Search Results (148)

Background and Objectives: Median nerve entrapment following elbow dislocation is a very rare but serious condition in children. We describe four cases of median nerve entrapment, highlight diagnostic challenges, evaluate the impact of delayed treatment, and propose a treatment algorithm. Materials and Methods: A retrospective review of a consecutive case series of four children with median nerve entrapment following elbow dislocation was conducted. Clinical records, imaging, operative findings, and outcomes were reviewed. Results: In all patients, the causative injury was posterolateral elbow dislocation with an associated medial epicondyle fracture. Three patients had delayed diagnosis of their median nerve entrapment (at 10–31 months post-injury) and required nerve resection with sural nerve grafting; none achieved full neurological recovery. One patient underwent early exploration and decompression (at 4 days post-injury) and made a full recovery. Misattribution of neurological symptoms to transient neurapraxia, under-recognition of radiographic signs, and inadequate early advanced imaging contributed to delayed diagnosis and management. Magnetic resonance imaging was the most useful imaging modality for diagnosing median nerve entrapment, though initial reporting missed entrapment in one case. Conclusions: Prompt recognition and early surgical decompression are critical for optimal recovery in paediatric median nerve entrapment after elbow dislocation. Delayed treatment is associated with poor outcomes. We propose a treatment algorithm to guide management and improve outcomes. Full article

Background: Tissue and plasma comprehensive genomic profiling (CGP) sample different compartments and may yield nonoverlapping clinically relevant findings. We assessed the added patient-level yield of paired testing in a pan-cancer cohort from India. Methods: This retrospective, single-institution study included patients with stage IV solid tumors who underwent paired tissue and plasma CGP between November 2022 and October 2024. The primary endpoint was detection in a patient of an OncoKB Level 1 or Level 2 or R1 alteration, high tumor mutational burden, or microsatellite instability. Molecularly informed therapy and laboratory turnaround time were assessed descriptively. Results: Of 158 patients, 123 matched evaluable pairs were analyzed. Either modality detected a clinically relevant finding in 46 of 123 patients (37.4%; 95% confidence interval, 29.4–46.2%): tissue CGP in 33 (26.8%), plasma CGP in 35 (28.5%), both in 22 (17.9%), tissue only in 11 (8.9%) and plasma only in 13 (10.6%). This complementarity remained bidirectional across exploratory analyses, including a subgroup that excluded non-small cell lung cancer (33.0%), all 139 technically successful pairs (33.8%), all 158 paired sample sets (32.3%), and even when the endpoint was restricted to OncoKB alterations (22.0%). Among 91 patients with therapy-alignment data, 25 (27.5%) received molecularly informed therapy. Median laboratory turnaround time was 7 days for plasma and 13 days for tissue CGP. Conclusions: Tissue and plasma CGP were complementary rather than interchangeable in this selected cohort: either modality alone would have missed clinically relevant findings. Whether this improves outcomes is untested; prospective studies linking testing strategy to treatment delivery and survival are needed. Full article

Osteoarthritis is a common cause of equine lameness and can involve periarticular osteophytes and subchondral bone (SB) changes. This exploratory ex vivo study evaluated the diagnostic performance of digital tomosynthesis (DT) for detecting these findings in the distal interphalangeal and metacarpophalangeal joints, using computed tomography (CT) classifications as the reference. Sixteen distal forelimbs from nine adult horses underwent imaging with both modalities under simulated weight-bearing conditions. Three independent observers graded osteophytes and SB changes, and the scores were dichotomised as absent or present for analysis. The primary analysis used a majority-based CT reference, defined by agreement of at least two observers. Observer-specific comparisons and within-observer DT–CT agreement were evaluated in a secondary analysis. In the primary analysis, sensitivity ranged from 23.1% (95% CI 9.5–39.3%) to 34.6% (95% CI 13.8–56.0%), specificity from 94.1% to 96.1%, Youden’s index from 19.2% to 30.7%, and Cohen’s κ from 0.249 to 0.377. Sensitivity was numerically higher for osteophytes than for SB changes for all three observers, without formal comparison. DT missed many CT-positive assessments but produced few false-positive classifications. Studies in live horses comparing DT with conventional radiography are needed to determine its additional diagnostic value. Full article

