Manual Annotation Research Articles

Text mining approach for feature extraction and cartilage disease grade classification using knee MRI radiology reports

MRI radiology reporting processes can be improved by exploiting structured and semantically labelled data that can be fed to artificial intelligence (AI) tools. AI-based tools assisting radiology reporting can help to automatically individuate cartilage grading in textual magnetic resonance imaging (MRI) reports, thus supporting clinicians' decisions regarding medical imaging utilisation, diagnosis and treatment. In this study, we extracted information (clinical findings, observations, anatomical regions, etc.) and classified knee cartilage degradation from medical reports utilising transfer-learning techniques applied to the Bidirectional Encoder Representations from Transformers (BERT) model and its variants, pre-trained on an Italian-language corpus. To realise this objective, we used a dataset of 750 MRI knee reports written by three radiologists who contributed to a manual annotation process to perform text classification (TC) and named entity recognition (NER) tasks. The dataset was obtained from an internal database of the IRCCS SYNLAB SDN. Seventy percent of the dataset was used for training, 10% was used for validation and 20% was used for testing. The best-performing configurations for NER and TC tasks were based on the pre-trained BERT model. The macro F1-scores obtained with the NER and TC models are 0.89 and 0.81, respectively. The accuracies calculated on the test set for both tasks are 0.96 and 0.99, respectively.

Evaluation of the impact of cardiac implantable electronic devices on cine MRI for real-time adaptive cardiac radioablation on a 1.5 T MR-linac.

Stereotactic arrhythmia radioablation (STAR) is a novel treatment approach for refractory ventricular tachycardia (VT). The risk of treatment-induced toxicity and geographic miss can be reduced with online MRI-guidance on an MR-linac. However, most VT patients carry cardiac implantable electronic devices (CIED), which compromise MRimages. Robust MR-linac imaging sequences are required for cardiac visualization and accurate motion monitoring in presence of a CIED during MRI-guided STAR. We optimized two clinically available cine sequences for cardiorespiratory motion estimation in presence of a CIED on a 1.5 T MR-linac. The image quality, motion estimation accuracy, and geometric fidelity using these cine sequences wereevaluated. Clinically available 2D balanced steady-state free precession (bSSFP, voxel size = 3.0 3.0 10 mm3, Tscan = 96 ms, bandwidth (BW) = 1884 Hz/px) and -spoiled gradient echo ( -GRE, voxel size = 4.0 4.0 10 mm3, Tscan = 97 ms, BW = 500 Hz/px) sequences were adjusted for real-time cardiac visualization and cardiorespiratory motion estimation on a 1.5 T Unity MR-linac (Elekta AB, Stockholm, Sweden), while complying with safety guidelines for MRI in presence of CIEDs (specific absorption rate 2 W/kg and 80 mT/s). Cine acquisitions were performed in five healthy volunteers, with and without an implantable cardioverter- defibrillator (ICD) placed on the clavicle, and a VT patient. Generalized divergence-curl (GDC) deformable image registration (DIR) was used for automated landmark motion estimation in the left ventricle (LV). Gaussian processes (GP), a machine-learning technique, was trained using GDC landmarks and deployed for real-time cardiorespiratory motion prediction. -mapping was performed to assess geometric image fidelity in the presence ofCIEDs. CIEDs introduced banding artifacts partially obscuring cardiac structures in bSSFP acquisitions. In contrast, the -GRE was more robust to CIED-induced artifacts at the expense of a lower signal-to-noise ratio. In presence of an ICD, image-based cardiorespiratory motion estimation was possible for 85% (100%) of the volunteers using the bSSFP ( -GRE) sequence. The in-plane 2D root-mean-squared deviation (RMSD) range between GDC-derived landmarks and manual annotations using the bSSFP (T1-GRE) sequence was 3.1-3.3 (3.3-4.1) mm without ICD and 4.6-4.6 (3.2-3.3) mm with ICD. Without ICD, the RMSD between the GP-predictions and GDC-derived landmarks ranged between 0.9 and 2.2 mm (1.3-3.0 mm) for the bSSFP (T1-GRE) sequence. With ICD, the RMSD between the GP-predictions and GDC-derived landmarks ranged between 1.3 and 2.2 mm (1.2-3.2 mm) using the bSSFP (T1-GRE) sequence resulting in an RMSD-increase of 42%-143% (bSSFP) and -61%-142% (T1-GRE). Lead-induced spatial distortions ranged between -0.2 and 0.2 mm (-0.7-1.2 mm) using the bSSFP ( -GRE) sequence. The 98th percentile range of the spatial distortions in the gross target volume of the patient was between 0.0 and 0.4 mm (0.0-1.8 mm) when using bSSFP ( -GRE). Tailored bSSFP and -GRE sequences can facilitate real-time cardiorespiratory estimation using GP trained with GDC-derived landmarks in the majority of landmark locations in the LV despite the presence of CIEDs. The need for high temporal resolution noticeably reduced achievable spatial resolution of the cine MRIs. However, the effect of the CIED-induced artifacts is device, patient and sequence dependent and requires specific assessment percase.

