Sensitivity And Specificity Research Articles

MTSA-SC: A multi-task learning approach for individual trip destination prediction with multi-trajectory subsequence alignment and space-aware loss functions.

Individual Trip Destination Prediction aims to accurately forecast an individual's future travel destinations by analyzing their historical trajectory data, holding significant application value in intelligent navigation, personalized recommendations, and urban traffic management. However, challenges such as data sparsity, low quality, and complex spatiotemporal volatility pose substantial difficulties for prediction tasks. Existing studies exhibit notable limitations in insufficient integration of sparsity handling and prediction tasks, constrained modeling capability for local volatility, and inadequate exploration of fine-grained spatial dependencies, struggling to balance global patterns and local features in trajectory data. To address these issues, this paper proposes an individual trip destination prediction method that integrates multi-task learning, a multi-trajectory subsequence alignment attention mechanism, and a spatially consistent constrained cross-entropy loss function. Leveraging a multi-task learning framework(MTSA-SC), our approach collaboratively addresses trajectory recovery and prediction tasks, enhancing prediction accuracy while improving robustness to missing data. The multi-trajectory subsequence alignment attention mechanism incorporates sliding windows and convolutional operations to dynamically capture local volatility and diverse patterns in trajectories. The spatially consistent constrained loss function strengthens spatial feature learning through differential error penalty adjustments. Experimental results on public datasets from Shenzhen and Xiamen demonstrate recall rates of 0.722 and 0.6 under complete and sparse trajectory scenarios, respectively, outperforming state-of-the-art baselines by an average of 15.64%. This research provides robust technical support for intelligent travel recommendations and traffic management.

ST-YOLO: a deep learning based intelligent identification model for salt tolerance of wild rice seedlings

BackgroundIn response to the limited models for salt tolerance detection in wild rice, the subtle leaf features, and the difficulty in capturing salt stress characteristics, resulting in low recognition and detection rates and accuracy, a deep learning-based ST-YOLO wild rice seedling salt tolerance phenotype evaluation and identification model is proposed.MethodIn order to improve accuracy and achieve model lightweighting, a multi branch structure DBB (Diverse Branch Block) is used to replace the convolutional layers in the C2f module, and a reparameterization module C2f DBB is proposed to replace some C2f modules. Diversified feature extraction paths are introduced to enhance the ability of feature extraction; Introducing CAFM (Context Aware Feature Modulation) convolution and attention fusion modules into the backbone network to enhance feature representation capabilities while improving the fusion of features at various scales; Design a more flexible and effective spatial pyramid pooling layer using deformable convolution and spatial information enhancement modules to improve the model’s ability to represent target features and detection accuracy.ResultsThe experimental results show that the improved algorithm improves the average precision by 2.7% compared with the original network; the accuracy rate improves by 3.5%; and the recall rate improves by 4.9%.ConclusionThe experimental results show that the improved model significantly improves in precision compared with the current mainstream model, and the model evaluates the salt tolerance level of wild rice varieties, and screens out a total of 2 varieties that are extremely salt tolerant and 7 varieties that are salt tolerant, which meets the real-time requirements, and has a certain reference value for the practical application.

Effect of incorporating symptom burden with mortality as a composite outcome on accuracy and bias in palliative care identification algorithms in oncology.

