Higher Confidence Scores Research Articles

Collective Decision of One-vs-Rest Networks for Open-Set Recognition.

Unknown examples that are unseen during training often appear in real-world pattern recognition tasks, and an intelligent self-learning system should be able to distinguish between known examples and unknown examples. Accordingly, open-set recognition (OSR), which addresses the problem of classifying knowns and identifying unknowns, has recently been highlighted. However, conventional deep neural networks (DNNs) using a softmax layer are vulnerable to overgeneralization, producing high confidence scores for unknowns. In this article, we propose a simple OSR method that is based on the intuition that the OSR performance can be maximized by setting strict and sophisticated decision boundaries that reject unknowns while maintaining satisfactory classification performance for knowns. For this purpose, a novel network structure, in which multiple one-vs-rest networks (OVRNs) follow a convolutional neural network (CNN) feature extractor, is proposed. Here, an OVRN is a simple feedforward neural network that is designed to assign confidence scores that are lower than those in the softmax layer to unknown samples so that unknown samples can be more effectively separated from known classes. Furthermore, the collective decision score is modeled by combining the multiple decisions reached by the OVRNs to alleviate overgeneralization. Extensive experiments were conducted on various datasets, and the experimental results show that the proposed method performs significantly better than the state-of-the-art methods by effectively reducing overgeneralization. The code is available at https://github.com/JaeyeonJang/Openset-collective-decision.

A Novel Single-Word Speech Recognition on Embedded Systems Using a Convolution Neuron Network with Improved Out-of-Distribution Detection

Advancements in AI have elevated speech recognition, with convolutional neural networks (CNNs) proving effective in processing spectrogram-transformed speech signals. CNNs, with lower parameters and higher accuracy compared to traditional models, are particularly efficient for deployment on storage-limited embedded devices. Artificial neural networks excel in predicting inputs within their expected output range but struggle with anomalies. This is usually harmful to a speech recognition system. In this paper, the neural network classifier for speech recognition is trained with a “negative branch” method, incorporating directional regularization with out-of-distribution training data, allowing it to maintain a high confidence score to the input within distribution while expressing a low confidence score to the anomaly input. It can enhance the performance of anomaly detection of the classifier, addressing issues like misclassifying the speech command that is out of the distribution. The result of the experiment suggests that the accuracy of the CNN model will not be affected by the regularization of the “negative branch”, and the performance of abnormal detection will be improved as the number of kernels of the convolutional layer increases.

MiRNA-protein-metabolite interaction network reveals the regulatory network and players of pregnancy regulation in dairy cows.

Pregnancy is a complex process involving complex molecular interaction networks, such as between miRNA-protein, protein-protein, metabolite-metabolite, and protein-metabolite interactions. Advances in technology have led to the identification of many pregnancy-associated microRNA (miRNA), protein, and metabolite fingerprints in dairy cows. An array of miRNA, protein, and metabolite fingerprints produced during the early pregnancy of dairy cows were described. We have found the in silico interaction networks between miRNA-protein, protein-protein, metabolite-metabolite, and protein-metabolite. We have manually constructed miRNA-protein-metabolite interaction networks such as bta-miR-423-3p-IGFBP2-PGF2α interactomes. This interactome is obtained by manually combining the interaction network formed between bta-miR-423-3p-IGFBP2 and the interaction network between IGFBP2-PGF2α with IGFBP2 as a common interactor with bta-miR-423-3p and PGF2α with the provided sources of evidence. The interaction between bta-miR-423-3p and IGFBP2 has many sources of evidence including a high miRanda score of 169, minimum free energy (MFE) score of -25.14, binding probability (p) of 1, and energy of -25.5. The interaction between IGFBP2 and PGF2α occurs at high confidence scores (≥0.7 or 70%). Interestingly, PGF2α is also found to interact with different metabolites, such as PGF2α-PGD2, PGF2α-thromboxane B2, PGF2α-PGE2, and PGF2α-6-keto-PGF1α at high confidence scores (≥0.7 or 70%). Furthermore, the interactions between C3-PGE2, C3-PGD2, PGE2-PGD2, PGD2-thromboxane B2, PGE2-thromboxane B2, 6-keto-PGF1α-thromboxane B2, and PGE2-6-keto-PGF1α were also obtained at high confidence scores (≥0.7 or 70%). Therefore, we propose that miRNA-protein-metabolite interactomes involving miRNA, protein, and metabolite fingerprints of early pregnancy of dairy cows such as bta-miR-423-3p, IGFBP2, PGF2α, PGD2, C3, PGE2, 6-keto-PGF1 alpha, and thromboxane B2 may form the key regulatory networks and players of pregnancy regulation in dairy cows. This is the first study involving miRNA-protein-metabolite interactomes obtained in the early pregnancy stage of dairy cows.