Evaluating large language models in analysing classroom dialogue

This study explores the use of Large Language Models (LLMs), specifically GPT-4, in analysing classroom dialogue—a key task for teaching diagnosis and quality improvement. Traditional qualitative methods are both knowledge- and labour-intensive. This research investigates the potential of LLMs to streamline and enhance this process. Using datasets from middle school mathematics and Chinese classes, classroom dialogues were manually coded by experts and then analysed with a customised GPT-4 model. The study compares manual annotations with GPT-4 outputs to evaluate efficacy. Metrics include time efficiency, inter-coder agreement, and reliability between human coders and GPT-4. Results show significant time savings and high coding consistency between the model and human coders, with minor discrepancies. These findings highlight the strong potential of LLMs in teaching evaluation and facilitation.

Generative AI-Driven Data Augmentation for Crack Detection in Physical Structures

The accurate segmentation of cracks in structural materials is crucial for assessing the safety and durability of infrastructure. Although conventional segmentation models based on deep learning techniques have shown impressive detection capabilities in these tasks, their performance can be restricted by small amounts of training data. Data augmentation techniques have been proposed to mitigate the data availability issue; however, these systems often have limitations in texture diversity, scalability over multiple physical structures, and the need for manual annotation. In this paper, a novel generative artificial intelligence (GAI)-driven data augmentation framework is proposed to overcome these limitations by integrating a projected generative adversarial network (ProjectedGAN) and a multi-crack texture transfer generative adversarial network (MCT2GAN). Additionally, a novel metric is proposed to evaluate the quality of the generated data. The proposed method is evaluated using three datasets: the bridge crack library (BCL), DeepCrack, and Volker. From the simulation results, it is confirmed that the segmentation performance can be improved by the proposed method in terms of intersection over union (IoU) and Dice scores across three datasets.

BrainQCNet: A Deep Learning attention-based model for the automated detection of artifacts in brain structural MRI scans

Abstract Analyses of structural MRI (sMRI) data depend on robust upstream data quality control (QC). It is also crucial that researchers seek to retain maximal amounts of data to ensure reproducible, generalizable models and to avoid wasted effort, including that of participants. The time-consuming and difficult task of manual QC evaluation has prompted the development of tools for the automatic assessment of brain sMRI scans. Existing tools have proved particularly valuable in this age of Big Data; as datasets continue to grow, reducing execution time for QC evaluation will be of considerable benefit. The development of Deep Learning (DL) models for artifact detection in structural MRI scans offers a promising avenue toward fast, accurate QC evaluation. In this study, we trained an interpretable Deep Learning model, ProtoPNet, to classify minimally preprocessed 2D slices of scans that had been manually annotated with a refined quality assessment (ABIDE 1; n = 980 scans). To evaluate the best model, we applied it to 2141 ABCD T1-weighted MRI scans for which gold-standard manual QC annotations were available. We obtained excellent accuracy: 82.4% for good quality scans (Pass), 91.4% for medium to low quality scans (Fail). Further validation using 799 T1w MRI scans from ABIDE 2 and 750 T1w MRI scans from ADHD-200 confirmed the reliability of our model. Accuracy was comparable to or exceeded that of existing ML models, with fast processing and prediction time (1 minute per scan, GPU machine, CUDA-compatible). Our attention model also performs better than traditional DL (i.e., convolutional neural network models) in detecting poor quality scans. To facilitate faster and more accurate QC prediction for the neuroimaging community, we have shared the model that returned the most reliable global quality scores as a BIDS-app (https://github.com/garciaml/BrainQCNet).

