Video Index Point Detection and Extraction Framework Using Custom YoloV4 Darknet Object Detection Model

We have put together a corpus of 242 abstracts of Arabic documents using the Proceedings of the Saudi Arabian National Conferences as a source. All these abstracts involve computer science and information systems. We also designed and built an automatic information retrieval system from scratch to handle Arabic data. The system was implemented in the C language using the GCC compiler and runs on IBM/PCs and compatible microcomputers. We have implemented both automatic and manual indexing techniques for this corpus. A long series of experiments using measures of recall and precision has demonstrated that automatic indexing is at least as effective as manual indexing and more effective in some cases. Since automatic indexing is both cheaper and faster, our results suggest that we can achieve a wider coverage of the literature with less money and produce as good results as with manual indexing. We have also compared the retrieval results using words as index terms versus stems and roots, and confirmed the results obtained by Al-Kharashi and Abu-Salem with smaller corpora that root indexing is more effective than word indexing.

The proliferation of computer technology and digital image‐acquisition hardware has led to the widespread use of image data across a variety of applications including astronomy, art, natural resources, engineering design, military, business operation, medicine, education, etc. Major research activities in the digital image databases surged after the U.S. government's Digital Library Initiative (DLI) from 1994 to 1998. The DLI research team at the University of California at Berkeley developed a work‐centered digital information system that contains 450 digital text documents, 200 air photos, and over 11,000 ground photographs. The system allowed a user or a working group to access its own collections of varying data types, and to generate new materials to be added to the collection. The DLI project at the University of California (UC) at Santa Barbara focused on the development of digital databases that contain geographically referenced materials such as maps and aerial photos. The research team developed a number of tools for browsing and retrieving map images at multiple resolution. Two other major efforts in developing digital image databases were at The National Library of Medicine (NLM), a component of the National Institutes of Health (NIH), and at Time Warner. More recently, research scientists at the University of California at Berkeley and the Fine Arts Museums of San Francisco successfully launched on the Internet the largest art image database in the world, the Thinker ImageBase.Automatic image indexing and retrieving technology is fundamental in digital image databases. There are two general strategies in image retrieval, browsing and searching. Browsing is an information retrieval strategy in which a user navigates through an ordered arrangement of images by making selections from the progressive levels of a hierarchy into which the available images have been logically grouped. Image browsing relies on the user's cognitive abilities to recognize images of interest without having to formulate a specific query. Searching is an information retrieval strategy in which the user communicates with an information retrieval system by an interface that requires the user to input a search query.Images can be indexed and retrieved by text‐tag information and/or by image content. Conventionally, images are indexed using text‐tag information, such as title, key words, date of the work, artist, author, photographer, legend, captions, etc. Images are often subject to a wide range of interpretations, and textual descriptions can only begin to capture the richness and complexity of the semantic content of a visual image. In many applications, the complexity of the information embedded in image content such as types of objects, object attributes, and spatial relationship of objects, etc, cannot be synthesized in a few key words. It has been reported that users querying an image collection tend to be much more specific in their requests and information needs than when querying a text database. Furthermore, text indexing requires large human effort in creating the meta‐data that enables visual queries and is language dependent. Large databases containing thousands and millions of digital images that can occupy gigabytes of space are almost impossible for manual indexing and searching. The demand for systems using pictorial information combined with textual description in image retrieval is growing and automatic indexing and retrieval based on image content has become the most promising techniques for large image databases and digital image libraries. This article describes the techniques developed for content‐based retrieval.

In the era of information explosion, efficiently retrieving relevant information from vast datasets has become a paramount challenge, with deep learning offering new avenues for enhancing information retrieval (IR) systems. This study introduces an innovative Recurrent Neural Network (RNN)-based approach to significantly improve the efficiency and accuracy of IR tasks, surpassing the limitations of traditional methods. By designing a specialized RNN architecture and comparing it against traditional and existing deep learning-based IR models across diverse datasets and various IR tasks—including document retrieval, question answering, and semantic search—our methodology demonstrates remarkable improvements in retrieval efficiency and accuracy. Specifically, the proposed model achieved substantial increases in precision, recall, and F1 score, highlighting the potential of RNNs in mastering the sequential and contextual nuances inherent in IR processes. This advancement offers a robust and efficient solution to the challenges of traditional IR systems, with future work aiming to integrate this model into broader IR applications and explore further enhancements using advanced neural network techniques.