12016 Background: Machine learning (ML) algorithms are increasingly used to identify patients for early palliative care (PC) or advance care planning (ACP). Most PC/ACP algorithms are trained using only mortality as an outcome. However, increasing availability of structured patient-reported outcomes (PROs) in electronic health record (EHR) databases can facilitate more comprehensive identification of palliative care need by training algorithms on composite outcomes of mortality and symptom burden. Methods: Our cohort consisted of patients with cancer seen at one of 18 practices in 2019 within a large academic cancer center. We leveraged structured EHR data, consisting of 153 demographic, laboratory, and comorbidity features and 12 symptom scores derived from CTCAE-PRO that were routinely reported at medical oncology encounters (72% response rate). Our Base Model was a random forest model predicting 180-day mortality from the date of an initial medical oncology encounter (index encounter) that was used in practice to prompt earlier ACP conversations. We retrained models using a Composite Label of mortality and/or severe symptoms (≥3 out of 5 in at least 1 symptom) within 180-days of an index encounter. We report performance for Base vs. Retrained models in 1,000 bootstrapped samples predicting the Composite Label using area under the precision-recall curve (AUPRC) and true positive rate (TPR) for All Patients, Black Patients, and White Patients. We hypothesized that Retrained models would improve performance and reduce Black-White disparities. Results: Our cohort consisted of 4908 patients (median age 64.1 years [IQR 17.5], 53.0% female, 59.6% solid tumor malignancies). Retrained Models improved TPR over Base Models for All (0.56 [95% CI 0.52-0.59] vs. 0.18 [95% CI 0.15-0.21]), Black (0.60 [95% CI 0.52-0.67] vs. 0.20 [95% CI 0.14-0.26]), and White (0.55 [95% CI 0.50-0.59] vs. 0.17 [95% CI 0.14-0.20]) patients, with similar AUPRC. TPR improvement was marginally greater for Black vs. White patients (0.02 [95% CI -0.01-0.05]). Conclusions: In this cohort study, retraining a PC identification algorithm on a Composite outcome of symptom burden + mortality significantly enhanced identification of PC need, with disproportionate improvements for Black patients. Incorporating PROs into PC identification model outcome labels should be strongly encouraged. Comparison of base vs. retrained model performance. All Patients a Black Patients a White Patients a Other b Model Base Retrained Base Retrained Base Retrained Difference in Black vs. White Improvement AUPRC 0.71 (0.68, 0.74) 0.71 (0.68, 0.74) 0.80 (0.73, 0.85) 0.79 (0.72, 0.85) 0.68 (0.65, 0.72) 0.68 (0.65, 0.72) 0 (-0.01, 0.08) TPR 0.18 (0.15, 0.21) 0.56 (0.52, 0.59) 0.20 (0.14, 0.26) 0.60 (0.52, 0.67) 0.17 (0.14, 0.20) 0.55 (0.50, 0.59) 0.02 (-0.01, 0.05) a Mean (95% CI) b Mean difference (95% CI).

DKCN-Net: Deep kronecker convolutional neural network-based lung disease detection with federated learning.

DKCN-Net: Deep kronecker convolutional neural network-based lung disease detection with federated learning.

Ensemble and deep learning via median method for learning disability classification

The study explores the classification of students with and without learning disabilities (LD) through machine learning techniques, utilizing a real dataset and implementing bootstrapping for data augmentation. Noteworthy findings reveal the Adam optimizer's superior performance among various optimizers, achieving a true positive rate (TPR) of 0.97 and a false positive rate (FPR) of 0.02, with high precision, recall, and f1-score values. Additionally, ensemble learning, employing the median method, combines models like Random-ForestClassifier and KerasClassifier, and BaggingClassifier with KerasClassifier, resulting in improved performance. However, the Median-Combined model, integrating AdaBoostClassifier and KerasClassifier, stands out with an accuracy of 99.6%, along with elevated precision, recall, and f1-score values. The comprehensive classification report showcases an overall FPR of 0.0 and TPR of 0.999, highlighting the enhanced performance of the combined model. The significance of this study lies in underscoring the power of fusion between ensemble learning and deep learning techniques, leveraging the median method. This combined model exhibits superior performance, excelling in accuracy, precision, recall, and overall classification effectiveness. The innovative approach of combining both ensemble and deep learning methods through the median method not only advances the understanding of learning disability classification but also emphasizes the practical importance of integrating diverse methodologies for enhanced model performance.

