MDIW-13: a New Multi-Lingual and Multi-Script Database and Benchmark for Script Identification

Script identification is a key step in document analysis and recognition in multilingual environments. This study proposed a new dataset for script identification algorithms, containing images of ancient documents in 12 different ethnic scripts, including Chinese script, Naxi Dongba script, Yi script, Shui script, Tangut script, ancient Zhuang script, ancient Buyi script, Tibetan script, Dai script, Chagatai script, Mongolian script, and Manchu script. Focusing on the high accuracy required for ancient script identification, this study proposed a method named multi-attention ghost pyramid fusion network (MAGPNet). MAGPNet consists of a feature extraction network, a channel feature pyramid, and a Multi-Headed Self-Attention Bottleneck Block. The feature extraction network utilizes lightweight convolutional modules and parameter-free attention modules to enhance MAGPNet's feature extraction capability while maintaining a lighter structure. The channel feature pyramid increases the model's robustness in processing ancient documents of different scales. The Multi-Headed Self-Attention Bottleneck Block, by introducing a Multi-Headed Self-Attention, focuses on effective features. Experiments demonstrate that MAGPNet achieves a 99.97% accuracy rate on the multilingual ancient document image script identification dataset, maintaining excellent classification performance across multiple datasets.

Script-based text document classification is an important field of research in the context of multilingual textual document processing. But, all script identification techniques available in the literature so far do not consider handwritten documents. Variations in the writing style, character size, inter-line and inter-word spacings, etc. make the recognition process difficult and unreliable when these script identification algorithms, more specifically visual appearance based approaches, are applied directly on hand-written documents. Therefore, in this paper, we propose to preprocess the input document images so as to compensate for the variations due to writing style and thereby making them suitable for analysis on the basis of their visual appearances. Accordingly, we apply denoising, thinning, pruning, m-connectivity and text size normalization in sequence. Multi-channel Gabor filtering is used to extract texture features that characterize the visual appearances of the document images. Experimental result proves the potentiality of our proposed method of script identification for hand-written text document classification.

Script identification is an important area in handwriting document image analysis field. The script identification at word level on documents written in multiple scripts is an open challenge for the scientific community and a real concern in countries with multiple official languages, e. G. The country like India. Such documents usually contain two scripts: the most of the document are written in the regional script while some words, acronyms or numbers are written in Roman script. In this case a word or even a character level script identification is required to locate the second script characters in the document. Here the major problem is the few script descriptors available for the script estimation which convey high error rates. The literatures try to address this problem by looking for more efficient descriptors. In this paper we propose a Multiple Training - One Test technique to alleviate this problem. Several classifiers are trained, each one with words of similar amount of information. A scale invariable word information index is defined for this sake. To identify the script of a query word, its word information index is worked out, and its script is identified with the most appropriate classifier. Accuracy improvements has been obtained with this promising technique, especially for the shorten words.

In this paper, we propose a new scheme for script and nature identification. The objective is to discriminate between machine-printed/handwritten and Latin/Arabic scripts at word level. It is relatively a complex task due to possible use of multi-fonts and sizes, complexity and variation in handwriting. In the proposed script identification system, we extract features from word images using Co-occurrence Matrix of Oriented Gradients (Co-MOG). The classification is done using different classifiers. Extensive experimentation has been carried on 24000 words, extracted from standard databases. An average identification accuracy of 99.85% is achieved by k Nearest Neighbors (k-NN) classifier which clearly outperforms results of some existing systems.

Script identification is the first step in most multilingual text-processing systems. To improve the time efficiency of script identification algorithms, whether there is content written in a certain script in the text is first determined; if so, the content written in that script is then obtained. Then, it is determined whether the total length of the texts corresponding to the identified scripts is equal to the original text length; if so, the script identification process ends. Finally, considering the frequencies of various scripts on the Internet, those that are more common are prioritized during script identification. Based on these three approaches, an improved script identification algorithm was designed. A comparison experiment was conducted using sentence-level text corpora in 263 languages written in 26 scripts. The testing times of the newly proposed method were reduced by 9.35-fold, while the F1 score for script identification was slightly higher than those reported in our earlier studies. The method proposed in this study effectively improves the time efficiency of script identification algorithms.

Script identification is an appealing research interest in the field of document image analysis during the last few decades. The accurate recognition of the script is paramount to many post-processing steps such as automated document sorting, machine translation and searching of text written in a particular script in multilingual environment. For automatic processing of such documents through Optical Character Recognition (OCR) software, it is necessary to identify different script words of the documents before feeding them to the OCR of individual scripts. In this paper, a robust word-level handwritten script identification technique has been proposed using texture based features to identify the words written in any of the seven popular scripts namely, Bangla, Devanagari, Gurumukhi, Malayalam, Oriya, Telugu, and Roman. The texture based features comprise of a combination of Histograms of Oriented Gradients (HOG) and Moment invariants. The technique has been tested on 7000 handwritten text words in which each script contributes 1000 words. Based on the identification accuracies and statistical significance testing of seven well-known classifiers, Multi-Layer Perceptron (MLP) has been chosen as the final classifier which is then tested comprehensively using different folds and with different epoch sizes. The overall accuracy of the system is found to be 94.7% using 5-fold cross validation scheme, which is quite impressive considering the complexities and shape variations of the said scripts. This is an extended version of the paper described in (Singh et al., 2014).