Automating field‐based floral surveys with machine learning

Abstract The abundance and diversity of flowering plant species are important indicators of pollinator habitat quality, but traditional field‐based surveying techniques are time‐intensive. Therefore, they are often biased due to under‐sampling and are difficult to scale. Aerial photography was collected across 10 sites located in and around Rouge National Urban Park, Toronto, Canada using a consumer‐grade drone. A convolutional neural network (CNN) was trained to semantically segment, or identify and categorize, pixel clusters which represent flowers in the collected aerial imagery. Specifically, flowers of the dominant taxa found in the depauperate fall flowering plant community were surveyed. This included yellow flowering Solidago spp., white Symphyotrichum ericoides/lanceolatum and purple Symphyotrichum novae‐angliae. The CNN was trained using 930 m2 of manually annotated data, ~1% of the mapped landscape. The trained CNN was tested on 20% of the manually annotated data concealed during training. In addition, it was externally validated by comparing the predicted drone‐derived floral abundance metrics (i.e. floral area (m2) and the number of floral patches) to the field‐based count of floral units estimated for 34 4 m2 plots. The CNN returned accurate multiclassification when evaluated against the testing data. It obtained a precision score of 0.769, a recall of 0.849, and an F1 score of 0.807. The automated floral abundance counting yielded estimates that were strongly correlated with field‐based manual counting. In addition, flower segmentation using the trained CNN was time‐efficient. On average, it took roughly the same amount of time to segment the flowers occurring in an entire drone scene as it took to complete the abundance count of a single quadrat. However, the training process, particularly manual data annotation, was the most time‐consuming component of the study. Practical implication: Overall, the analysis provided valuable insights into automated flower classification and abundance estimation using drone imagery and machine learning. The results demonstrate that these tools can be used to provide accurate and scalable estimates of pollinator habitat quality. Further research should consider diverse wildflower systems to develop the generalizability of the methods.

Variation in forest root image annotation by experts, novices, and AI

BackgroundThe manual study of root dynamics using images requires huge investments of time and resources and is prone to previously poorly quantified annotator bias. Artificial intelligence (AI) image-processing tools have been successful in overcoming limitations of manual annotation in homogeneous soils, but their efficiency and accuracy is yet to be widely tested on less homogenous, non-agricultural soil profiles, e.g., that of forests, from which data on root dynamics are key to understanding the carbon cycle. Here, we quantify variance in root length measured by human annotators with varying experience levels. We evaluate the application of a convolutional neural network (CNN) model, trained on a software accessible to researchers without a machine learning background, on a heterogeneous minirhizotron image dataset taken in a multispecies, mature, deciduous temperate forest.ResultsLess experienced annotators consistently identified more root length than experienced annotators. Root length annotation also varied between experienced annotators. The CNN root length results were neither precise nor accurate, taking ~ 10% of the time but significantly overestimating root length compared to expert manual annotation (p = 0.01). The CNN net root length change results were closer to manual (p = 0.08) but there remained substantial variation.ConclusionsManual root length annotation is contingent on the individual annotator. The only accessible CNN model cannot yet produce root data of sufficient accuracy and precision for ecological applications when applied to a complex, heterogeneous forest image dataset. A continuing evaluation and development of accessible CNNs for natural ecosystems is required.

Translation of a story into Chilean Sign Language: productions by deaf co-teachers

IntroductionThe incorporation of narratives in Chilean Sign Language enriches classroom practices, preserving and transmitting Deaf culture. However, that within the Chilean educational context, the narratives used for instructing Deaf students are translations from written Spanish. This study aims to describe and analyze the narrative structure and the use of Highly Iconic Structures in these translations conducted by Deaf co-teachers into Chilean Sign Language.MethodThe research adopts a qualitative approach and a descriptive case study design, involving two deaf teachers from an Inclusive Education Program in a school. The data analyzed focus on the video-recorded LSCh translations of the short story “The Greedy Squirrel”. Manual annotations were made on the corpus gloss transcription to first segment by narrative structure for description and then to identify the transfer operations.ResultsThe co-teachers make variations to the narrative structure of the translated text and incorporate specific visual-gestural elements such as the change of narrator to first person; the 45% of the translation is composed of transfer operations.DiscussionThese findings suggest that the variations in structure and the new visual-gestural information provided by the co-teachers reflect their understanding and use of sign language narrative norms for translation. The use of transfer operations enhances the storytelling experience, although it cannot be conclusively stated how and why teachers decide to use them in translation. The results are limited, as they require comparison with other corpora on translations from written Spanish to LSCh and software-assisted analysis to standardize data, which implies further research on the topic. For now, it is essential that the time allocated to the preparation of educational materials in LSCh be more extensive and better planned.