Comparative analysis of deep learning model artificial intelligence and radiologists in breast tumor classification: A study in Uzbekistan.

e13650 Background: To evaluate and compare the diagnostic performance of a deep learning-based artificial intelligence (AI) system versus three radiologists in the detection of breast cancer using digital mammography, specifically within the context of Uzbekistan, and to determine if AI can serve as a reliable tool in this setting. Methods: This retrospective study utilized a dataset of mammograms, sourced from Uzbekistan, which were independently assessed by three radiologists and an AI system. The AI model, based on deep neural networks, was designed for automated breast cancer detection. The radiologists’ interpretations and the AI predictions were compared against a reference standard of biopsy results. The primary outcome measures included the area under the receiver operating characteristic curve (AUC), accuracy, and specificity for both the AI system and radiologists. The data underwent rigorous statistical analysis to establish the significance of the observed differences. The model was trained using data from multiple institutions in multiple countries. Results: The AI system demonstrated a significantly higher area under the curve (AUC of 0.89) compared to the average of three radiologists (AUC of 0.82). The AI also showed higher specificity (e.g., 93.0% versus 77.6%), and the recall rate for AI was three times lower than that of radiologists. The AI was more sensitive in detecting cancers with mass, distortion, or asymmetry and better at detecting T1 or node-negative cancers. This result underscores AI's potential to reduce false positives, but also demonstrates that it can detect cancers missed by radiologists. The AI system's performance aligns with other studies showing AI sensitivity to be non-inferior to, or surpassing, radiologists. AI systems can detect more cancers with mass or distortion than radiologists. The statistical analysis showed that the AI system achieved robust accuracy and demonstrated potential as a reliable tool to enhance breast cancer screening outcomes. A study also showed that AI can reduce the number of reads in a screening program by 41.4%. Conclusions: In this study the AI system outperformed the group of radiologists in terms of AUC, specificity, recall rates, and positive predictive value. These findings suggest that deep learning-based AI can significantly improve the detection of breast cancer in mammography and may serve as a valuable tool in the Uzbekistan healthcare setting. Additional studies that include larger, more heterogenous datasets are warranted and it is important to continue researching AI integration, including risk management and real-world follow up of performance. Future studies should examine the impact of AI on screening performance when used by radiologists and assess the value of different models for various conditions.

48-month comperative evaluation of a novel glass ionomer cement and a resin composite in restoring non-carious cervical lesions of patients with systemic diseases: A randomized clinical trial.

48-month comperative evaluation of a novel glass ionomer cement and a resin composite in restoring non-carious cervical lesions of patients with systemic diseases: A randomized clinical trial.

Polymer weathering under simulated solar radiation and comparison to stormwater and estuarine microplastics.

Polymer weathering under simulated solar radiation and comparison to stormwater and estuarine microplastics.

Mammography results in male BRCA carriers.