Script identification at the word level is challenging because of complex backgrounds and low resolution of video. The presence of graphics and scene text in video makes the problem more challenging. In this paper, we employ gradient angle segmentation on words from video text lines. This paper presents new Gradient-Angular-Features (GAF) for video script identification, namely, Arabic,Chinese, English, Japanese, Korean and Tamil. This work enables us to select an appropriate OCR when the frame has words of multi-scripts. We employ gradient directional features for segmenting words from video text lines. For each segmented word, we study the gradient information in effective ways to identify text candidates. The skeleton of the text candidates is analyzed to identify Potential Text Candidates (PTC) by filtering out unwanted text candidates. We propose novel GAF for the PTC to study the structure of the components in the form of cursiveness and softness. The histogram operation on the GAF is performed in different ways to obtain discriminative features. The method is evaluated on 760 words of six scripts having low contrast, complex background, different font sizes, etc. in terms of the classification rate and is compared with an existing method to show the effectiveness of the method. We achieve 88.2% average classification rate.

In this paper we employ appearance based models for document script identification. They are employed to identify scripts at both paragraph and word level. Elaborate experimentation has been conducted which has revealed that they are robust enough to handle highly confusing scripts and their performance does not degrade drastically even in the presence of noise. A generic script identification has been attempted, to identify both Asian and European scripts by considering a dataset of twenty different languages.

Classifying social media (SM) messages into relevant or irrelevant categories is challenging due to data sparsity, imbalance, and ambiguity. This study aims to improve Intelligent Transport Systems (ITS) by enhancing short text classification of traffic-related SM data. Deep learning methods such as RNNs, CNNs, and BERT are effective at capturing context, but they can be computationally expensive, struggle with very short texts, and perform poorly with rare words. On the other hand, transfer learning leverages pre-trained knowledge but may be biased towards the pre-training domain. To address these challenges, we propose DLCTC, a novel system combining character-level, word-level, and context features with BiLSTM and TextCNN-based attention. By utilizing external knowledge, DLCTC ensures an accurate understanding of concepts and abbreviations in traffic-related short texts. BiLSTM captures context and term correlations; TextCNN captures local patterns. Multi-level attention focuses on important features across character, word, and concept levels. Experimental studies demonstrate DLCTC’s effectiveness over well-known short-text classification approaches based on CNN, RNN, and BERT.

The clinical utility of regional strain measurements in echocardiography is challenged by suboptimal reproducibility. In this study, we aimed to evaluate the test-retest reproducibility of regional longitudinal strain (RLS) per coronary artery perfusion territory (RLSTerritory) and basal-to-apical level of the left ventricle (RLSLevel), measured by a novel fully automated deep learning (DL) method based on point tracking. We measured strain in a dual-centre test-retest data set that included 40 controls and 40 patients with suspected non-ST elevation acute coronary syndrome. Two consecutive echocardiograms per subject were recorded by different operators. The reproducibility of RLSTerritory and RLSLevel measured by the DL method and by three experienced observers using semi-automatic software (2D Strain, EchoPAC, GE HealthCare) was evaluated as minimal detectable change (MDC). The DL method had MDC for RLSTerritory and RLSLevel ranging from 3.6 to 4.3%, corresponding to a 33-35% improved reproducibility compared with the inter- and intraobserver scenarios (MDC 5.5-6.4% and 4.9-5.4%). Furthermore, the DL method had a lower variance of test-retest differences for both RLSTerritory and RLSLevel compared with inter- and intraobserver scenarios (all P < 0.001). Bland-Altman analyses demonstrated superior reproducibility by the DL method for the whole range of strain values compared with the best observer scenarios. The feasibility of the DL method was 93% and measurement time was only 1 s per echocardiogram. The novel DL method provided fully automated measurements of RLS, with improved test-retest reproducibility compared with semi-automatic measurements by experienced observers. RLS measured by the DL method has the potential to advance patient care through a more detailed, more efficient, and less user-dependent clinical assessment of myocardial function.

