Character Recognition Process Research Articles

A Dilated MultiRes Visual Attention U-Net for historical document image binarization

The task of binarization of historical document images has been in the forefront of image processing research, during the digital transition of libraries. The process of storing and transcribing valuable historical printed or handwritten material can salvage world cultural heritage and make it available online without physical attendance. The task of binarization can be viewed as a pre-processing step that attempts to separate the printed/handwritten characters in the image from possible noise and stains, which will assist in the Optical Character Recognition (OCR) process. Many approaches have been proposed before, including deep learning based approaches. In this article, we propose a U-Net style deep learning architecture that incorporates many other developments of deep learning, including residual connections, multi-resolution connections, visual attention blocks and dilated convolution blocks for upsampling. The novelties in the proposed DMVAnet lie in the use of these elements in combination in a novel U-Net style architecture and the application of DMVAnet in image binarization for the first time. In addition, the proposed DMVAnet is a very computationally lightweight network that performs very close or even better than the state-of-the-art approaches with a fraction of the network size and parameters. Finally, it can be used on platforms with restricted processing power and system resources, such as mobile devices and through scaling can result in inference times that allow for real-time applications.

License Plate Recognition: A Brief Overview

Abstract: The License Plate Recognition system is a technological solution designed to automatically capture and interpret license plate information from vehicles. Utilizing a blend of object detection, character segmentation, optical character recognition (OCR) and image processing, ANPR systems play a pivotal role in bolstering security, optimizing traffic management, and supporting diverse applications such as toll collection and parking management. Equipped with highresolution cameras and advanced algorithms, these systems are integral to modern surveillance and transportation, providing solutions that enhance safety and operational efficiency.

Planetary Knowledge Base: Semantic Transcription Using Graph Neural Networks

The Natural History Museum, London (NHM), in collaboration with Amazon Web Services (AWS), has embarked on a project to build the Planetary Knowledge Base (PKB), a comprehensive graph network comprising data on all specimens, collectors, and localities. In the initial prototype, we have concentrated on botanical specimens, using all plant taxa and specimens within the Global Biodiversity Information Facility (GBIF), combined with geographic data from GeoNames and biographic data from WikiData, Bionomia, Harvard Index of Botany, TL2, and Tropicos. Development of the PKB is a huge undertaking—our first proof of concept has more than 100 million nodes. The primary application of this knowledge graph (KG) is powering the automated transcription of specimen labels. Using Graph Convolutional Neural Networks, textual information from labels can be aligned to the entities in the graph, creating structured semantic data from the raw text. Text is extracted from images using services from the AWS ecosystem, including Optical Character Recognition and Natural Language Processing to identify the units of information, creating a high-throughput auto-digitisation workflow for extracting structured data. The PKB graph network enables new ways to interrogate collections. It can help identify species that may require re-examination or re-identification due to taxonomic updates or inconsistencies. It can also flag potential discrepancies or conflicts in the data, such as cases where the same species is recorded under different names or classifications across various sources. Moreover, the PKB can detect possible errors and outliers in the knowledge graph and point out specimens that could represent new species misidentified within the collection. By cross-validating species with the International Union for Conservation of Nature (IUCN) Red List, it can also assist in analysing species populations with insufficient data. The PKB is being developed as a cloud service, so researchers and other institutions can experiment with this transformative technology, using it to support their own digitisation efforts.

Development of a Benchmark Odia Handwritten Character Database for an Efficient Offline Handwritten Character Recognition with a Chronological Survey

A good benchmark dataset is a primary requirement in the offline handwritten character recognition (HCR) process. Only three handwritten numerals and alphabet datasets from Odia are publicly accessible for study, although many writers have used several datasets in their experiments. In this article, two tasks are done to address this issue. Those are the following: First, an extensive survey focused on various datasets is provided with the methodologies used in chronological order. The second factor is a solution to the lack of publicly available handwritten characters and numeral datasets. A new dataset of handwritten Odia characters with numerals has been developed. Anyone can access this dataset by sending an email to the authors of the article. This dataset was created with the help of 150 volunteers of various age groups, races, and qualifications. Some homogeneous experiments are conducted using deep learning models to evaluate the consistency of the dataset. One heterogeneous trial has also been performed to estimate the complexities of the characters present in the dataset by comparing them with the existing benchmark datasets.