10618 Background: Patients assigned male at birth (AMAB) with pathogenic germline variants (PGVs) in BRCA1 or BRCA2 ( BRCA+ ) have a 1-10% lifetime risk of developing breast cancer (BC). NCCN guidelines currently recommend that male BRCA1/2 carriers consider annual mammograms. However, there is minimal data on the mammography among male BRCA1/2 carriers. Methods: We identified a cohort of 489 BRCA + male patients with male gender identity from the electronic health record (EHR) at Penn Medicine who had no prior BC diagnosis. Charts were reviewed to determine indication for and results of mammography episodes between 2008-2024. A mammography episode was defined as all breast imaging studies obtained for a specific reason, i.e. asymptomatic screening or for symptoms within six months. An independent cohort of 1808 BRCA negative AMAB patients with male gender identity and no prior BC diagnosis from the Penn Medicine EHR was analyzed for the true positive rate of BI-RADS 4/5 findings. Results: Of 489 BRCA + individuals, 85 (17%) patients completed at least one mammography episode and 46 (9%) had at least one subsequent mammography episode during the study period. Of 85 BRCA + patients, 71% had BRCA2 PGVs. Of 404 patients who did not complete a mammogram, 270 were at least 50 years old at the time of data abstraction. Of these 270 patients, 50% had no discussion of mammograms in their charts, 8% of patients had a physician ordered mammogram that was not completed by the patient, and 42% had a shared decision-making discussion between the physician and patient indicating a decision against mammography. The first observed and subsequent mammography episodes were ordered for asymptomatic screening in 65% and 83% of 85 BRCA + individuals, respectively. In BRCA neg individuals, 92% of mammography episodes were for symptoms. Nine (11%) and one (2%) BRCA+ individuals were diagnosed with BC after the first observed or subsequent mammography episode, respectively. No breast cancers were identified on mammography episodes among asymptomatic patients. Combining all mammography data, the true positive rate of BI-RADS 4 mammograms was significantly higher in BRCA + vs BRCA neg individuals (71% vs 11%, p=0.0007); whereas the true positive rate of BI-RADS 5 mammograms was similar in BRCA + vs BRCA -neg individuals (100% vs 82%, p=0.54). Hormone receptor status and clinical stage of identified BC were similar between BRCA + and BRCA neg individuals. Conclusions: The majority of male BRCA1/2 carriers in our cohort did not complete mammography. All BC diagnosed in BRCA+ individuals were identified on mammography episodes obtained for symptoms. The true positive rate of a BI-RADS 4 mammogram was significantly higher in BRCA + compared to BRCA neg individuals. Additional data is needed regarding whether mammography identifies asymptomatic BC in male BRCA1/2 carriers and whether mammograms improve clinical outcomes.

Vehicle trajectory fractal theory for macro-level highway crash rate analysis.

Vehicle trajectory fractal theory for macro-level highway crash rate analysis.

Optimizing elderly care: A data-driven AI model for predicting polypharmacy risk in the elderly using SHARE data.

Optimizing elderly care: A data-driven AI model for predicting polypharmacy risk in the elderly using SHARE data.

A semantic segmentation network for red tide detection based on enhanced spectral information using HY-1C/D CZI satellite data.

A semantic segmentation network for red tide detection based on enhanced spectral information using HY-1C/D CZI satellite data.

Evaluating the Performance of a Regularized Differential Item Functioning Method for Testlet-Based Polytomous Items

This study investigated the effect of testlets on regularization-based differential item functioning (DIF) detection in polytomous items, focusing on the generalized partial credit model with lasso penalization (GPCMlasso) DIF method. Five factors were manipulated: sample size, magnitude of testlet effect, magnitude of DIF, number of DIF items, and type of DIF-inducing covariates. Model performance was evaluated using false-positive rate (FPR) and true-positive rate (TPR). Results showed that the simulation had effective control of FPR across conditions, while the TPR was differentially influenced by the manipulated factors. Generally, the small testlet effect did not noticeably affect the GPCMlasso model’s performance regarding FPR and TPR. The findings provide evidence of the effectiveness of the GPCMlasso method for DIF detection in polytomous items when testlets were present. The implications for future research and limitations were also discussed.

Improved People Counting System Using Deep Learning

Abstract: With the rapid rise in population, public areas such as malls, supermarkets, and transport hubs are becoming increasingly crowded. Businesses depending on customer footfallpatterns requireaccurate datatooptimize operations.Toaddress this, we developed a people counting and tracking system that detects, tracks, and identifies individuals in real-time.The system uses Faster R-CNN for robust people detection, offering high accuracy even in dense environments. To ensure consistent monitoring, DeepSORT assigns unique IDs to each individual andtracks them across frames. Additionally, DeepFace is integrated for face recognition, enabling the system tomatch detected faces with previously registered identities.A face registrationmodule (register_faces.py)allowswebcam- based registration, making it user-friendly. The evaluation module (evaluate.py) computes key performance metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The model was tested on a dataset comprising 2416 positive and 1218 negative image samples. It achieved a True Positive Rate (TPR) of 95.03%, a False Positive Rate (FPR) of 0.08%, and an overall accuracy of 97.08%. While the model performs well, challenges such as overlapping subjects, varying clothing, and lighting conditions may occasionally affect results. This system provides a reliable and scalable solution for people counting, face tracking, and identity verification