Revisiting the functional annotation of TriTryp using sequence similarity tools

Trypanosomatids are the causative agents of deadly diseases in humans and livestock. Given the high phylogenetic distance of trypanosomatids from model organisms, these organisms have ample unannotated genes. Manual functional annotation is time-consuming, highlighting the importance of automated functional annotation tools. The development of automated functional tools is a hot research topic, and multiple tools have been developed for the task. PANNZER2 is an automated functional annotation tool that merely relies on the sequence similarity of the query to the annotated proteins. We tried PANNZER2 on Trypanosoma brucei, the most studied organism among trypanosomatids, to see if it could improve our knowledge of the functions of the genes.Even with the availability of automated annotation tools like InterPro2GO in databases such as TriTrypDB, PANNZER2 has made surprisingly confident predictions for some hypothetical proteins in T. brucei. In this study, we identify gaps in such annotations because of not employing pairwise sequence alignment tools in TriTrypDB's automated annotation process. Our findings demonstrate that even the use of stringent cutoffs can successfully annotate a significant number of proteins. Additionally, we discovered that adjusting the open reading frames in certain genes leads to sequences with increased sequence signature coverage—characterized by the length covered by at least one sequence signature—compared to the original sequences. This enhanced sequence signature coverage suggests these genomic fragments could be pseudogenes. To facilitate further exploration, we developed a script to help identify potential pseudogenes within an organism's genome, offering researchers a new tool for genomic analysis and understanding. We extended all our analysis to Trypanosoma cruzi and Leishmania major to assess the impact of this approach across different species.Our study demonstrates that by utilizing pairwise sequence similarity alignment, even with stringent cutoffs, we can attribute 2986, 3953, and 3798 new GO terms to the genomes of T. brucei, T. cruzi, and L. major. Additionally, we found that 210, 239, and 29 genes exhibit increased sequence signature coverage following frame correction, suggesting the presence of pseudogenes.

Clinical knowledge-guided hybrid classification network for automatic periodontal disease diagnosis in X-ray image

Accurate classification of periodontal disease through panoramic X-ray images carries immense clinical importance for effective diagnosis and treatment. Recent methodologies attempt to classify periodontal diseases from X-ray images by estimating bone loss within these images, supervised by manual radiographic annotations for segmentation or keypoint detection. However, these annotations often lack consistency with the clinical gold standard of probing measurements, potentially causing measurement inaccuracy and leading to unstable classifications. Additionally, the diagnosis of periodontal disease necessitates exceptional sensitivity. To address these challenges, we introduce HC-Net, an innovative hybrid classification framework devised for accurately classifying periodontal disease from X-ray images. This framework comprises three main components: tooth-level classification, patient-level classification, and a learnable adaptive noisy-OR gate. In the tooth-level classification, we initially employ instance segmentation to individually identify each tooth, followed by tooth-level periodontal disease classification. For patient-level classification, we utilize a multi-task strategy to concurrently learn patient-level classification and a Classification Activation Map (CAM) that signifies the confidence of local lesion areas within the panoramic X-ray image. Eventually, our adaptive noisy-OR gate acquires a hybrid classification by amalgamating predictions from both levels. In particular, we incorporate clinical knowledge into the workflows used by professional dentists, targeting the enhanced handling of sensitivity of periodontal disease diagnosis. Extensive empirical testing on a dataset amassed from real-world clinics demonstrates that our proposed HC-Net achieves unparalleled performance in periodontal disease classification, exhibiting substantial potential for practical application.