OCR Of Devanagari Script Using CNN

Devanagari script is widely used across India. It forms many languages like Hindi, Marathi, Nepali and Sanskrit languages. As the Devanagari characters are similar to the hindi character the national language of India. It is important to recognize the characters to understand the message that particular tries to tell. The automatic character recognition system is thus developing for the Devanagari script. The character recognition process converts an image of a character into machine-readable format also its English corresponds. In this paper, we are using Convolutional Neural Network for developing the character recognition system. Convolutional neural network learns directly from data. It is a type of Deep learning neural network architecture. CNN is useful as it does not require any human intervention and performs the identification of important features on its own. The proposed paper uses a CNN algorithm applied to a dataset of 49 characters of Devanagari script. The dataset contains of total 4018 Images. The algorithm of the Convolutional Neural Network is applied to train the dataset. The input image to be predicted is first preprocessed and then the model predicts the output result. The system is designed in Jupyter Lab using Python. The Convolutional Neural Network model's overall accuracy is 98%.

DIFFERENT APPROACHES OF HANDWRITTEN BENGALI CHARACTER RECOGNITION FROM DOCUMENT IMAGE

Handwritten character recognition from a document image is no-doubt a challenging task and active research filed of interest because of its several application domain. It becomes more challenging in pattern recognition domain because of variability of human writing style, skew and orientation. Noises, smears and faded ink makes it more complicated. OCR research success has limited to very few scripts like Roman, English and Chinese. Among Indian script, Bengali Handwritten Script identification is still an open and active research area. In this paper some popular approach are discussed to indentify the handwritten Bengali character recognition process.

Key–Value Pair Identification from Tables Using Multimodal Learning

Computer vision and optical character recognition techniques have rapidly advanced in order to accurately capture text and other features from paper documents. While state-of-the-art tools in these fields now yield high accuracy, analyzing their outputs requires more research. Since tables are common in such documents, a new pipeline, based on multimodal learning, is proposed to better extract key–value pairs from tables. Its performance is evaluated with a synthetically generated dataset with randomly generated tables and a dataset of mechanical part documents provided by SiliconExpert Technologies. Its performance is also compared with another state-of-the-art model built for similar tasks, LayoutLM. The proposed pipeline provides a fully automated, end-to-end scalable solution, beginning with image processing and computer vision components to a machine learning model that uses data from optical character recognition and natural language processing to make the final decisions. In the best configuration, the pipeline achieved a 96.26% accuracy on a large, synthetically generated training and test set. When comparing the proposed pipeline with LayoutLM, the proposed pipeline performed similarly on the synthetic dataset and better on the real dataset. These results show the potential of the multimodal approach in extracting key–value pairs from tables from real paper documents.

Моделирование процесса распознавания символов в нормативных документах организации

Improving the quality of classification of different documents is a purpose of modeling the optical character recognition. Non-digital documents, such as scanned or photographed documents, are difficult to classify correctly in electronic document management systems. A decision was made to simulate the process of optical character recognition in the regulatory documents of the organization. There have been considered various methods of modeling the process. The structure of departments for the electronic document management system is given. Methods of implementing optical character recognition (OCR) are considered. The stages of the OCR system development are revealed: image processing, segmentation, recognition. The methods of image processing are analyzed. The main processes associated with image processing are disclosed: alignment, blurring, binarization, finding contours, removing extra lines. Comparison of image blur methods is made. Two stages of image binarization are defined: conversion of a color image into a gray image, binarization of a gray image. The Kenny operator is proposed as a second stage of binarization, which is used to detect the boundaries of the image. The last stage of image processing is the process of removing extra lines. Algorithms for dividing text areas into segments are considered. 3 stages of segmentation are identified: string segmentation, word segmentation, character segmentation. A segmentation algorithm is defined based on calculating the average brightness of image pixels to search for different intervals: line spacing, word spacing, character spacing. Available popular online OCR services as well as some popular desktop programs are considered. A connection has been found between an artificial neural network and optical object recognition. To implement the recognition stage, it is proposed to use an artificial neural network.

Clinical implications of tumor-based next-generation sequencing in high-grade epithelial ovarian cancer.