SIGIR, from its very onset in the early 1960's, has been concerned with the development of automatic information retrieval systems and with improving the effectiveness of automatic indexing. Automatic indexing is a subsystem, or component, of an automatic information retrieval system. The term "automatic" implies that the process is to be accomplished by a set of computer programs rather than by the intellectual effort of skilled people. The essential questions that need to be answered are how automatic indexing can:• Adequately represent the subject content of a document;• Improve recall by increasing the number of relevant documents retrieved;• Improve precision by decreasing the number of non-relevant documents retrieved.In this tutorial, I will review the basic and improved procedures that have been devised to respond to each of these questions. Finally, after we have reviewed developments in automatic indexing, we will consider the more radical questions of:• Whether we can rely exclusively on automatic indexing to achieve adequate retrieval effectiveness in online interactive document retrieval systems, and• Whether the traditional recall and precision ratios are the best criteria for evaluating the effectiveness of online retrieval systems.

Objectives: The objective of this paper is to find out the abrupt transitions between consecutive shots in a video with less false detection and high F1 score. Method/Analysis: This paper presents a video shot boundary detection approach using Gray Level Cooccurrence Matrix (GLCM). The proposed system can roughly be divided into feature extraction using GLCM and the application of the abrupt shot boundary detection. In the first step, the frames are converted into gray level and GLCM is calculated from each frame in the video. Secondly, correlation coefficient is calculated from the GLCM of two consecutive frames of the video. A threshold is set to identify the shot boundaries of the video. The proposed system can detect abrupt transitions effectively with less false detection in the uncompressed domain. Findings: The proposed system can able to achieve an average F1 score of 93.51%, which is achieve due to the reduced false detection. Novelty/ Improvement: The proposed system uses the GLCM matrix directly instead of calculating the contrast, entropy,etc, i.e., the proposed system is purely based on the correlation of the pixel’s co-occurrence. The proposed system also reduces the false detection thereby increasing the precision and F1 score. Keywords: Abrupt, Gradual, Gray Level Cooccurrence Matrix, Shot Boundary Detection,Video Segmentation

To develop and test a scalable, performant, and rule-based model for identifying 3 major domains of social needs (residential instability, food insecurity, and transportation issues) from the unstructured data in electronic health records (EHRs). We included patients aged 18 years or older who received care at the Johns Hopkins Health System (JHHS) between July 2016 and June 2021 and had at least 1 unstructured (free-text) note in their EHR during the study period. We used a combination of manual lexicon curation and semiautomated lexicon creation for feature development. We developed an initial rules-based pipeline (Match Pipeline) using 2 keyword sets for each social needs domain. We performed rule-based keyword matching for distinct lexicons and tested the algorithm using an annotated dataset comprising 192 patients. Starting with a set of expert-identified keywords, we tested the adjustments by evaluating false positives and negatives identified in the labeled dataset. We assessed the performance of the algorithm using measures of precision, recall, and F1 score. The algorithm for identifying residential instability had the best overall performance, with a weighted average for precision, recall, and F1 score of 0.92, 0.84, and 0.92 for identifying patients with homelessness and 0.84, 0.82, and 0.79 for identifying patients with housing insecurity. Metrics for the food insecurity algorithm were high but the transportation issues algorithm was the lowest overall performing metric. The NLP algorithm in identifying social needs at JHHS performed relatively well and would provide the opportunity for implementation in a healthcare system. The NLP approach developed in this project could be adapted and potentially operationalized in the routine data processes of a healthcare system.

