individualized learning using technology remains a challenge without the proper tools or experience. The TacPic system was developed as an online platform to create tactile educational materials (TEM) based on the image inputs of users who do not have prior experience in tactile photo development or 3D printing. The TacPic system allows the users to simply upload images to a website and uses AI cloud computing on the Amazon Web Services platform. First, it segments and labels the images. Then, the text label is converted into braille words. Subsequently, surface rendering and consolidation of the image and text is performed, before it is converted into a single file that is ready for 3D printing. Currently, the types of TEM that can be created are tactile flashcards, tactile maps, and tactile peg puzzles, which can be developed within a few hours. This is in contrast to a development period of weeks using traditional methods. Furthermore, the tactile educational materials were tested by two VIB teachers and six VIB students. It was found that those who are congenitally blind need more time to identify the object and rely more on the braille labels compared to students who became blind at a later age. Teachers also suggested producing TEM that use simpler images, and TEM that are suitable for both sighted and VIB students. In conclusion, the researchers successfully developed a platform that allows more educators or parents to develop personalized and individualized TEM. In the future, further optimization of the algorithms to improve segmentation and the inclusion of other features, such as color, could be undertaken. Finally, new printing materials and methods are needed to improve printing efficiency.

Smart Glass for Visually Challenged Peoples to Read the Books using Raspberry Pi

The objective of this paper is to presents new design on assistive smart glasses for visually impaired. The objective is to assist in multiple daily tasks using the advantage of wearable design format. The proposed method is a camera based assistive text reading to help to blind in person in reading the text present on the text labels, printed notes and products in their own respective languages. It combines the concept of Optical Character Recognition (OCR), text to Speech Synthesizer (TTS) and translator in Raspberry pi. Optical character recognition (OCR) is the identification of printed characters using photoelectric devices and computer software. It converts images of typed, handwritten or printed text into machine encoded text from scanned document or from subtitle text superimposed on an image. Text-to-Speech conversion is a method that scans and reads any language letters and numbers that are in the image using OCR technique and then translates it into any desired language and at last it gives audio output of the translated text. The audio output is heard through the raspberry pi's audio jack using speakers or earphones.

AEMF: An Attention-Based Efficient and Multifeature Fast Text Detector

The label from industrial commodity packaging usually contains important data, such as production date, manufacturer, and other commodity-related information. As such, those labels are essential for consumers to purchase goods, help commodity supervision, and reveal potential product safety problems. Consequently, packaging label detection, as the prerequisite for product label identification, becomes a very useful application, which has achieved promising results in the past decades. Yet, in complex industrial scenarios, traditional detection methods are often unable to meet the requirements, which suffer from many problems of low accuracy and efficiency. In this paper, we propose a multifeature fast and attention-based algorithm using a combination of area suggestion and semantic segmentation. This algorithm is an attention-based efficient and multifeature fast text detector (termed AEMF). The proposed approach is formed by fusing segmentation branches and detection branches with each other. Based on the original algorithm that can only detect text in any direction, it is possible to detect different shapes with a better accuracy. Meanwhile, the algorithm also works better on long-text detection. The algorithm was evaluated using ICDAR2015, CTW1500, and MSRA-TD500 public datasets. The experimental results show that the proposed multifeature fusion with self-attention module makes the algorithm more accurate and efficient than existing algorithms. On the MSRA-TD500 dataset, the AEMF algorithm has an F-measure of 72.3% and a frame per second (FPS) of 8. On the CTW1500 dataset, the AEMF algorithm has an F-measure of 62.3% and an FPS of 23. In particular, the AEMF algorithm has achieved an F-measure of 79.3% and an FPS of 16 on the ICDAR2015 dataset, demonstrating the excellent performance in detecting label text on industrial packaging.