Tumor-based next-generation sequencing is used inconsistently as a tool to tailor treatment of ovarian cancer, yet beyond detection of somatic BRCA1 and BRCA2 mutations, the clinical benefit is not well established. This study aimed to assess the clinical relevance of tumor-based next-generation sequencing (tbNGS) in patients with ovarian cancer. This retrospective study included patients with high-grade epithelial ovarian carcinoma. tbNGS results were identified in the electronic medical record using optical character recognition and natural language processing. Genetic, clinical, and demographic information was collected. Progression-free survival (PFS) and overall survival were calculated and compared using log-rank tests. Multivariate Cox regression and clustering analyses were used to identify patterns of genetic alterations associated with survival. Of 1092 patients in the described population, 409 (37.5%) had tbNGS results. Nearly all (96.1% [393/409]) had one or more genetic alterations. In 25.9% (106/409) of patients, an alteration that aligned with a targeted treatment was identified, and in an additional 48.7% (199/409), tbNGS results suggested eligibility for an investigational agent or clinical trial. The most frequent alterations were TP53, PIK3CA, and NF1 mutations, and CCNE1 amplification. Together, BRCA1 and BRCA2 mutations were associated with longer PFS (hazard ratio [HR], 0.62; 95% confidence interval [CI], 0.42-0.92; p=.02), whereas AKT2 amplification was associated with shorter PFS (HR, 3.86; 95% CI, 1.002-14.88; p<.05). Multivariate Cox regression and clustering analyses identified several combinations of genetic alterations that corresponded to outcomes in patients with high-grade serous carcinoma. tbNGS often yields clinically relevant information. Detailed analysis of population-level tumor genomics may help to identify therapeutic targets and guide development of clinical decision support tools. Although more and more patients with ovarian cancer are undergoing tumor-based next-generation sequencing to identify genetic mutations in their tumors, the benefits of such testing are not well established. In a group of over 400 patients with ovarian cancer who underwent tumor-based next-generation sequencing in the course of their treatment, nearly all patients had one or more genetic alterations detected, and one out of four patients had a mutation that qualified them for a personalized treatment option.

An Intelligent Telugu Handwritten Character Recognition Using Multi-Objective Mayfly Optimization with Deep Learning–Based DenseNet Model

The handwritten character recognition process has gained significant attention among research communities due to its application in assistive technologies for visually impaired people, human–robot interaction, automated registry for business documents, and so on. Handwritten character recognition of the Telugu language is difficult owing to the absence of a massive dataset and a trained convolutional neural network (CNN). This article introduces an intelligent Telugu character recognition process using a multi-objective mayfly optimization with deep learning (MOMFO-DL) model. The proposed MOMFO-DL technique involves the DenseNet-169 model as a feature extractor to generate a useful set of feature vectors. A functional link neural network (FLNN) is used as a classification model to recognize and classify the printer characters. The design of the MOMFO technique for the parameter optimization of the DenseNet model and FLNN model shows the novelty of the work. The use of MOMFO technique helps to optimally tune the parameters in such a way that the overall performance can be improved. The extensive experimental analysis takes place on benchmark datasets and the outcomes are examined with respect to different measures. The experimental results pointed out the supremacy of the MOMFO technique over the recent state-of-the-art methods.

Validation of Argo (Automatic record generator for Onco-Hematology), a New App Supporting the Automatic Conversion of Paper-Based Pathology Reports in Standardized Ecrfs