As the world rapidly transitions to digitization and cashless transactions, the use of credit cards has surged. Alongside this growth, fraudulent activities have also increased, leading to significant financial losses for institutions. Therefore, it is crucial to identify and assess fraudulent transactions to distinguish them from legitimate ones. In this study, we propose combining Deep Learning with Reinforcement Learning to effectively detect online payment fraud. Specifically, we employ a Long Short-Term Memory recurrent neural network in conjunction with a Deep Q Network, naming this hybrid model DQN-LSTM. A key challenge in this study is achieving high accuracy using an unbalanced credit card fraud dataset, compared to existing methods. Our main goal is to eliminate false positives and false negatives and improve the overall F1 score without using any data balancing techniques. For feature selection from imbalanced datasets, we utilize a deep autoencoder. The deep auto encoder generates accurate data representations that can be used for classification and reconstruction. Experimental results show that the proposed hybrid model consistently outperforms other classifiers as well as state of the arts, achieving better precision, recall and f1 score using the train-test method. Additionally, our model excels in 5-fold cross-validation, demonstrating its robustness and reliability. SHAP (SHapley Additive exPlanations) values enhance model interpretability by quantifying the contribution of each feature to the model's predictions. The SHAP summary plot provides a global view of feature importance, identifying key drivers such as “V19” and “V15” that influence the model's decisions across the European cardholder dataset. IEEE and German datasets are also tested by SHAP. On the other hand, SHAP force plots focus on local interpretability, illustrating how specific features contribute to individual predictions, making it easier to understand why a particular transaction was flagged as fraudulent.

Hypertension is the one of the leading causes of cardiovascular morbidity and mortality. Early detection of individuals at elevated risk is critical, yet conventional prediction models fail to capture nonlinear and high-dimensional interactions among epidemiological and behavioral determinants. Machine learning (ML) offers new opportunities for accurate, population-level risk stratification. We conducted a secondary analysis of the Tamil Nadu STEPS survey 2020, applying supervised ML algorithms—including boosting, k-nearest neighbours, decision tree, random forest, and support vector machine—to predict hypertension risk. Predictors included demographic, anthropometric, lifestyle, and behavioral variables. Models were implemented in JASP, and their performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUC), precision, recall, and F1 score. Among the algorithms tested, the random forest classifier demonstrated the most balanced performance (accuracy 65.5%, AUC 0.708, precision 64.2%, recall 65.5%, F1 score 64.7%). Feature importance analysis identified age as the strongest predictor, followed by waist circumference, while diet and physical activity contributed minimally. The confusion matrix confirmed the model’s balanced sensitivity and specificity, reducing both false negatives and false positives. This study highlights the potential of machine learning, particularly random forest models, for hypertension risk prediction in Indian populations. By leveraging routinely collected survey data, ML can enable scalable, non-invasive screening and inform targeted public health interventions. Integration of richer clinical and genetic features and ensemble methods may further improve predictive accuracy.

Sidhom and Hassoun discuss the crucial role of NLP (Natural Language Processing) tools in Knowledge Extraction and Management as well as in the design of Information Retrieval Systems. The authors focus more specifically on the morpho-syntactic issues by describing their morpho-syntactic analysis platform, which has been implemented to cover the automatic indexing and information retrieval topics. To this end they implemented the Cascaded Augmented Transition Network (ATN). They used this formalism in order to analyse French text descriptions of Multimedia documents. An implementation of an ATN parsing automaton is briefly described. The Platform in its logical operation is considered as an investigative tool towards the knowledge organization (based on an NP -Noun Phrase- recognition model) and management of multiform e(electronic)-documents (text, multimedia, audio, image) using their text descriptions.

Previous studies identified black race as a risk factor for stroke in cancer settings. Yet, no studies have sought to identify the risk factors associated with stroke as a cause of death in black cancer patients. This study aimed to use advanced machine learning techniques to predict stroke mortality in black cancer patients and to identify significant risk factors. We utilized The Surveillance, Epidemiology, and End Results 17 registry to predict stroke as a cause of death in Black cancer patients from January 1, 2000, to December 31, 2020. Various machine learning models were employed, including Random Forest, Logistic Regression, XGBoost, LightGBM, and Gradient Boosting. The dataset included features such as age, sex, cancer type, year of diagnosis, and treatment modalities. Data preprocessing involved cleaning, feature selection, and 80-20 stratified train-test splitting. The models were combined into a voting ensemble to leverage their strengths. The performance of each model was evaluated using metrics including accuracy, AUC (macro, micro, weighted), precision, recall, and F1 scores. Feature importance was analyzed to identify the most contributing variables in predicting stroke death. A total of 292,680 black cancer patients were included, with a mean age of 57.36 ± 13.68 years. The cohort comprised 152,991 males (52.27%) and 139,689 females (47.73%). Among them, 5,874 patients (2.01%) died from stroke, while 286,806 patients are alive. The voting ensemble model achieved an accuracy of 0.98, with an AUC macro of 0.88, AUC micro of 0.99, and AUC weighted of 0.88. The average precision scores were 0.64 (macro), 0.99 (micro), and 0.98 (weighted). The F1 scores were 0.64 (macro), 0.98 (micro), and 0.97 (weighted). The precision scores were 0.78 (macro), 0.98 (micro), and 0.97 (weighted). Significant predictors of stroke death included an earlier year of diagnosis, older age, lack of any cancer treatment, and specific cancer types (notably gastrointestinal and male genital cancers). Our findings suggest that the voting ensemble machine learning model can effectively predict stroke mortality in black cancer patients, with high accuracy and robust performance metrics. These insights could inform targeted interventions to reduce stroke risk in this population, contributing to improved clinical outcomes and survival rates. Future studies should assess the impact of immunotherapy on stroke risk to further refine treatment approaches.