MedLSAM: Localize and segment anything model for 3D CT images

Recent advancements in foundation models have shown significant potential in medical image analysis. However, there is still a gap in models specifically designed for medical image localization. To address this, we introduce MedLAM, a 3D medical foundation localization model that accurately identifies any anatomical part within the body using only a few template scans. MedLAM employs two self-supervision tasks: unified anatomical mapping (UAM) and multi-scale similarity (MSS) across a comprehensive dataset of 14,012 CT scans. Furthermore, we developed MedLSAM by integrating MedLAM with the Segment Anything Model (SAM). This innovative framework requires extreme point annotations across three directions on several templates to enable MedLAM to locate the target anatomical structure in the image, with SAM performing the segmentation. It significantly reduces the amount of manual annotation required by SAM in 3D medical imaging scenarios. We conducted extensive experiments on two 3D datasets covering 38 distinct organs. Our findings are twofold: (1) MedLAM can directly localize anatomical structures using just a few template scans, achieving performance comparable to fully supervised models; (2) MedLSAM closely matches the performance of SAM and its specialized medical adaptations with manual prompts, while minimizing the need for extensive point annotations across the entire dataset. Moreover, MedLAM has the potential to be seamlessly integrated with future 3D SAM models, paving the way for enhanced segmentation performance. Our code is public at https://github.com/openmedlab/MedLSAM.

Using Artificial Intelligence to Support Informed Decision-Making on BRAF Mutation Testing.

Precision oncology relies on accurate and interpretable reporting of testing and mutation rates. Focusing on the BRAFV600 mutations in advanced colorectal carcinoma, non-small-cell lung carcinoma, and cutaneous melanoma, we developed a platform displaying testing and mutation rates reported in the literature, which we annotated using an artificial intelligence (AI) and natural language processing (NLP) pipeline. Using AI, we identified publications that likely reported a testing or mutation rate, filtered publications for cancer type, and identified sentences that likely reported rates. Rates and covariates were subsequently manually curated by three experts. The AI performance was evaluated using precision and recall metrics. We used an interactive platform to explore and present the annotated testing and mutation rates by certain study characteristics. The interactive dashboard, accessible at the BRAF dimensions website, enables users to filter mutation and testing rates with relevant options (eg, country of study, study type, mutation type) and to visualize annotated rates. The AI pipeline demonstrated excellent filtering performance (>90% precision and recall for all target cancer types) and moderate performance for sentence classification (53%-99% precision; ≥75% recall). The manual annotation of testing and mutation rates revealed inter-rater disagreement (testing rate, 19%; mutation rate, 70%), indicating unclear or nonstandard reporting of rates in some publications. Our AI-driven NLP pipeline demonstrated the potential for annotating biomarker testing and mutation rates. The difficulties we encountered highlight the need for more advanced AI-powered literature searching and data extraction, and more consistent reporting of testing rates. These improvements would reduce the risk of misinterpretation or misunderstanding of testing and mutation rates by AI-based technologies and the health care community, with beneficial impacts on clinical decision-making, research, and trial design.

Improved ground truth annotation by multimodal image registration from 3D ultrasound to histopathology for resected tongue carcinoma.

This study's objectives are (1) to investigate the registration accuracy from intraoperative ultrasound (US) to histopathological images, (2) to assess the agreement and correlation between measurements in registered 3D US and histopathology, and (3) to train a nnUNet model for automatic segmentation of 3D US volumes of resected tongue specimens. Ten 3D US volumes were acquired, including the corresponding digitalized histopathological images (n = 29). Based on corresponding landmarks, the registrations between 3D US and histopathology images were calculated and evaluated using the target registration error (TRE). Tumor thickness and resection margins were measured based on three annotations: (1) manual histopathological tumor annotation (HTA), manual 3D US tumor annotation, and (2) the HTA registered in the 3D US. The agreement and correlation were computed between the measurements based on the HTA and those based on the manual US and registered HTA in US. A deep-learning model with nnUNet was trained on 151 3D US volumes. Segmentation metrics quantified the model's performance. The median TRE was 0.42mm. The smallest mean difference was between registered HTA in US and histopathology with 2.16mm (95% CI - 1.31; 5.63) and a correlation of 0.924 (p < 0.001). The nnUNet predicted the tumor with a Dice similarity coefficient of 0.621, an average surface distance of 1.15mm, and a Hausdorff distance of 3.70mm. Multimodal image registration enabled the HTA's registration in the US images and improved the agreement and correlation between the modalities. In the future, this could be used to annotate ground truth labels accurately.