Background and aims. The scarce accessibility to integrated systems limits the advantage of using real-world medical data for translational research purposes. ARGO (Automatic Report Generation for Onco-hematology) converts paper-based pathology reports in electronic Case Report Forms (eCRFs) exploiting Optical Character Recognition and Natural Language Processing technologies [Zaccaria et al., Sci. Rep., 2021]. Here, we present for the first time the App version of ARGO, designed to support physicians and data entries in rapidly generating eCRFs in a standardized and filtrable fashion. For scalability purposes, we tested ARGO App by processing n. 501 pathology reports from eight Italian Centers, including Hodgkin (HL), Diffuse Large B-Cell (DLBCL), Follicular (FCL), Mantle Cell (MCL), and T-Cell (TCL) lymphoma diagnoses. Methods. Validation involved six expert hematologists who generated eCRFs by simply acquiring photographs of each paper-based report using commercially available camera-equipped smartphones (Apple® iPhones, IOS version 15). The set included n. 347 and n. 154 reports from IRCCS Istituto Tumori 'Giovanni Paolo II’ (internal series, IS) and seven Italian cooperative Centers (external series, ES), respectively. Overall, they comprised n. 139 HL, n. 154 DLBCL, n. 109 FL, n. 76 MCL, n. 6 TCL, n. 17 unclassified describing major immunohistochemistry markers (IMs) as MYC, BCL2, BCL6, CD10, CD20, Cyclin D1, CD79a, CD15, CD30, PAX5, CD5, CD3, and Ki-67 proliferation index. The automatic process of diagnosis assignment by the ARGO algorithm (developed in Python) was imposed to depend on the highest matching rate between the detected IMs and corresponding classification as from the National Institute of Health in accordance with the International Classification of Diseases, 10th (ICD-10) and oncology (ICD-O) versions. To overcome potential misdiagnosis risk, a Random Forest (RF) model was trained on the IMs set of the IS, tested on the ES, and combined with ARGO algorithm. The performance of the App was assessed for accuracy and F1-score, which is a more sensitive metric. Results. The ARGO App includes two use-cases at prospectically acquiring and retrospectively reading reports by users (physicians and/or data-managers). The first use-case allows users to acquire pathology reports via mobile phone's camera. The second use-case leads to search patients’ data filtering by "Report ID", "Name", "Surname” and "Type of diagnosis". For each new record, ARGO converts information about patients’ demography, diagnosis, tissue of origin of samples (lymph-node, extra-nodal, bone marrow, and peripheral blood), and IMs expressions. ARGO successfully converted 490 (97.8%) reports into structured eCRFs (overall, n. 18,816 data). In terms of accuracy (Fig. 1A), MYC, Cyclin D1, CD79a, CD15, EMA, BCL2 (by fluorescent in situ hybridization), IgD, IgM, EBV, Ki-67 detection achieved among 85.7% and 99.4% in both series. BCL2, BCL6, CD10, CD20, CD30, PAX5, CD23, CD5, and CD45/LCA achieved among 65.1% and 85.6% for the IS and among 71.0% and 90.9% for the ES. MUM1 and CD3 achieved 56.2% and 71.4%, and 72.0% and 79.9% for IS and ES, respectively. Concerning the F1-score (Fig. 1B), although no significant differences were observed between the two series, on average, biomarkers gave a score that was lower of 20.2% for IS and 35.3% for ES compared to accuracy. Interestingly, Ki-67 proliferation index, MYC, CD10, CD20, Cyclin D1, CD79a, CD15, CD23, and CD5 achieved among 73.5% and 85.9% for the IS and among 72.4% and 85.5% for the ES. The capturing of diagnosis achieved 87.3% and 82.5% of accuracy, and 87.3% and 83.0% of F1-score for IS and ES, respectively. Focusing on individual diagnoses in the ES, HL, MCL, DLBCL, FCL, and TCL reached 90.0%, 88.5%, 84.9%, 76.5%, 33.3%, respectively. Of these n. 154 reports, n. 52 (34%) were detected by the sole ARGO algorithm, n. 28 (18%) by RF, n. 67 (43%) by a combination of both, while n. 7 (5%) remained unclassified. Conclusions. We validated ARGO App that robustly converts paper-based pathology reports of major lymphoma subtypes into structured eCRFs. ARGO is feasible and easily transferable into the daily practice to generate standardized patients’ clinical records for clinical and translational research purposes. Ongoing efforts are aiming at enlarging the TCL cohort of pathology reports and developing a multilanguage version for other languages than Italian. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Automated Analysis of Treatment Log Files and Videos for MR-Guided Stereotactic Body Radiation Therapy Treatments of Thoracic and Abdominal Tumors

<h3>Purpose/Objective(s)</h3> MR-guided stereotactic body radiation therapy (MRgSBRT) is a novel method of treating mobile tumors with robust real-time gating and on-table adaptive planning. Delivery log files and cine videos of the treatment generated by the software system can be a rich source of information for assessing MRgSBRT treatment sessions. We hypothesized that developing scripts for automated review of MRgSBRT treatments using delivery logs and cine videos would be feasible. We have included a summary of our findings from our initial experience using this automated analysis. <h3>Materials/Methods</h3> We conducted a retrospective review of patients undergoing MRgSBRT for treatment of liver, lung, and pancreas tumors. Custom Python scripts were written to automatically extract treatment delivery information from log files and export data neatly into to a spreadsheet. A separate Python script was written to extract information from delivery cine videos using optical character recognition (OCR), color thresholding, and morphological processing as well. <h3>Results</h3> Using our automated system we were able to extract the following data points from the log files: treatment site, date and time of treatment, tracking algorithm configuration, average treatment duration, total dose, fraction (fx) number, MU per fx, couch shifts, and need for plan re-optimization. Additionally, through automated analysis of treatment cine video files we were able to extract frame by frame information for tracking structure centroid displacement, beam on/off status, frame-frame similarity, and tracking confidence (TC) over the course of treatment. Table 1 is an example of the data output. <h3>Conclusion</h3> Automated analysis of MRgSBRT delivery log files and log videos is feasible and can serve as a rich resource for QA/QI to improve treatment efficiency on an MR-LINAC by quantifying a multitude of measurable parameters and formulating strategies to address any unrealized bottlenecks in the process. Future plans include expanding our automation to examine total patient time on table, monitor units per beam segment, and predicting internal tumor volumes (ITV) based on tracking target displacement obtained from the delivery videos.