Towards Generalized Few-Shot Open-Set Object Detection.

Open-set object detection (OSOD) aims to detect the known categories and reject unknown objects in a dynamic world, which has achieved significant attention. However, previous approaches only consider this problem in data-abundant conditions, while neglecting the few-shot scenes. In this paper, we seek a solution for the generalized few-shot open-set object detection (G-FOOD), which aims to avoid detecting unknown classes as known classes with a high confidence score while maintaining the performance of few-shot detection. The main challenge for this task is that few training samples induce the model to overfit on the known classes, resulting in a poor open-set performance. We propose a new G-FOOD algorithm to tackle this issue, named Few-shOt Open-set Detector (FOOD), which contains a novel class weight sparsification classifier (CWSC) and a novel unknown decoupling learner (UDL). To prevent over-fitting, CWSC randomly sparses parts of the normalized weights for the logit prediction of all classes, and then decreases the co-adaptability between the class and its neighbors. Alongside, UDL decouples training the unknown class and enables the model to form a compact unknown decision boundary. Thus, the unknown objects can be identified with a confidence probability without any threshold, prototype, or generation. We compare our method with several state-of-the-art OSOD methods in few-shot scenes and observe that our method improves the F-score of unknown classes by 4.80%-9.08% across all shots in VOC-COCO dataset settings.

Unsupervised Domain Adaptation via Risk-Consistent Estimators

Unsupervised domain adaptation (UDA) attempts to learn domain invariant representations and has achieved significant progress, whereas self-training-based UDA methods have shown powerful performance. However, due to the domain gap, pseudo-labels selected through high confidence scores or uncertainty inevitably contain noise, leading to inaccurate predictions. To address this issue, we propose a novel risk-consistent training method. Specifically, both clean and noisy classifiers are introduced to estimate the noise transition matrix. The clean classifier is exploited to assign pseudo-labels for target data in each iteration. The noisy classifier is then trained with noisy target samples, and the optimal parameters are obtained through a closed-form solution. Heuristically, we also pre-train a domain predictor to select a target-like source example for the noise transition matrix estimation. In addition, we design an uncertainty-guided regularization to generate soft pseudo-labels and avoid overconfident predictions. Extensive experimental results show the effectiveness of our method, and state-of-the-art performance has been achieved. Codes are available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/feifei-cv/RCE.</uri>

A novel method for real-time object-based copy-move tampering localization in videos using fine-tuned YOLO V8

The research community faces challenges for video forgery detection techniques as advancements in multimedia technology have made it easy to alter the original video content and share it on electronic and social media with false propaganda. The copy-move attack is the most commonly practiced type of attack in videos/images, where an object is copied and moved into the current frame or any other frame of the video. Hence an illusion of recreation can be created to forge the content. It is very difficult to differentiate to uncover the forgery traces by the naked eye. Hence, a passive method-based algorithm is proposed to scientifically investigate the statistical properties of the video by normalizing the median difference of the frames at the pixel level, and graphical analysis successfully shows the clear peak in the forged region. After that, a new deep learning approach, “You Only Look at Once”, the latest eighth version of YOLO, is tuned and trained for the localization of forged objects in the real-time domain. The validation and testing results obtained from the trained YOLO V8 are successfully able to detect and localize the forged objects in the videos with mean average precision (mAP) of 0.99, recall is 0.99, precision is 0.99, and highest confidence score. The proposed YOLO V8 is fine-tuned in three different ways, and the performance of the proposed method outperforms existing state-of-the-art techniques in terms of inference speed, accuracy, precision, recall, testing, and training time.