Classification of Multi-Region Bone Fractures from X-ray Images Using Transfer Learning with ResNet18

Fracture detection in radiographic images is a critical task in orthopaedic diagnostics, often requiring timely and accurate interpretation by medical professionals. However, manual evaluation of X-rays is time-consuming and prone to subjective bias. This study proposes an automated deep learning approach for binary classification of bone fractures using a pre-trained ResNet18 architecture. The model was trained and validated on a multi-region X-ray dataset consisting of 10,580 images categorized into fractured and non-fractured classes. To improve generalization, data augmentation techniques such as rotation and horizontal flipping were applied during pre-processing. The final model achieved a validation accuracy of 97.59%, with high true positive and true negative rates as confirmed by the confusion matrix analysis. The results demonstrate the effectiveness of transfer learning in handling radiographic image classification tasks while maintaining computational efficiency. This research contributes to the development of reliable and scalable computer-aided diagnostic tools that can support clinical decision-making, especially in environments with limited resources.

An Improved Lithium-Ion Battery Fire and Smoke Detection Method Based on the YOLOv8 Algorithm

This paper introduces a novel algorithm—YOLOv8 (You Only Look Once version 8) + FRMHead (a multi-branch feature refinement head) + Slimneck (a lightweight bottleneck module), abbreviated as YFSNet—for lithium-ion battery fire and smoke detection in complex backgrounds. By integrating advanced modules for richer feature extraction and streamlined architecture, YFSNet significantly enhances detection precision and real-time performance. A dataset of 2300 high-quality images was constructed for training and validation, and experimental results demonstrate that YFSNet boosts detection precision from 95.6% in the traditional YOLOv8n model to 99.6%, while the inference speed shows a marked improvement with FPS increasing from 49.75 to 116.28. Although the recall rate experienced a slight drop from 97.7% to 93.1%, the overall performance in terms of F1-score and detection accuracy remains robust, underscoring the method’s practical value for reliable and efficient battery fire detection in fire safety systems.

Improving Aerobics Posture Evaluation by Transfer Learning: Humanized Computational Application of BERT-PTA Domain Adaptive Methods

Due to the influence of datasets, traditional pose evaluation methods have insufficient generalization ability, high computational resource requirements, and low efficiency. To address this issue, this article applied transfer learning into the field of aerobics posture evaluation and achieved automation and objectivity of posture evaluation through BERT-PTA (Bidirectional Encoder Representations from Transformers-Prototype-based Transfer Assistants) domain adaptive methods. BERT and PTA methods were chosen because BERT’s bidirectional language understanding and transfer learning capabilities can effectively adapt to the language instructions in the field of aerobics, and PTA’s posture tracking and humanized computing features provide an accurate and user-friendly solution specifically for the assessment of aerobics poses. First, a BERT-PTA model was established based on the collection of aerobics posture data. Second, the BERT-PTA model was used to extract features from the preprocessed posture data. Next, a convolutional neural network was used to construct a key point localization model for aerobics poses, and transfer learning was used to train and fine-tune the model. Finally, experimental verification was conducted on using transfer learning to improve aerobics posture evaluation. The results showed that the precision of using transfer learning to improve aerobics posture evaluation was 4.88% and 8.86% higher than that of the other two methods, respectively. The recall rate of using transfer learning to improve aerobics posture evaluation was 3.45% and 7.14% higher than that of the other two methods, respectively. The evaluation efficiency of using transfer learning to improve aerobics posture evaluation was 6.52–7.69% higher than that of the other two methods, respectively. In short, using transfer learning to improve aerobics can provide more scientific guidance for aerobics sports.