Enhancement of license plate recognition performance using Xception with Mish activation function.

The current breakthroughs in the highway research sector have resulted in a greater awareness and focus on the construction of an effective Intelligent Transportation System (ITS). One of the most actively researched areas is Vehicle Licence Plate Recognition (VLPR), concerned with determining the characters contained in a vehicle’s Licence Plate (LP). Many existing methods have been used to deal with different environmental complexity factors but are limited to motion deblurring. The aim of our research is to provide an effective and robust solution for recognizing characters present in license plates in complex environmental conditions. Our proposed approach is capable of handling not only the motion-blurred LPs but also recognizing the characters present in different types of low resolution and blurred license plates, illegible vehicle plates, license plates present in different weather and light conditions, and various traffic circumstances, as well as high-speed vehicles. Our research provides a series of different approaches to execute different steps in the character recognition process. The proposed approach presents the concept of Generative Adversarial Networks (GAN) with Discrete Cosine Transform (DCT) Discriminator (DCTGAN), a joint image super resolution and deblurring approach that uses a discrete cosine transform with low computational complexity to remove various types of blur and complexities from licence plates. License Plates (LPs) are detected using the Improved Bernsen Algorithm (IBA) with Connected Component Analysis(CCA). Finally, with the aid of the proposed Xception model with transfer learning, the characters in LPs are recognised. Here we have not used any segmentation technique to split the characters. Four benchmark datasets such as Stanford Cars, FZU Cars, HumAIn 2019 Challenge datasets, and Application-Oriented License Plate (AOLP) dataset, as well as our own collected dataset, were used for the validation of our proposed algorithm. This dataset includes the images of vehicles captured in different lighting and weather conditions such as sunny, rainy, cloudy, blurred, low illumination, foggy, and night. The suggested strategy does better than the current best practices in both numbers and quality.

The devil you know: A population-level analysis of tumor-based next-generation sequencing of high-grade epithelial ovarian cancer (151)

<b>Objectives:</b> Clinical use of tumor-based next-generation sequencing (tbNGS) is expanding, but the indications and implications remain vague. We aimed to assess the clinical relevance of tbNGS data in a large cohort of patients with high-grade epithelial ovarian cancer. <b>Methods:</b> Electronic medical records (both internal notes and external scanned documents) of 905 patients with high-grade serous or endometrioid ovarian cancer were queried for tbNGS data using optical character recognition (OCR) and natural language processing (NLP) technologies. Identified results and clinical-demographic information were entered into an institutional database. Data were analyzed with descriptive statistics and log-rank tests comparing selected genetic alterations to progression-free survival (PFS). <b>Results:</b> Three hundred and twelve (34.4%) patients had their tbNGS results identified. Tests used had variable breadth, ranging from 46 genes to whole-exome sequencing. The most common was an institutional panel that evaluated 146 genes. Nearly all (99.7%) patients had their genomic alterations (mutations or copy number variations) identified. On average, each tumor had 2.38 (±2.46) aberrations. Two hundred and fifty-two (80.8%) patients had actionable alterations, defined as relaying eligibility for targeted treatment or clinical trial, according to the Precision Oncology Decision Support System (PODSS) database. Twenty-seven (10.7% of those with actionable abnormalities) enrolled in relevant clinical trials. In total, 630 mutations were detected in the population, with half of these (58.6%) considered actionable. The most common type of abnormality encountered was missense substitution (52.9%), followed by amplification (14.7%) and a frameshift mutation (9.72%). The most frequently mutated genes were <i>TP53</i> (93.0%), <i>PIK3CA</i> (9.35%), <i>BRCA1/2</i> (8.30%), <i>NF1</i> (4.90%), and <i>KRAS</i> (4.52%). Among these, only <i>BRCA1/2</i> was associated with a significant improvement in PFS (20.1 vs 14.1 months, p=0.046). <i>CCNE1</i> amplification, the most prevalent copy number variant, was present in 9.4% of those tested. Specific mutation types or copy number variations were not correlated with PFS. <b>Conclusions:</b> Clinical use of tbNGS is prevalent in ovarian cancer. Of patients undergoing tbNGS at our institution, four out of five gained the opportunity for an FDA-approved treatment or genotypically relevant clinical trial. Description of population-level tumor genomics could help identify therapeutic targets and guide the development of clinical decision support tools.