Mapping Chromatin Occupancy of Ppp1r1b-lncRNA Genome-Wide Using Chromatin Isolation by RNA Purification (ChIRP)-seq.

Long non-coding RNA (lncRNA) mediated transcriptional regulation is increasingly recognized as an important gene regulatory mechanism during development and disease. LncRNAs are emerging as critical regulators of chromatin state; yet the nature and the extent of their interactions with chromatin remain to be fully revealed. We have previously identified Ppp1r1b-lncRNA as an essential epigenetic regulator of myogenic differentiation in cardiac and skeletal myocytes in mice and humans. We further demonstrated that Ppp1r1b-lncRNA function is mediated by the interaction with the chromatin-modifying complex polycomb repressive complex 2 (PRC2) at the promoter of myogenic differentiation transcription factors, TBX5 and MyoD1. Herein, we employed unbiased chromatin isolation by RNA purification (ChIRP) and high throughput sequencing to map the repertoire of Ppp1r1b-lncRNA chromatin occupancy genome-wide in the mouse muscle myoblast cell line. We uncovered a total of 99732 true peaks corresponding to Ppp1r1b-lncRNA binding sites at high confidence (p-value < 1E-5) and enrichment score ≥ 10). The Ppp1r1b-lncRNA-binding sites averaged 558 bp in length and were distributed widely within the coding and non-coding regions of the genome. Approximately 46% of these true peaks were mapped to gene elements, of which 1180 were mapped to experimentally validated promoter sequences. Importantly, the promoter-mapped binding sites were enriched in myogenic transcription factors and heart development while exhibiting focal interactions with known motifs of proximal promoters and transcription initiation by RNA Pol-II, including TATA-box, transcription initiator motif, CCAAT-box, and GC-box, supporting Ppp1r1b-lncRNA role in transcription initiation of myogenic regulators. Remarkably, nearly 40% of Ppp1r1b-lncRNA-binding sites mapped to gene introns were enriched with the Homeobox family of transcription factors and exhibited TA-rich motif sequences, suggesting potential motif-specific Ppp1r1b-lncRNA-bound introns. Lastly, more than 136521 enhancer sequences were detected in Ppp1r1b-lncRNA-occupancy sites at high confidence. Among these enhancers, 3390 (12%) exhibited cell type/tissue-specific enrichment in fetal heart and muscles. Together, our findings provide further insights into the genome-wide Ppp1r1b-lncRNA: Chromatin interactome that may dictate its function in myogenic differentiation and potentially other cellular and biological processes.

Defects Detection on 110 MW AC Wind Farm’s Turbine Generator Blades Using Drone-Based Laser and RGB Images with Res-CNN3 Detector

An effective way to perform maintenance on the wind turbine generator (WTG) blades installed in grid-connected wind farms is to inspect them using Unmanned Aerial Vehicles (UAV). The ability to detect wind turbine blade defects from these laser and RGB images captured by drones has been the subject of numerous studies. The issue that most applied techniques battle with is being able to locate different wind turbine blade defects with high confidence scores and precision. The accuracy of these models’ defect detection decreases due to varying testing image scales. This article proposes the Res-CNN3 technique for detecting wind turbine blade defects. In Res-CNN3, defect region detection is achieved through a bipartite process that processes the laser delta and RGB delta structure of a wind turbine blade image with an integration of residual networks and concatenated CNNs to determine the presence of typical defect regions in the image. The loss function is logistic regression, and a Selective Search (SS) algorithm is used to predict the regions of interest (RoI) of the input images for defects detection. Several experiments are conducted, and the outcomes prove that the proposed model has a high prospect for accuracy in solving the problem of defect detection in a manner similar to the advanced benchmark methods.