Reliable and robust multimodal crisis classification is challenged by degraded, corrupted, or missing image–text evidence, because reliability-blind fusion can propagate an unreliable modality and amplify local evidence failure into joint prediction error. To address this problem, we propose Clean-Preserving Residual Reliability Adaptation (CP-RRA), a preserve–diagnose–intervene framework for robust multimodal crisis classification. First, CP-RRA preserves the intact evidence decision pathway as a stable anchor; second, it estimates modality reliability and intervention strength to construct a reliability-aware target representation; finally, it applies bounded residual adaptation between the target and the preserved anchor, with the intervention strength controlling the residual update. Evaluation on a five-class CrisisMMD v2.0 subset covers image and text corruptions, training-unseen perturbations, and complete modality loss. Under the original three-seed protocol, the pre-specified locked CP-RRA configuration (λ_rel = 1.0) achieves 76.17% worst Weighted-F1; under complete image or text loss, it recovers more than 44 percentage points relative to the preserved pathway. Post-lock Train/Dev ablation shows that removing explicit reliability supervision (λ_rel = 0) increases worst Weighted-F1 to 79.70%, while substantially weakening reliability diagnostics, revealing a prediction–diagnostic trade-off. Reliability diagnostics further show that the learned scores identify the usable evidence channel. These results indicate that CP-RRA improves continuity and resilience under asymmetric evidence failure while providing auditable signals for crisis information screening, prioritization, and analyst verification. Full article

Multi-modal 3D object detection is critical for autonomous driving perception. While Bird’s Eye View (BEV) fusion methods effectively integrate LiDAR and camera features, they primarily focus on single-frame fusion and neglect temporal context. We propose CamT-BEV, a camera-temporal-enhanced BEV fusion framework for improved multi-modal 3D object detection. Our key insight is that temporal modeling is particularly critical for the camera branch to resolve monocular depth ambiguity and object occlusion, while single-frame LiDAR representation already provides accurate instantaneous geometry. We thus propose a camera-centric temporal enhancement module via ego-motion warping and ConvLSTM temporal encoding. Extensive experiments on the nuScenes dataset demonstrate that CamT-BEV achieves competitive perception performance, attaining 0.6971 NDS and 0.6683 mAP, with notable relative AP gains on challenging categories such as bicycles (+27.3%) and motorcycles (+7.66%) evaluated under category-level mAP (averaged across 0.5 m to 4.0 m distance thresholds). Furthermore, evaluations under fog and miss-beam conditions in nuScenes-C confirm its improved robustness against specific visual and sensor degradations. Crucially, these gains are achieved with low additional computational and memory overhead, demonstrating that targeted camera-temporal fusion is a practical solution for 3D perception. Full article

Early-childhood digital screening, clinician-supervised assessment, and monitoring use smartphone images, home video, interaction audio, caregiver questionnaires, and non-contact sensors. Yet “multimodal” is sometimes applied when distinct patient-level inputs are not jointly modeled. This structured narrative review examines selected decentralized domains—neonatal jaundice, autism-related screening or assessment, and cry or sleep monitoring—in children from birth through six years. A structured targeted update through 29 July 2026 used PubMed, IEEE Xplore, version-of-record pages, citation chaining, and official guidance. Multimodality required the joint use of two or more conceptually distinct patient-level modalities for model development, inference, or decision generation. Of 22 retained empirical reports, 9 used multimodal fusion, representing four system lineages. Validation was task-specific; prospective or external evaluation was more developed in neonatal jaundice and autism-related applications, while remote-home, postmarket, or implementation evidence was concentrated in selected autism-related lineages. Limitations included matched unimodal comparisons, calibration, determinate result coverage, missing-input handling, subgroup transportability, independent replication, referral outcomes, and deployment evaluation. A non-aggregated decision-oriented matrix organizes these requirements across technical, clinical, operational, equity, caregiver, privacy, and safety dimensions. Future studies should test whether multimodal fusion improves clinically relevant decisions over matched unimodal baselines, and whether gains persist across independent settings and care pathways. Full article

Short-term cross-border prediction is of significant importance for global supply-chain risk management, international trade decision-making, and stability. However, existing approaches mainly rely on individual price sequences or limited structured variables, making it difficult to effectively perceive complex external factors, including cross-border trade flows, logistics state variations, exchange-rate fluctuations, and policy-driven shocks. To address these challenges, an artificial intelligence-driven sensing-oriented reliability- and consistency-aware multimodal framework for short-term cross-border prediction is proposed. Market conditions, exchange rates, cross-border trade activities, logistics operations, and news–policy events are jointly modeled as multisource intelligent sensing signals. In the proposed framework, the reliability and cross-modal consistency perception module is first developed to dynamically evaluate the credibility of different data sources and suppress the interference caused by missing, delayed, and conflicting information. Subsequently, the asynchronous cross-border temporal interaction module is introduced to capture the time-dependent propagation relationships among trade, logistics, exchange rates, and states. Furthermore, the dynamic trade and policy event perception module is constructed to identify short-term disturbances induced by tariff adjustments, trade restrictions, port disruptions, and major international events, thereby enabling intelligent prediction under complex cross-border environments. Based on a multisource cross-border sensing dataset constructed from January 2022 to December 2025, the performance of the proposed framework is systematically evaluated through three tasks, including short-term direction prediction, volatility forecasting, and risk level prediction. Experimental results demonstrate that the proposed method achieves an accuracy of 0.842, precision of 0.836, recall of 0.829, macro-F1 of 0.832, and AUC of 0.913 in the short-term direction prediction task, significantly outperforming ARIMA, XGBoost, LSTM, TCN, Transformer, PatchTST, and existing multimodal fusion models. Ablation studies further verify the critical contributions of reliability modeling, consistency constraints, asynchronous temporal interaction, and dynamic event perception modules to improving prediction performance. Full article