Natural language-based detection of semantic execution anomalies in event logs

Anomaly detection in process mining aims to recognize outlying or unexpected behavior in event logs for purposes such as the removal of noise and identification of conformance violations. Existing techniques for this task are primarily frequency-based, arguing that behavior is anomalous because it is uncommon. However, such techniques ignore the semantics of recorded events and, therefore, do not take the meaning of potential anomalies into consideration. In this work, we overcome this caveat and focus on the detection of anomalies from a semantic perspective, arguing that anomalies can be recognized when process behavior does not make sense. To achieve this, we propose an approach that exploits the natural language associated with events. Our key idea is to detect anomalous process behavior by identifying semantically inconsistent execution patterns. To detect such patterns, we first automatically extract business objects and actions from the textual labels of events. We then compare these against a process-independent knowledge base. By populating this knowledge base with patterns from various kinds of resources, our approach can be used in a range of contexts and domains. We demonstrate the capability of our approach to successfully detect semantic execution anomalies through an evaluation based on a set of real-world and synthetic event logs and show the complementary nature of semantics-based anomaly detection to existing frequency-based techniques.

Joint Representations of Texts and Labels with Compositional Loss for Short Text Classification

Short text classification is an important foundation for natural language processing (NLP) tasks. Though, the text classification based on deep language models (DLMs) has made a significant headway, in practical applications however, some texts are ambiguous and hard to classify in multi-class classification especially, for short texts whose context length is limited. The mainstream method improves the distinction of ambiguous text by adding context information. However, these methods rely only the text representation, and ignore that the categories overlap and are not completely independent of each other. In this paper, we establish a new general method to solve the problem of ambiguous text classification by introducing label embedding to represent each category, which makes measurable difference between the categories. Further, a new compositional loss function is proposed to train the model, which makes the text representation closer to the ground-truth label and farther away from others. Finally, a constraint is obtained by calculating the similarity between the text representation and label embedding. Errors caused by ambiguous text can be corrected by adding constraints to the output layer of the model. We apply the method to three classical models and conduct experiments on six public datasets. Experiments show that our method can effectively improve the classification accuracy of the ambiguous texts. In addition, combining our method with BERT, we obtain the state-of-the-art results on the CNT dataset.

Inter-Rater Reliability of Unstructured Text Labeling: Artificially vs. Naturally Intelligent Approaches.

Unstructured medical text labeling technologies are expected to be highly demanded since the interest in artificial intelligence and natural language processing arises in the medical domain. Our study aimed to assess the agreement between experts who judged on the fact of pulmonary embolism (PE) in neurosurgical cases retrospectively based on electronic health records and assess the utility of the machine learning approach to automate this process. We observed a moderate agreement between 3 independent raters on PE detection (Light's kappa = 0.568, p = 0). Labeling sentences with the method we proposed earlier might improve the machine learning results (accuracy = 0.97, ROC AUC = 0.98) even in those cases that could not be agreed between 3 independent raters. Medical text labeling techniques might be more efficient when strict rules and semi-automated approaches are implemented. Machine learning might be a good option for unstructured text labeling when the reliability of textual data is properly addressed. This project was supported by the RFBR grant 18-29-22085.

TSRGAN: Real-world text image super-resolution based on adversarial learning and triplet attention

The text in a low-resolution (LR) image is usually hard to read. Super-resolution (SR) is an intuitive solution to this issue. Existing single image super-resolution (SISR) models are mainly trained on synthetic datasets whose LR images are obtained by performing bicubic interpolation or gaussian blur on high-resolution (HR) images. However, these models can hardly generalize to practical scenarios because real-world LR images are more difficult to super-resolve. The newly proposed TextZoom dataset is the first dataset for real-world text image super-resolution. We propose a new model termed TSRGAN trained on this dataset. First, a discriminator is designed to prevent the SR network from generating over-smoothed images. Second, we introduce triplet attention into the SR network for better representational ability. Moreover, besides L2 loss and adversarial loss, wavelet loss is incorporated to help reconstruct sharper character edges. Since TextZoom provides text labels, the recognition accuracy of scene text recognition (STR) model can be used to evaluate the quality of SR images. It can reflect the performance of text image SR models better than traditional SR evaluation metrics such as PSNR and SSIM. Comprehensive experiments show the superiority of our TSRGAN. Compared with the state-of-the-art method, the proposed TSRGAN improves the average recognition accuracy of ASTER, MORAN and CRNN by 0.8%, 1.5% and 3.2% on TextZoom respectively.