Bioinformatics strategies to identify differences in molecular biomarkers for ischemic stroke and myocardial infarction.

Ischemic strokes (ISs) are commonly treated by intravenous thrombolysis using a recombinant tissue plasminogen activator; however, successful treatment can only occur within 3 hours after the stroke. Therefore, it is crucial to determine the causes and underlying molecular mechanisms, identify molecular biomarkers for early diagnosis, and develop precise preventive treatments for strokes. We aimed to clarify the differences in gene expression, molecular mechanisms, and drug prediction approaches between IS and myocardial infarction (MI) using comprehensive bioinformatics analysis. The pathogenesis of these diseases was explored to provide directions for future clinical research. The IS (GSE58294 and GSE16561) and MI (GSE60993 and GSE141512) datasets were downloaded from the Gene Expression Omnibus database. IS and MI transcriptome data were analyzed using bioinformatics methods, and the differentially expressed genes (DEGs) were screened. A protein-protein interaction network was constructed using the STRING database and visualized using Cytoscape, and the candidate genes with high confidence scores were identified using Degree, MCC, EPC, and DMNC in the cytoHubba plug-in. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses of the DEGs were performed using the database annotation, visualization, and integrated discovery database. Network Analyst 3.0 was used to construct transcription factor (TF) - gene and microRNA (miRNA) - gene regulatory networks of the identified candidate genes. The DrugBank 5.0 database was used to identify gene-drug interactions. After bioinformatics analysis of IS and MI microarray data, 115 and 44 DEGS were obtained in IS and MI, respectively. Moreover, 8 hub genes, 2 miRNAs, and 3 TFs for IS and 8 hub genes, 13 miRNAs, and 2 TFs for MI were screened. The molecular pathology between IS and MI presented differences in terms of GO and KEGG enrichment pathways, TFs, miRNAs, and drugs. These findings provide possible directions for the diagnosis of IS and MI in the future.

Identification of key miRNAs as regulatory biomarkers of gonadotropins leading to infertility in males

Introduction. Infertility is a highly fatal reproductive system disorder that affects the ability of a couple to reproduce. Over the past decades, a drastic uplift has been recorded in infertility cases among males ranging from 20 to 70 % indicating spermatogenesis impairment.Aim: to identify key microRNAs (miRNAs) as regulatory biomarkers of gonadotropins involved in dysregulation of fertility-related genes to propose potential therapeutic strategies that would combat the action of oncogenic miRNAs (oncomiRs).Materials and Methods. Interaction analysis was performed between miRNAs and fertility-related genes namely luteinizing hormone choriogonadotropin receptor (LHCGR), gonadotropin-releasing hormone receptor (GnRHR), follicle-stimulating hormone receptor (FSHR) and cystic fibrosis transmembrane conductance regulator (CFTR) to identify key miRNAs as regulatory biomarkers of gonadotropins leading to infertility in males.Results. A total of 10, 13, 31 and 18 strong and potential binding sites were predicted for miRNAs-LHCGR, miRNAs-GnRHR, miRNAs-FSHR, and miRNAs-CFTR respectively employing miRWalk (comprehensive genetic database including miRNA targets) followed by identification of 6, 18, 55 and 17 significant interactions through RNA22. Subsequently shortlisted miRNAs and messenger RNA (mRNA) regions were subjected to Vfold-Pipeline and RNAComposer individually for 3D structure prediction. Additionally molecular docking was carried out between miRNAs and mRNAs models that discovered potential and stable interactions elucidating miR-6880-FSHR(R2) as a highly stable complex with least binding affinity (-566.3) and high confidence score (0.999).Conclusion. Hence this study proposes key oncomiRs as a diagnostic biomarker and therapeutic target to bring about a promising treatment strategy against male factor infertility. However wet lab investigations are required for further validations of proposed study.