Hierarchical Semi-Supervised Approach for Classifying Activities of Workers Utilising Indoor Trajectory Data

Activity recognition refers to the process of automatically identifying or interpreting activities of objects based on the data captured from different sensing devices. While previous research on indoor activity recognition predominantly relies on visual data like images or video recordings, we present a novel approach based on spatiotemporal trajectory data recorded by IoT based sensors. The proposed approach is tailored for indoor manufacturing applications leveraging trajectory partitioning, hierarchical clustering, and convolutional neural networks. Moreover, a vast majority of activity recognition models that have been used in different industrial settings are supervised methods requiring large manually labelled datasets to be collected. This manual annotation process is unwieldly and labour-intensive, and hence, often infeasible to deploy for practical applications. In contrast, our proposed activity recognition approach is semi-supervised meaning it can be trained with far less labelled data; significantly reducing the effort and costs associated with the manual annotation process. The proposed approach is evaluated using two indoor trajectory datasets related to different manufacturing assembly processes. Experimental results demonstrate its effectiveness for activity recognition: the classification accuracy (measured using F-score) varies between 0.81 to 0.95 and 0.88 to 0.92 across indoor trajectory datasets. A comparison with a baseline model indicates that it achieves up to a 18% improvement in classification accuracy. Furthermore, the classification results enable insights into factory floor states, aiding in decision-making for operational efficiency and resource allocation.

Few-Shot Symbol Detection in Engineering Drawings

ABSTRACT Recently, there has been significant interest in digitizing engineering drawings due to their complexity and practical benefits. Symbol digitization, a critical aspect in this field, is challenging as utilizing Deep Learning-based methods to recognize symbols of interest requires a large number of training instances for each class of symbols. Acquiring and annotating sufficient diagrams is difficult due to concerns about confidentiality and availability. The conventional manual annotation process is time-consuming, costly, and prone to human error. Additionally, obtaining an adequate number of samples for rare classes proves to be exceptionally challenging. This paper introduces a few-shot framework to address these challenges. Several experiments with fewer than ten, and sometimes just one, training instance per class using complex engineering drawings from industry sources were carried out. The results suggest that our method not only significantly improves symbol detection performance compared to other state-of-the-art methods but also decreases the necessary number of training instances.

A Corpus-based Behavioral Profile Study of Synonymous Expressions Meaning “to Start”

Synonymy is an extremely significant component in language learning, and the usage of synonymy is rather flexible and difficult to distinguish. In this research, the author employs a corpus-based behavioral profile approach to study the usage of a set of English synonymous expressions meaning “to start”, namely, start, begin, commence, launch, set out. The author first searches in American Contemporary English Corpus to explore the overall usage frequency and specific genre of each synonymous expression, then randomly selects 500 sentences, and makes manual annotation with 12 ID tag and 46 ID tag levels. Later, hierarchical clustering analysis and correspondence analysis have been employed to examine the semantic differences and usage structures of this set of expressions.The results shows that: ① The overall usage frequency varies greatly—from the highest to the lowest are start, begin, launch, set out, commence. ② They are used in various genres. Start is mainly used in informal styles including blogs, while the other four are mainly used in formal styles, including fictions and magazines. ③ Begin and set out have the greatest semantic similarity; start, begin, and set out form one semantic branch, and commence and launch form another, and there is a huge difference between these two branches. ④ There are great differences between the five expressions and their morphological, syntactic and semantic changes.Quantitative analysis method has been used to reveal the different semantic features and usage of a set of synonymous expressions, which is conducive to both promote the study of synonymous relations, and verify the effectiveness and applicability of the behavioral profile approach in semantic study.

Robust and smooth Couinaud segmentation via anatomical structure-guided point-voxel network

Precise Couinaud segmentation from preoperative liver computed tomography (CT) is crucial for surgical planning and lesion examination. However, this task is challenging as it is defined based on vessel structures, and there is no intensity contrast between adjacent Couinaud segments in CT images. To solve this challenge, we design a multi-scale point-voxel fusion framework, which can more effectively model the spatial relationship of points and the semantic information of the image, producing robust and smooth Couinaud segmentations. Specifically, we first segment the liver and vessels from the CT image and generate 3D liver point clouds and voxel grids embedded with the vessel structure. Then, our method with two input-specific branches extracts complementary feature representations from points and voxels, respectively. The local attention module adaptively fuses features from the two branches at different scales to balance the contribution of different branches in learning more discriminative features. Furthermore, we propose a novel distance loss at the feature level to make the features in the segment more compact, thereby improving the certainty of segmentation between segments. Our experimental results on three public liver datasets demonstrate that our proposed method outperforms several state-of-the-art methods by large margins. Specifically, in out-of-distribution (OOD) testing of LiTS dataset, our method exceeded the voxel-based 3D UNet by approximately 20% in Dice score, and outperformed the point-based PointNet2Plus by approximately 8% in Dice score. Our code and manual annotations of the public datasets presented in this paper are available online: https://github.com/xukun-zhang/Couinaud-Segmentation.