The Smart Assistant for Library Management and Book Reader for Blind People Using Raspberry Pi

This system proposes a camera-based assistive text reading framework to help blind persons ad text labels and product packaging from hand-held objects in their daily lives. To isolate the object from cluttered backgrounds or other surrounding objects in the camera view, we rest propose an ancient and active motion-based method to dine a Region Of Interest (ROI) in the video by asking the user to shake the object. In the extracted ROI, text localization and recognition are conducted to acquire text information. To automatically localize the text regions from the object ROI, we propose a novel text localization algorithm by learning gradient features of stroke orientations and distributions of edge pixels in an Ad boost model. Text characters in the localized text regions are then binaries and recognized by the shelf Optical Character Recognition (OCR) software.

Measuring destination image: a novel approach based on visual data mining. A methodological proposal and an application to European islands

Availability of User Generated Content and the development of Big Data and machine learning algorithms have paved the way to collecting and analysing great volumes of data. We scan imagery data from traveling-related posts on Instagram to identify the key features of the destination image and of its dynamics. Specifically, we exploit a newly introduced Visual Object Recognition tool (Google Cloud Vision) to convert into textual labels the content of about 860,000 travel-related pictures posted on Instagram in Summer 2019 for several European islands. The output, a vector of labels’ frequencies on a very fine-grained scale, is used to proxy the destination image at different points in time. We then introduce the Index of Distance in Destination Image, a metric built on the pictures’ labels ranking, and aimed at providing a quantitative measure of (dis)similarity between destination images. We show that the analysis of labels and the index are fit to compare destinations cross-sectionally and over time, providing a useful tool for researchers, marketers and DMOs. We also deliver evidence on how external shocks (like extreme events linked to climate change) or the organization of events modify the cognitive sphere of the destination image, with repercussions on activities undertaken by tourists and relevant implications for local policies. • A Visual Object Recognition tool is applied to Instagram pictures posted by tourists. • The output is a vector of labels frequency for each destination and point in time. • An Index of Distance in Destination Images is proposed and applied to the output. • The metrics allows the investigation of differences and similarities across destinations. • Variation over time in destination image triggered by events is also found.

Multi-View Data Integration Methods for Radiotherapy Structure Name Standardization.

Simple SummaryStructure names associated with radiotherapy treatments need standardization to develop data pipelines enabling personalized treatment plans. Automatic classification of structure names based on the currently available TG-263 nomenclature can help with data aggregation from both retrospective and future data sources. The aim of our proposed machine learning-based data integration methods is to achieve highly accurate structure name classification to automate the data aggregation process. Our multi-view models can overcome the challenges of integrating different data types associated with radiotherapy structures, such as the physician-given text labels and geometric or image data. The models exhibited high accuracy when tested on multi-center and multi-institutional lung and prostate cancer patients data and outperformed the models built on any single data type. This highlights the importance of combining different types of data in building generalizable models for structure name standardization.Standardization of radiotherapy structure names is essential for developing data-driven personalized radiotherapy treatment plans. Different types of data are associated with radiotherapy structures, such as the physician-given text labels, geometric (image) data, and Dose-Volume Histograms (DVH). Prior work on structure name standardization used just one type of data. We present novel approaches to integrate complementary types (views) of structure data to build better-performing machine learning models. We present two methods, namely (a) intermediate integration and (b) late integration, to combine physician-given textual structure name features and geometric information of structures. The dataset consisted of 709 prostate cancer and 752 lung cancer patients across 40 radiotherapy centers administered by the U.S. Veterans Health Administration (VA) and the Department of Radiation Oncology, Virginia Commonwealth University (VCU). We used randomly selected data from 30 centers for training and ten centers for testing. We also used the VCU data for testing. We observed that the intermediate integration approach outperformed the models with a single view of the dataset, while late integration showed comparable performance with single-view results. Thus, we demonstrate that combining different views (types of data) helps build better models for structure name standardization to enable big data analytics in radiation oncology.