Knowledge, practice, counseling confidence, and intention to use AAR model of smoking cessation among respiratory therapists: A cross-sectional study.

There is a paucity of research on knowledge, practice, counseling confidence, and intention to use ask, advice, and refer (AAR) model of smoking cessation among respiratory therapists (RTs). Thus, we aimed to analyze the characteristics and factors that may influence them. We collected data using online questionnaires from convenience sample of active licensed RTs in Saudi Arabia. We included 206 participants. A descriptive analysis of the demographic information and characteristics of smoking cessation counseling practices and confidence were conducted. Multiple linear regression was used to test whether demographic variables and AAR model components significantly predicted the RTs' calculated cumulative score of tobacco counseling confidence skills. Our results showed a deficiency in tobacco knowledge among RTs. Most RTs did not have certifications or attend lectures or seminars related to tobacco treatment. RTs were unfamiliar with the smoking cessation program contact information and mobile smoking cessation clinics but reported a high tobacco counseling confidence score. Clinical experience (P = .008), familiarity with smoking cessation program contact information (P = .02), inquiry regarding smoking status (P < .001), and advice regarding smoking status (P = .03) significantly predicted tobacco counseling confidence levels in RTs. RT experience, knowledge, and awareness of smoking cessation programs could enhance the confidence level among them in implementing AAR model.

Robust visual question answering via semantic cross modal augmentation

Recent advances in vision-language models have resulted in improved accuracy in visual question answering (VQA) tasks. However, their robustness remains limited when faced with out-of-distribution data containing unanswerable questions. In this study, we first construct a simple randomised VQA dataset, incorporating unanswerable questions from the VQA v2 dataset, to evaluate the robustness of a state-of-the-art VQA model. Our findings reveal that the model struggles to predict the “unknown” answer or provides inaccurate responses with high confidence scores for irrelevant questions. To address this issue without retraining the large backbone models, we propose Cross Modal Augmentation (CMA), a model-agnostic, test-time-only, multi-modal semantic augmentation technique. CMA generates multiple semantically-consistent but heterogeneous instances from the visual and textual inputs, which are then fed to the model, and the predictions are combined to achieve a more robust output. We demonstrate that implementing CMA enables the VQA model to provide more reliable answers in scenarios involving unanswerable questions, and show that the approach is generalisable across different categories of pre-trained vision language models.

Scorpion Sting Envenomation, a Neglected Tropical Disease: A Nationwide Survey Exploring Perspectives and Attitudes of Resident Doctors from India.

Scorpion sting envenomation (SSE) is a commonly encountered and a significant problem in the tropics, affecting rural and marginalized communities. However, it is not formally listed as a neglected tropical disease (NTD) by the WHO. We designed this cross-sectional study to explore medical graduates' and resident doctors' perspectives on SSE as an NTD and to assess their experiences, knowledge, and confidence in managing these patients. An online questionnaire was developed, validated, and administered to interns and resident doctors across India. Adjusted odds ratios were calculated for factors predicting high self-reported confidence scores for managing scorpion stings using multivariable logistic regression. The final questionnaire contained 26 items including participant background, perspectives about SSE being an NTD, experiences, knowledge, and skills needed to manage, and prevent stings effectively. Of 454 participants, 69% opined that SSE was an NTD, and >60% felt that SSE was inadequately addressed within undergraduate training. Predictors of high self-reported confidence scores in management competencies were residency in a clinical branch that commonly encountered SSE (internal/emergency medicine or pediatrics, P < 0.0001), having ever managed an SSE patient alone or as a part of a team (P < 0.0001), and attending a class or teaching session on SSE during undergraduate training (P = 0.048). Our results suggest that residents across India believe that there is an urgent need to declare SSE an NTD to increase its visibility, further paving the way for innovative multilevel cross-cutting solutions for mitigation. Designing authentic learning experiences can help produce competent and empathetic physicians in managing and preventing SSE.