Design of a Configurable Acoustic Sensor Network for Privacy‐Compliant Urban Soundscape Recordings

ABSTRACTEnvironmental acoustics, particularly urban soundscape monitoring, has gained increasing consideration since the United Nations established the Sustainable Development Goals in 2015, and to an even greater extent with the rise of privacy concerns following the introduction of global regulations such as GDPR. As a result, privacy‐compliant devices have become essential for soundscape monitoring in urban environments. In this paper, we present the design and implementation of a portable AI‐driven, privacy‐compliant urban sound recording device that locally captures and processes acoustic data on the edge. In more detail, this device operates as a sensor that captures soundscapes and processes them through a pipeline, which employs a pre‐trained open‐source AI model to anonymize human voices, ensuring privacy without compromising the integrity of the acoustic environment. The anonymization process alters human speech in a way that protects identity while maintaining environmental audio quality. The device can function as a standalone sensor or as part of a synchronized network of distributed sensors. Privacy‐focused evaluation of the device's recordings indicates that, while the anonymization process impacts speech intelligibility, it preserves the overall soundscape with a recall rate of 96%. The system was deployed in a real‐world setting with four temporally synchronized sensors. While network synchronization was achieved, a 1 to 2‐s deviation was occasionally observed in the first duty cycle interval, reflecting timing variability inherent to Cron‐based script triggering. This limitation has been identified for future refinement. This work demonstrates the feasibility of deploying privacy‐compliant, edge‐based soundscape sensors in urban environments, contributing to privacy preservation, and enhanced public safety.

Automated Detection of Micro-Scale Porosity Defects in Reflective Metal Parts via Deep Learning and Polarization Imaging

Aiming at the key technology of defect detection in precision additive manufacturing of highly reflective metal materials, this study proposes an enhanced SCK-YOLOV5 framework, which combines polarization imaging and deep learning methods to significantly improve the intelligent identification ability of small metal micro and nano defects. This framework introduces the SNWD (Selective Network with attention for Defect and Weathering Degradation) Loss function, which combines the SIOU Angle Loss with the NWD distribution sensing characteristics. It is specially designed for automatic positioning and identification of micrometer hole defects. At the same time, we employ global space construction with a dual-attention mechanism and multi-scale feature refining technique with selection kernel convolution to extract multi-scale defect information from highly reflective surfaces stably. Combined with the polarization imaging preprocessing and the comparison of enhancement defects under high reflectivity, the experimental results show that the proposed method significantly improves the precision, recall rate, and mAP50 index compared with the YOLOv5 baseline (increased by 0.5%, 1.2%, and 1.8%, respectively). It is the first time that this improvement has been achieved among the existing methods based on the YOLO framework. It creates a new paradigm for intelligent defect detection in additive manufacturing of high-precision metal materials and provides more reliable technical support for quality control in industrial manufacturing.

Comparison of Directional and Diffused Lighting for Pixel-Level Segmentation of Concrete Cracks

Visual inspections of concrete infrastructure in low-light environments require external lighting to ensure adequate visibility. Directional lighting sources, where an image scene is illuminated with an angled lighting source from one direction, can enhance the visibility of surface defects in an image. This paper compares directional and diffused scene illumination images for pixel-level concrete crack segmentation. A novel directional lighting image segmentation algorithm is proposed, which applies crack segmentation image processing techniques to each directionally lit image before combining all images into a single output, highlighting the extremities of the defect. This method was benchmarked against two diffused lighting crack detection techniques across a dataset with crack widths typically ranging from 0.07 mm to 0.4 mm. When tested on cracked and uncracked data, the directional lighting method significantly outperformed other benchmarked diffused lighting methods, attaining a 10% higher true-positive rate (TPR), 12% higher intersection over union (IoU), and 10% higher F1 score with minimal impact on precision. Further testing on only cracked data revealed that directional lighting was superior across all crack widths in the dataset. This research shows that directional lighting can enhance pixel-level crack segmentation in infrastructure requiring external illumination, such as low-light indoor spaces (e.g., tunnels and containment structures) or night-time outdoor inspections (e.g., pavement and bridges).