Less than 10% of patients with cancer have detectable pathogenic germline alterations, which may be partially due to incomplete pathogenic variant detection. To evaluate if deep learning approaches identify more germline pathogenic variants in patients with cancer. A cross-sectional study of a standard germline detection method and a deep learning method in 2 convenience cohorts with prostate cancer and melanoma enrolled in the US and Europe between 2010 and 2017. The final date of clinical data collection was December 2017. Germline variant detection using standard or deep learning methods. The primary outcomes included pathogenic variant detection performance in 118 cancer-predisposition genes estimated as sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The secondary outcomes were pathogenic variant detection performance in 59 genes deemed actionable by the American College of Medical Genetics and Genomics (ACMG) and 5197 clinically relevant mendelian genes. True sensitivity and true specificity could not be calculated due to lack of a criterion reference standard, but were estimated as the proportion of true-positive variants and true-negative variants, respectively, identified by each method in a reference variant set that consisted of all variants judged to be valid from either approach. The prostate cancer cohort included 1072 men (mean [SD] age at diagnosis, 63.7 [7.9] years; 857 [79.9%] with European ancestry) and the melanoma cohort included 1295 patients (mean [SD] age at diagnosis, 59.8 [15.6] years; 488 [37.7%] women; 1060 [81.9%] with European ancestry). The deep learning method identified more patients with pathogenic variants in cancer-predisposition genes than the standard method (prostate cancer: 198 vs 182; melanoma: 93 vs 74); sensitivity (prostate cancer: 94.7% vs 87.1% [difference, 7.6%; 95% CI, 2.2% to 13.1%]; melanoma: 74.4% vs 59.2% [difference, 15.2%; 95% CI, 3.7% to 26.7%]), specificity (prostate cancer: 64.0% vs 36.0% [difference, 28.0%; 95% CI, 1.4% to 54.6%]; melanoma: 63.4% vs 36.6% [difference, 26.8%; 95% CI, 17.6% to 35.9%]), PPV (prostate cancer: 95.7% vs 91.9% [difference, 3.8%; 95% CI, -1.0% to 8.4%]; melanoma: 54.4% vs 35.4% [difference, 19.0%; 95% CI, 9.1% to 28.9%]), and NPV (prostate cancer: 59.3% vs 25.0% [difference, 34.3%; 95% CI, 10.9% to 57.6%]; melanoma: 80.8% vs 60.5% [difference, 20.3%; 95% CI, 10.0% to 30.7%]). For the ACMG genes, the sensitivity of the 2 methods was not significantly different in the prostate cancer cohort (94.9% vs 90.6% [difference, 4.3%; 95% CI, -2.3% to 10.9%]), but the deep learning method had a higher sensitivity in the melanoma cohort (71.6% vs 53.7% [difference, 17.9%; 95% CI, 1.82% to 34.0%]). The deep learning method had higher sensitivity in the mendelian genes (prostate cancer: 99.7% vs 95.1% [difference, 4.6%; 95% CI, 3.0% to 6.3%]; melanoma: 91.7% vs 86.2% [difference, 5.5%; 95% CI, 2.2% to 8.8%]). Among a convenience sample of 2 independent cohorts of patients with prostate cancer and melanoma, germline genetic testing using deep learning, compared with the current standard genetic testing method, was associated with higher sensitivity and specificity for detection of pathogenic variants. Further research is needed to understand the relevance of these findings with regard to clinical outcomes.

Deep learning methods have become popular among researchers in the field of fault detection. However, their performance depends on the availability of big datasets. To overcome this problem researchers started applying transfer learning to achieve good performance from small available datasets, by leveraging multiple prediction models over similar machines and working conditions. However, the influence of negative transfer limits their application. Negative transfer among prediction models increases when the environment and working conditions are changing continuously. To overcome the effect of negative transfer, we propose a novel deep transfer learning method, coined deep boosted transfer learning, for wind turbine gearbox fault detection that prevents negative transfer and only focuses on relevant information from the source machine. The proposed method is an instance-based deep transfer learning method that updates the weights of the source and the target machine training samples separately. The weights of different source training samples are gradually decreased to reduce the impact on the final model. The proposed method is verified by the Case Western Reserve University bearing and real field wind farm datasets. The results show that the proposed method ignores negative transfer and achieves higher accuracy compared to standard deep learning and deep transfer learning methods.