Assessment of knowledge, attitude, and practice of primary healthcare physicians in Qatar towards hypertension management: An online cross-sectional study.

Primary health care is integral to diagnosing and managing hypertension. This study aimed to assess the knowledge, attitude, practice, priority, and confidence of primary care physicians in Qatar toward hypertension diagnosis and management, and to measure the determinants of good knowledge, desirable attitudes, and desirable practices. We conducted a cross-sectional web-based survey using a modified version of the World Hypertension League Questionnaire during the period from August 30th to October 23rd, 2020. All primary care physicians working in any of Qatar's 27 publicly run primary healthcare centers were invited to participate in the survey. Out of the 450 primary care physicians working at that time, 197 completed the study questionnaire with a response rate of 43.8%. Most respondents stated that they followed local or international treatment guidelines for hypertension management (96.4%). Primary care physicians were highly confident and prioritized hypertension management. The overall desirable practice score was 73.8%, with the highest score (95.4%) for assessing adherence to antihypertensive medications, whereas the lowest desirable practice score was 33.5% for counseling on home blood pressure monitoring. Fifty-one-point eight percentage and 62.4% correctly identified 140 mm Hg and 90 mm Hg as the systolic and diastolic blood pressure threshold for diagnosing hypertension in most patients. The lowest knowledge scores were for hypertension epidemiology in Qatar and recommended dietary modification for hypertensive patients. Respondents had positive attitudes toward task sharing with nonphysician healthcare workers for most items (58.9%-78.2%), except for drug prescriptions (30.5%). Primary care physicians in Qatar had positive attitudes, with high overall confidence and conviction (priority) scores toward hypertension. However, their knowledge scores and certain practices were suboptimal. Healthcare policymakers in Qatar should raise primary care physicians awareness of the current burden of untreated hypertension in the country by organizing targeted educational programs and emphasizing the importance of following national clinical practice guidelines in the diagnosis and management of hypertension.

Automatic Localization of the Pons and Vermis on Fetal Brain MR Imaging Using a U-Net Deep Learning Model.

An MRI of the fetus can enhance the identification of perinatal developmental disorders, which improves the accuracy of ultrasound. Manual MRI measurements require training, time, and intra-variability concerns. Pediatric neuroradiologists are also in short supply. Our purpose was developing a deep learning model and pipeline for automatically identifying anatomic landmarks on the pons and vermis in fetal brain MR imaging and suggesting suitable images for measuring the pons and vermis. We retrospectively used 55 pregnant patients who underwent fetal brain MR imaging with a HASTE protocol. Pediatric neuroradiologists selected them for landmark annotation on sagittal single-shot T2-weighted images, and the clinically reliable method was used as the criterion standard for the measurement of the pons and vermis. A U-Net-based deep learning model was developed to automatically identify fetal brain anatomic landmarks, including the 2 anterior-posterior landmarks of the pons and 2 anterior-posterior and 2 superior-inferior landmarks of the vermis. Four-fold cross-validation was performed to test the accuracy of the model using randomly divided and sorted gestational age-divided data sets. A confidence score of model prediction was generated for each testing case. Overall, 85% of the testing results showed a ≥90% confidence, with a mean error of <2.22 mm, providing overall better estimation results with fewer errors and higher confidence scores. The anterior and posterior pons and anterior vermis showed better estimation (which means fewer errors in landmark localization) and accuracy and a higher confidence level than other landmarks. We also developed a graphic user interface for clinical use. This deep learning-facilitated pipeline practically shortens the time spent on selecting good-quality fetal brain images and performing anatomic measurements for radiologists.