Research on Handwritten Chinese Character Recognition Based on BP Neural Network

The application of pattern recognition technology enables us to solve various human-computer interaction problems that were difficult to solve before. Handwritten Chinese character recognition, as a hot research object in image pattern recognition, has many applications in people’s daily life, and more and more scholars are beginning to study off-line handwritten Chinese character recognition. This paper mainly studies the recognition of handwritten Chinese characters by BP (Back Propagation) neural network. Establish a handwritten Chinese character recognition model based on BP neural network, and then verify the accuracy and feasibility of the neural network through GUI (Graphical User Interface) model established by Matlab. This paper mainly includes the following aspects: Firstly, the preprocessing process of handwritten Chinese character recognition in this paper is analyzed. Among them, image preprocessing mainly includes six processes: graying, binarization, smoothing and denoising, character segmentation, histogram equalization and normalization. Secondly, through the comparative selection of feature extraction methods for handwritten Chinese characters, and through the comparative analysis of the results of three different feature extraction methods, the most suitable feature extraction method for this paper is found. Finally, it is the application of BP neural network in handwritten Chinese character recognition. The establishment, training process and parameter selection of BP neural network are described in detail. The simulation software platform chosen in this paper is Matlab, and the sample images are used to train BP neural network to verify the feasibility of Chinese character recognition. Design the GUI interface of human-computer interaction based on Matlab, show the process and results of handwritten Chinese character recognition, and analyze the experimental results.

GUI Based Optical Character Recognition using Neural Networks

The Optical Character Recognition (OCR) process means the transition from examined primer or written images to a machine- adjudged document. The American Standard Code for Information Interchange (ASCII) in cognitive processing uses OCR. The challenge is two primary clans word segmentation by letters and character recognition. Apply a new approach to include the two functions by Scale-Invariant Transforming Feature (SIFT) descriptors. To compare SIFT descriptors (RootSIFT), construct a new procedure, that offers outstanding results without accelerating calculation or repository conditions. In order to identify English characters, proposed system suggests that the reverse propagation neural network for bracket of character be employed. Conducted trials with further than 10 expedients aimed for every character and tried the delicacy for numerical figures, map letters, small letters, and alphanumeric symbols. The interpretation analysis of optimized neural network algorithm has attained an outside.

Vehicle Number Plate Detection System for Indian Vehicle using OCR

Automatic Plate recognition is one of the techniques used for vehicle identification purposes and it will provide all details about particular vehicle owner This process usually comprises of three steps. First step is the vehicle plate localization, regardless of the vehicle-plate size and orientation. The second step is the segmentation of the characters and last step is the recognition of the characters from the vehicle plate. Thus, this project uncovers the fundamental idea of various algorithms required to accomplish character recognition from the vehicle plate. This feature of the algorithm mentioned above helped in achieving faster character recognition of the license plate. This process of character recognition consists of steps like Image processing, Defragmentation, Resizing and Character localization that are required to be performed on the image.

Clinical implications of tumor-based next-generation sequencing in ovarian cancer.