This review paper examines the current state of lung disease diagnosis based on deep learning (DL) methods. Lung diseases, such as Pneumonia, TB, Covid-19, and lung cancer, are significant causes of morbidity and mortality worldwide. Accurate and timely diagnosis of these diseases is essential for effective treatment and improved patient outcomes. DL methods, which utilize artificial neural networks to extract features from medical images automatically, have shown great promise in improving the accuracy and efficiency of lung disease diagnosis. This review discusses the various DL methods that have been developed for lung disease diagnosis, including convolutional neural networks (CNNs), deep neural networks (DNNs), and generative adversarial networks (GANs). The advantages and limitations of each method are discussed, along with the types of medical imaging techniques used, such as X-ray and computed tomography (CT). In addition, the review discusses the most commonly used performance metrics for evaluating the performance of DL for lung disease diagnosis: the area under the curve (AUC), sensitivity, specificity, F1-score, accuracy, precision, and the receiver operator characteristic curve (ROC). Moreover, the challenges and limitations of using DL for lung disease diagnosis, including the limited availability of annotated data, the variability in imaging techniques and disease presentation, and the interpretability and generalizability of DL models, are highlighted in this paper. Furthermore, strategies to overcome these challenges, such as transfer learning, data augmentation, and explainable AI, are also discussed. The review concludes with a call for further research to address the remaining challenges and realize DL's full potential for improving lung disease diagnosis and treatment.

ObjectivePituitary adenomas are the most common type of pituitary disorders, which usually occur in young adults and often affect the patient’s physical development, labor capacity and fertility. Clinical free texts noted in electronic medical records (EMRs) of pituitary adenomas patients contain abundant diagnosis and treatment information. However, this information has not been well utilized because of the challenge to extract information from unstructured clinical texts. This study aims to enable machines to intelligently process clinical information, and automatically extract clinical named entity for pituitary adenomas from Chinese EMRs.MethodsThe clinical corpus used in this study was from one pituitary adenomas neurosurgery treatment center of a 3A hospital in China. Four types of fine-grained texts of clinical records were selected, which included notes from present illness, past medical history, case characteristics and family history of 500 pituitary adenoma inpatients. The dictionary-based matching, conditional random fields (CRF), bidirectional long short-term memory with CRF (BiLSTM-CRF), and bidirectional encoder representations from transformers with BiLSTM-CRF (BERT-BiLSTM-CRF) were used to extract clinical entities from a Chinese EMRs corpus. A comprehensive dictionary was constructed based on open source vocabularies and a domain dictionary for pituitary adenomas to conduct the dictionary-based matching method. We selected features such as part of speech, radical, document type, and the position of characters to train the CRF-based model. Random character embeddings and the character embeddings pretrained by BERT were used respectively as the input features for the BiLSTM-CRF model and the BERT-BiLSTM-CRF model. Both strict metric and relaxed metric were used to evaluate the performance of these methods.ResultsExperimental results demonstrated that the deep learning and other machine learning methods were able to automatically extract clinical named entities, including symptoms, body regions, diseases, family histories, surgeries, medications, and disease courses of pituitary adenomas from Chinese EMRs. With regard to overall performance, BERT-BiLSTM-CRF has the highest strict F1 value of 91.27% and the highest relaxed F1 value of 95.57% respectively. Additional evaluations showed that BERT-BiLSTM-CRF performed best in almost all entity recognition except surgery and disease course. BiLSTM-CRF performed best in disease course entity recognition, and performed as well as the CRF model for part of speech, radical and document type features, with both strict and relaxed F1 value reaching 96.48%. The CRF model with part of speech, radical and document type features performed best in surgery entity recognition with relaxed F1 value of 95.29%.ConclusionsIn this study, we conducted four entity recognition methods for pituitary adenomas based on Chinese EMRs. It demonstrates that the deep learning methods can effectively extract various types of clinical entities with satisfying performance. This study contributed to the clinical named entity extraction from Chinese neurosurgical EMRs. The findings could also assist in information extraction in other Chinese medical texts.

According to a recently conducted survey on surgical complication mortality rate, 47% of such cases are due to anesthetics overdose. This indicates that there is an urgent need to moderate the level of anesthesia. Recently deep learning (DL) methods have played a major role in estimating the depth of Anesthesia (DOA) of patients and has played an essential role in control anesthesia overdose. In this paper, Electroencephalography (EEG) signals have been used for the prediction of DOA. EEG signals are very complex signals which may require months of training and advanced signal processing techniques. It is a point of debate whether DL methods are an improvement over the already existing traditional EEG signal processing approaches. One of the DL algorithms is Convolutional neural network (CNN) which is very popular algorithm for object recognition and is widely growing its applications in processing hierarchy in the human visual system. In this paper, various decomposition methods have been used for extracting the features EEG signal. After acquiring the necessary signals values in image format, several CNN models have been deployed for classification of DOA depending upon their Bispectral Index (BIS) and the signal quality index (SQI). The EEG signals were converted into the frequency domain using and Empirical Mode Decomposition (EMD), and Ensemble Empirical Mode Decomposition (EEMD). However, because of the inter mode mixing observed in EMD method; EEMD have been utilized for this study. The developed CNN models were used to predict the DOA based on the EEG spectrum images without the use of handcrafted features which provides intuitive mapping with high efficiency and reliability. The best trained model gives an accuracy of 83.2%. Hence, this provides further scope and research which can be carried out in the domain of visual mapping of DOA using EEG signals and DL methods.