Identification of spoilage yeasts in cashew apple (Annacardium occidentale L.) using MALDI-TOF MS

Cashew (Annacardium occidentale L.) is a hardy and golden crop of arid regions, and the pseudo fruit, cashew apple is a vital source of sugars and vitamins that is discarded as a waste during cashew nut processing. Cashew apples start fermenting soon after the harvest, and due to the presence of a wide spectrum of polyphenols and tannins, the change of colour of cashew apples leads to an unappealing marketability and storage stability. The present study is aimed to rapid identification of different spoilage yeasts and to identify them using MALDI-TOF Mass Spectroscopy. Spoilage yeasts like Candida krusei, C. tropicalis, Pichia norvegensis, Brettannomyces bruxellensis were the spoilage yeasts found to have high confidence score values of 2.00 - 3.00 with high consistency.

Sparse fooling images: Fooling machine perception through unrecognizable images

Fooling images are potential threats to deep neural networks (DNNs). These images cannot be recognized by humans as natural objects, e.g., dogs and cats. However, they are misclassified by DNNs as natural object classes with high confidence scores. Despite their original design concept, existing fooling images, if closely examined, can be seen to retain some features that are characteristic of the target objects. Hence, DNNs can react to these features. In this study, we evaluate whether fooling images with no characteristic pattern of natural objects, either locally or globally, can exist. As a minimal case, we introduce single-color images with a few pixels altered, called sparse fooling images (SFIs). We first prove that SFIs always exist under mild conditions for linear and nonlinear models and reveal that complex models are more likely to be vulnerable to SFI attacks. Using two SFI generation methods, we demonstrate that in deeper layers, SFIs have features similar to those of natural images. Therefore, they fool DNNs successfully. Among the other layers, we discover that the max-pooling layer causes vulnerability to SFIs. The defense against SFIs and transferability are also discussed. This study highlights a new vulnerability of DNNs by introducing a novel class of images that are distributed extremely far from natural images.

Using Deep Learning to Detect the Presence and Location of Hemoperitoneum on the Focused Assessment with Sonography in Trauma (FAST) Examination in Adults.

Abdominal ultrasonography has become an integral component of the evaluation of trauma patients. Internal hemorrhage can be rapidly diagnosed by finding free fluid with point-of-care ultrasound (POCUS) and expedite decisions to perform lifesaving interventions. However, the widespread clinical application of ultrasound is limited by the expertise required for image interpretation. This study aimed to develop a deep learning algorithm to identify the presence and location of hemoperitoneum on POCUS to assist novice clinicians in accurate interpretation of the Focused Assessment with Sonography in Trauma (FAST) exam. We analyzed right upper quadrant (RUQ) FAST exams obtained from 94 adult patients (44 confirmed hemoperitoneum) using the YoloV3 object detection algorithm. Exams were partitioned via fivefold stratified sampling for training, validation, and hold-out testing. We assessed each exam image-by-image using YoloV3 and determined hemoperitoneum presence for the exam using the detection with highest confidence score. We determined the detection threshold as the score that maximizes the geometric mean of sensitivity and specificity over the validation set. The algorithm had 95% sensitivity, 94% specificity, 95% accuracy, and 97% AUC over the test set, significantly outperforming three recent methods. The algorithm also exhibited strength in localization, while the detected box sizes varied with a 56% IOU averaged over positive cases. Image processing demonstrated only 57-ms latency, which is adequate for real-time use at the bedside. These results suggest that a deep learning algorithm can rapidly and accurately identify the presence and location of free fluid in the RUQ of the FAST exam in adult patients with hemoperitoneum.