5545 Background: Epithelial ovarian cancer is genetically heterogeneous, both among and within histologic subtypes. Advances in next-generation sequencing have made it feasible to ascertain the somatic genetic signature of each patient, however, critical analysis of population-level sequencing results is required to maximize the potential of this technology. Here, we aimed to assess the clinical relevance of tumor-based next-generation sequencing (tbNGS) in a large cohort of patients with high-grade epithelial ovarian cancer. Methods: Our study population comprised patients with high-grade serous (n = 972), clear cell (n = 33), endometrioid (n = 28), mucinous (n = 4), and mixed (n = 34) or unspecified (n = 21) epithelial ovarian carcinoma diagnosed between April 2013 and September 2021. tbNGS results were identified within the electronic medical record using optical character recognition and natural language processing. Genetic, clinical, and demographic information was collected for patients who had undergone tbNGS. Progression-free survival (PFS) and overall survival (OS) were calculated from date of first treatment to date of first recurrence and date of death, respectively. Data were analyzed using descriptive statistics, univariate and multivariate Cox regression models, and clustering analyses. Results: Of 1092 patients in the described population, 409 (37.5%) had tbNGS results identified. Nearly all (96.1%) revealed one or more genetic aberrations. Most patients (74.6%) had an actionable mutation, defined as relaying eligibility for a targeted treatment or clinical trial. The most frequent alterations were TP53, PIK3CA, and NF1 mutations; and CCNE1 amplification. Ten different targeted institutional and commercial panels were employed, covering a range of 35 to 600+ gene loci. The median time from diagnosis to testing was 14.5 months, likely corresponding to time of recurrence. Though no standalone alterations were significantly related to survival, multivariate and clustering analyses identified several genetic patterns which corresponded to patient outcomes. Mutation of BRAF, PIK3R1, NOTCH3, MET, and/or ATR was correlated with shorter PFS (HR 1.84, p = 0.001); mutation of ATM, RB1, CDKN2A, FGFR1, and/or FGFR2 was associated with improved PFS (HR 0.64, p = 0.04), as was mutation of NBN and/or ATRX (HR 0.54, p &lt; 0.05). MYC, NOTCH3, and/or CREBBP mutations were significantly correlated with worse OS (HR 1.95, p = 0.02). In our population, 40 patients (9.78%) were enrolled in genotypically-relevant clinical trials. Conclusions: tbNGS is prevalent at our institution, and often yields actionable information. We identified several mutational patterns that correlate to patient survival. Detailed analysis of population-level tumor genomics may help to identify therapeutic targets and guide development of clinical decision support tools.

Non-monotonic developmental trend of holistic processing in visual expertise: the case of Chinese character recognition

Holistic processing has been identified as an expertise marker of face and object recognition. By contrast, reduced holistic processing is purportedly an expertise marker in recognising orthographic characters in Chinese. Does holistic processing increase or decrease in expertise development? Is orthographic recognition a domain-specific exception to all other kinds of recognition (e.g. face and objects)? In two studies, we examined the developmental trend of holistic processing in Chinese character recognition in Chinese and non-Chinese children, and its relationship with literacy abilities: Chinese first graders—with emergent Chinese literacy acquired in kindergarten—showed increased holistic processing perhaps as an inchoate expertise marker when compared with kindergartners and non-Chinese first graders; however, the holistic processing effect was reduced in higher-grade Chinese children. These results suggest a non-monotonic inverted U-shape trend of holistic processing in visual expertise development: An increase in holistic processing due to initial reading experience followed by a decrease in holistic processing due to literacy enhancement. This result marks the development of holistic and analytic processing skills, both of which can be essential for mastering visual recognition. This study is the first to investigate the developmental trend of holistic processing in Chinese character recognition using the composite paradigm.

An OCR Post-Correction Approach Using Deep Learning for Processing Medical Reports

According to a recent Deloitte study, the COVID-19 pandemic continues to place a huge strain on the global health care sector. Covid-19 has also catalysed digital transformation across the sector for improving operational efficiencies. As a result, the amount of digitally stored patient data such as discharge letters, scan images, test results or free text entries by doctors has grown significantly. In 2020, 2314 exabytes of medical data was generated globally. This medical data does not conform to a generic structure and is mostly in the form of unstructured digitally generated or scanned paper documents stored as part of a patient’s medical reports. This unstructured data is digitised using Optical Character Recognition (OCR) process. A key challenge here is that the accuracy of the OCR process varies due to the inability of current OCR engines to correctly transcribe scanned or handwritten documents in which text may be skewed, obscured or illegible. This is compounded by the fact that processed text is comprised of specific medical terminologies that do not necessarily form part of general language lexicons. The proposed work uses a deep neural network based self-supervised pre-training technique: Robustly Optimized Bidirectional Encoder Representations from Transformers (RoBERTa) that can learn to predict hidden (masked) sections of text to fill in the gaps of non-transcribable parts of the documents being processed. Evaluating the proposed method on domain-specific datasets which include real medical documents, shows a significantly reduced word error rate demonstrating the effectiveness of the approach.