Postpancreatectomy hemorrhage (PPH) is a rare yet dreaded complication following pancreaticoduodenectomy (PD). This retrospective study aimed to explore a machine learning (ML) model for predicting PPH in PD patients. A total of 284 patients who underwent open PD at our institute were included in the analysis. To address the issue of imbalanced data, the adaptive synthetic sampling (ADASYN) technique was employed. The best-performing ML model was selected using the PyCaret library in Python and evaluated based on recall, precision, and F1 score metrics. In addition to assessing the model's performance on the test data, bootstrap validation (n = 1000) with the original dataset was conducted. PPH occurred in 11 patients (3.9%), with a median onset time of 22days postoperatively. These minority cases were oversampled to 85 using ADASYN. The extra trees classifier demonstrated superior performance with recall, precision, and F1 score of 0.967, 0.914, and 0.937, respectively. Both validation using the test data and bootstrap resampling consistently demonstrated recall, precision, and F1 score exceeding 0.9. The model identified the peak value of C-reactive protein during the first 7 postoperative days as the most significant feature, followed by the preoperative neutrophil-to-lymphocyte ratio. This study highlights the potential of the ML approach to predict PPH occurrence following PD. Vigilance and early interventions guided by such model predictions could positively impact outcomes for high-risk patients.

Objective The risk of gastric cancer can be predicted by gastroscopic manifestation recognition and the Kyoto Gastritis Score. This study aims to validate the applicability of AI approaches for recognizing gastroscopic manifestations according to the definition of Kyoto Gastritis Score, with the goal of improving early gastric cancer detection and reducing gastric cancer mortality. Methods In this retrospective study, 29013 gastric endoscopy images were collected and carefully annotated into five categories according to the Kyoto Gastritis Score, i.e. atrophy (A), diffuse redness (DR), enlarged folds (H), intestinal metaplasia (IM), and nodularity (N). As a multi-label recognition task, we propose a deep learning approach composed of five GAM-EfficientNet models, each performing a multiple classification to quantify gastroscopic manifestations, i.e. no presentation or the severity score 0–2. This approach was compared with endoscopists of varying years of experience in terms of accuracy, specificity, precision, recall, and F1 score. Results The approach demonstrated good performance in identifying the five manifestations of the Kyoto Gastritis Score, with an average accuracy, specificity, precision, recall, and F1 score of 78.70%, 91.92%, 80.23%, 78.70%, and 0.78, respectively. The average performance of five experienced endoscopists was 72.63%, 90.00%, 77.68%, 72.63%, and 0.73, while that of five less experienced endoscopists was 66.60%, 87.44%, 70.88%, 66.60%, and 0.66, respectively. The sample t-test indicates that the approach’s average accuracy, specificity, precision, recall, and F1 score for identifying the five manifestations were significantly higher than those of less experienced endoscopists, experienced endoscopists, and all endoscopists on average (p < 0.05). Conclusion Our study demonstrates the potential of deep learning approaches on gastric manifestation recognition over junior, even senior endoscopists. Thus, the deep learning approach holds potential as an auxiliary tool, although prospective validation is still needed to assess its clinical applicability.