CARVING-DETC: A network scaling and NMS ensemble for Balinese carving motif detection method

Balinese carvings are cultural objects that adorn sacred buildings. The carvings consist of several motifs, each representing the values adopted by the Balinese people. Detection of Balinese carving motifs is challenging due to the unavailability of a Balinese carving dataset for detection tasks, high variance, and tiny-size carving motifs. This research aims to improve carving motif detection performance on challenging Balinese carving motifs detection task through a modification of YOLOv5 to support a digital carving conservation system. We proposed CARVING-DETC, a deep learning-based Balinese carving detection method consisting of three steps. First, the data generation step performs data augmentation and annotation on Balinese carving images. Second, we proposed a network scaling strategy on the YOLOv5 model and performed non-maximum suppression (NMS) on the model ensemble to generate the most optimal predictions. The ensemble model utilizes NMS to produce higher performance by optimizing the detection results based on the highest confidence score and suppressing other overlap predictions with a lower confidence score. Third, performance evaluation on scaled-YOLOv5 versions and NMS ensemble models. The research findings are beneficial in conserving the cultural heritage and as a reference for other researchers. In addition, this study proposed a novel Balinese carving dataset through data collection, augmentation, and annotation. To our knowledge, it is the first Balinese carving dataset for the object detection task. Based on experimental results, CARVING-DETC achieved a detection performance of 98%, which outperforms the baseline model.

OPTIMISING HOLISTIC CARE OF WOMEN AT RISK OF OR LIVING WITH BIA‐ALCL: A SURVEY OF UK HEALTH CARE PROFESSIONALS

Introduction: Breast implant-associated anaplastic large cell lymphoma (BIA-ALCL) was classified by the World Health Organisation in 2016. As of December 2021 there have been 86 reported cases within the UK. Raising awareness amongst clinicians of this novel entity is crucial in order to provide effective clinical care. Psycho-social support forms a vital arm of care, both for affected women but also for the now millions at risk or those considering new implants. This survey aims to assess the current awareness and experiences of health care providers potentially involved in the care of patients with BIA-ALCL with a particular focus on psycho-social support requirements. Methods: We performed an online survey using categorical and open-ended questions. The questionnaire was distributed amongst professional associations that have an interest in the topic researched, e.g., The Association of Breast Surgeons (ABS), British Association of Plastic, Reconstructive and Aesthetic Surgeons (BAPRAS). In addition the survey was distributed amongst multidisciplinary teams that are likely to encounter these patients, e.g. Breast and Lymphoma. The survey was designed in Qualtrics and sent via a link. Data has been collated and analysed within SPSS. Research approval was obtained from Leeds Beckett University. Results: There were a total of 78 respondents who completed the survey. Breast surgeons made up the largest proportion of respondents (36%, 28/78) with over half of respondents having cared for a patient with BIA-ALCL (58% 42/78). The largest proportion of respondents reporting using ABS/BAPRAS joint guidelines in their clinical practice (46%, 36/78). High median confidence scores were reported in both diagnosing/managing, 7/10 (3) and having conversations with patients regarding BIA-ALCL, 7/10 (5) although a large range in responses was noted. The most frequently selected test to work up a case of suspected BIA-ALCL was ‘core biopsy of any associated peri-implant mass’ (82%, 64/78) with mammography the least selected (42%, 33/78). 45% (35/78) reported seeing women requesting reassurance or implant removal despite an absence of symptoms or signs of BIA-ALCL. The average degree of distress upon receiving a diagnosis was reported as 7/10 (2). 70% (54/77) of respondents felt patients diagnosed with BIA-ALCL or those awaiting test results would benefit from psychological support. Only 38 (29/76) % of respondents however, confirmed that breast implant patients have access to any psychological support as part of routine care. Conclusion: This survey demonstrates that although there is relatively high confidence in managing and conducting discussions with patients regarding BIA-ALCL, there is a large variability amongst healthcare providers. The findings suggest there is urgent need for psycho-social support for both patients diagnosed with BIA-ALCL and asymptomatic women seeking reassurance. Keyword: Aggressive T-cell non-Hodgkin lymphoma Conflicts of interests pertinent to the abstract D. Cunningham Consultant or advisory role Ovibio on scientific advisory board Research funding: Medimmune/AZ, Clovis, Eli Lilly, 4SC, Bayer, Celgene, Roche - institution recipient