Evaluation of emotion classification schemes in social media text: an annotation-based approach

BackgroundEmotion analysis of social media texts is an innovative method for gaining insight into the mental state of the public and understanding social phenomena. However, emotion is a complex psychological phenomenon, and there are various emotion classification schemes. Which one is suitable for textual emotion analysis?MethodsWe proposed a framework for evaluating emotion classification schemes based on manual annotation experiments. Considering both the quality and efficiency of emotion analysis, we identified five criteria, which are solidity, coverage, agreement, compactness, and distinction. Qualitative and quantitative factors were synthesized using the AHP, where quantitative metrics were derived from annotation experiments. Applying this framework, 2848 Sina Weibo posts related to public events were used to evaluate the five emotion schemes: SemEval’s four emotions, Ekman’s six basic emotions, ancient China’s Seven Emotions, Plutchik’s eight primary emotions, and GoEmotions’ 27 emotions.ResultsThe AHP evaluation result shows that Ekman’s scheme had the highest score. The multi-dimensional scaling (MDS) analysis shows that Ekman, Plutchik, and the Seven Emotions are relatively similar. We analyzed Ekman’s six basic emotions in relation to the emotion categories of the other schemes. The correspondence analysis shows that the Seven Emotions’ joy aligns with Ekman’s happiness, love demonstrates a significant correlation with happiness, but desire is not significantly correlated with any emotion. Compared to Ekman, Plutchik has two more positive emotions: trust and anticipation. Trust is somewhat associated with happiness, but anticipation is weakly associated with happiness. Each emotion of Ekman’s corresponds to several similar emotions in GoEmotions. However, some emotions in GoEmotions are not clearly related to Ekman’s, such as approval, love, pride, amusement, etc.ConclusionEkman’s scheme performs best under the evaluation framework. However, it lacks sufficient positive emotion categories for the corpus.

Reaping the Chemical Diversity of Morinagamyces vermicularis Using Feature-Based Molecular Networking.

Moringadepsin (6) and chaetone B (7) were isolated by us in the course of a conventional chemical screening of Morinagamyces vermicularis CBS 303.81, a fungus belonging to the relatively underexplored family Schizotheciaceae of the phylum Ascomycota. Since these metabolites did not account for the antifungal activity observed in a crude extract of this fungus, we utilized an MS/MS-based molecular networking approach to get a thorough insight into the secondary metabolites produced by this strain. Manual annotation of high-resolution fragmentation mass spectra by CANOPUS classified a major molecular family as putatively new thiodiketopiperazines. However, these results were opposite to the results of ChemWalker analysis based solely on MS/MS data, assigning these metabolites as various polyketides. Thus, targeted preparative HPLC isolation focusing on the most abundant features within this major molecular family resulted in the isolation of five secondary metabolites. Their structures were elucidated based on HRMS and NMR data as four new thiodiketopiperazine derivatives, botryosulfuranols D-G (1-4), alongside the known botryosulfuranol A (5). Compounds 1-3 and 5 exhibited moderate to weak antifungal activity against different test strains, accounting for the initial antifungal activity observed for its crude extract. Our study stressed the importance of full NMR-based structure elucidation for metabolomics research.

Automatic detection of problem-gambling signs from online texts using large language models.

Problem gambling is a major public health concern and is associated with profound psychological distress and economic problems. There are numerous gambling communities on the internet where users exchange information about games, gambling tactics, as well as gambling-related problems. Individuals exhibiting higher levels of problem gambling engage more in such communities. Online gambling communities may provide insights into problem-gambling behaviour. Using data scraped from a major German gambling discussion board, we fine-tuned a large language model, specifically a Bidirectional Encoder Representations from Transformers (BERT) model, to predict signs of problem-gambling from forum posts. Training data were generated by manual annotation and by taking into account diagnostic criteria and gambling-related cognitive distortions. Using cross-validation, our models achieved a precision of 0.95 and F1 score of 0.71, demonstrating that satisfactory classification performance can be achieved by generating high-quality training material through manual annotation based on diagnostic criteria. The current study confirms that a BERT-based model can be reliably used on small data sets and to detect signatures of problem gambling in online communication data. Such computational approaches may have potential for the detection of changes in problem-gambling prevalence among online users.