Prostate cancer (PC) mortality in Latin America is reported to be twice as high as in developed countries. Combination therapies, such as androgen deprivation therapy (ADT) combined with new hormonal therapies (NHT), are known to improve survival, but their effectiveness in Latin American populations remains understudied. This knowledge limitation can be addressed using artificial intelligence (AI). In this study, an AI and Natural Language Processing (NLP) tool was developed by the Hospital Italiano de Buenos Aires's (HIBA) healthcare AI program in collaboration with the oncology department, to extract, encode, and analyze electronic health record (EHR) data. The data were collected from patients diagnosed with metastatic castration-sensitive prostate cancer (mCSPC) at HIBA, from 2010 to 2020, with a focus on first-line treatment patterns. Performance metrics, including precision, recall, and F1 score, were evaluated for several key variables. For "prostate cancer diagnosis," recall was 0.95, and for "metastasis development detection," recall was 0.91. Regarding "initiation of first-line treatment," precision, recall, and F1 score were 0.94, 0.97, and 0.95, respectively. However, metrics for more complex variables, such as "castration resistance detection," were lower (precision 0.87, recall 0.46, F1 score 0.60). The tool demonstrated adequate recall for metastatic disease diagnosis, although further refinements are needed to enhance performance.

Introduction: Acute coronary syndrome is a time sensitive condition requiring rapid identification, but nurses only identify 40%-50% of high-risk patients at emergency department (ED) triage. Symptoms are the driving force behind prompt ED care of patients with ACS. Symptom data are recorded as free text in nursing narratives. Standardized extraction of symptoms via natural language processing will create usable clinical data that could be used in predictive analytics, to improve early identification of ACS and ensure life-saving treatments. Hypothesis: The Apache Clinical Text Analysis and Knowledge Extraction System (cTAKES) will recognize symptom concepts in nursing narratives with an acceptable F1 score of greater than or equal to 0.9. Methods: A nurse scientist and a physician annotated 1200 ED nursing narratives of patients with suspicion of ACS for the following symptoms predictive of ACS: chest pain, chest discomfort, shortness of breath, fatigue and sweating. Inter-rater reliability (Cohen’s kappa statistic) was calculated with 10% overlap. cTAKES, which is an open-source natural language processing platform, was applied to extract evidence of the 5 annotated symptoms. Recall (sensitivity), precision (positive predictive value) and F1 score (harmonic mean of precision and recall) were calculated to assess extraction quality for each symptom. Results: Recall, precision and F1 score for each symptom are as follows: chest pain: 0.71, 0.97, 0.823; chest discomfort: 0.6, 0.9, 0.72; fatigue: 0.9, 0.26, 0.41; shortness of breath: 0.93, 0.96, 0.95; and sweating: 0.54, 1, 0.7, respectively. Cohen’s kappa was 0.9. Conclusions: cTAKES had fair results across all 5 ACS predictive symptoms, with shortness of breath meeting our hypothesized F1 score goal. Failure to recognize symptoms at a clinically acceptable value might be due to the linguistic and content variation present in ED nursing initial narratives. Customized natural language processing pipelines trained specifically for ED nursing narratives have significant potential for accurately identifying symptoms and improving early identification of ACS.

Distal radius fractures are among the most common upper extremity injuries. While convolutional neural networks (CNNs) have shown promise in fracture detection, no models have specifically addressed the need for reduction, which is a critical distinction for clinical decision-making. Current AI models detect fractures but fall short in guiding treatment, highlighting an opportunity to enhance triage through AI. We retrospectively assembled 495 patients with AP, lateral, and oblique wrist radiographs. To mirror real-world deployment, images were smartphone-style screenshots standardized to 224 × 224 grayscale. We trained BoraeNet, a three-branch DenseNet-169 model (one branch per view) using 5-fold CV on 90% of patients and evaluated a held-out internal test set (n = 50). BoraeNet demonstrated strong performance in classifying fractures needing reduction. Training metrics included accuracy (0.93), precision (0.95), recall (0.91), and F1 score (0.93), with consistent performance across folds. On the held-out internal validation set (n = 50), the model achieved an AUC of 0.89 and average precision of 0.91, classifying 15 true positives, 27 true negatives, 6 false positives, and 2 false negatives. The BoraeNet model offers a promising tool using low-fidelity images to predict need for reduction with strong discrimination, supporting triage in resource-limited settings. Future work should include larger datasets with demographic information, object detection, and multi-center external validation and integration into PACS/clinical workflow. Level of Evidence Level III.