Automating the identification of International Classification of Diseases (ICD) codes from electronic health records (EHRs) presents a critical challenge, particularly for rare diseases where existing computational methods severely underperform due to extreme long-tail label distributions. To address this, we propose a multi-modal deep learning framework known as MMAC-Net, designed to enhance the retrospective assignment of ICD-9 codes to admissions involving rare pathologies. The model integrates unstructured clinical narratives with structured auxiliary data, specifically pharmacological prescriptions and microbiology events, using a convolutional attention-based architecture. Through a late fusion mechanism, it synthesizes attention-weighted textual representations with dense embeddings of the structured data types. Validation on the MIMIC-III dataset shows consistent improvements over a matched text-only baseline evaluated under an identical protocol. On the full dataset of 8930 ICD codes, the framework achieved a Micro-AUC of 0.997 and Precision@8 of 0.875. On the subset of admissions carrying at least 1 of 568 rare codes, adding the two structured modalities to the text encoder raises Macro-F1 from 0.011 to 0.084 and Micro-F1 from 0.368 to 0.513 relative to the text-only baseline, corresponding to relative increases of 6.69 and 0.39, respectively, while Precision@8 rises from 0.092 to 0.159 and Micro-AUC from 0.966 to 0.985. While extreme class imbalance remains a formidable obstacle, these findings underscore that incorporating structured clinical context partially mitigates the limitations of purely natural language processing approaches. Practically, the framework is intended as a decision-support component that presents a ranked shortlist of candidate codes to a human coder or clinician; by recovering rare codes that text-only systems miss, it targets the under-coding of low-prevalence conditions that degrades registry completeness and downstream epidemiological estimates. Full article

Background: Tumor heterogeneity arises from complex interactions among diverse biological factors, posing a major challenge for the development of robust multi-omics data integration methods. While the existing Tumor Multi-Omics pre-trained Network (TMO-Net) enables the fusion of multi-omics features into unified representations, its practical utility is constrained by issues such as missing modalities, incomplete within-omics data, and high-dimensional noise. To overcome these limitations, we propose TMO-Net+, an enhanced architecture specifically designed to improve the robustness and reliability of multi-omics modeling. Methods: TMO-Net+ introduces several coordinated architectural enhancements. First, a feature attention encoder is applied to each omics data type to reduce the influence of modality-dependent input variation. Second, we combine a gated Mixture-of-Experts (MoE) module with a Product-of Experts (PoE) mechanism to capture sample-specific contributions and enable robust inference even when partial omics data are available. Additionally, a supervised deep classification head with a tailored loss function is incorporated to enhance the separability of learned embeddings in the latent space. Results: Extensive experiments on pan-cancer datasets demonstrate that TMO-Net+ consistently outperforms the original TMO-Net, as measured by LogME scores. Furthermore, in various downstream tasks (e.g., pan-cancer classification, primary/metastatic site prediction, and prognostic modeling), TMO-Net+ achieves superior performance under partial-omics settings, which proves that it enhances the robustness and cross-cancer transferability of the multi-omics representations. Conclusions: The proposed TMO-Net+ improves the robustness and cross-cancer transferability of multi-omics representations within the evaluated TCGA cohorts. Biological interpretability analyses further show that TMO-Net+ prioritizes established cancer-driver genes, preserves cancer-dependent molecular-state information, and adaptively redistributes relative modality contributions across molecular states. By addressing modality-level missingness and modality-dependent input variation, it offers a reliable framework for integrative tumor analysis within the evaluated TCGA cohorts. Full article

Automated concrete-surface classification increasingly employs multimodal sensing, although a wide range of systems assume that all sensors remain continuously available, reliable, and suitable for fusion. This study presents Reliability-Aware Selective Multimodal Fusion (RASMF), an image-first framework in which supplementary acoustic evidence is requested conditionally, assessed for signal-quality anomalies, fused when appropriate, and rejected when necessary. RASMF was evaluated on 4094 class-associated image-acoustic observations using 5 duplicate-aware outer folds and three random seeds under clean, simulated-degradation, and missing-modality conditions. At the nominal 10% acquisition budget, realised acoustic acquisition was 25.67%. Relative to the predefined missing-aware image-first baseline, mean error decreased from 1.9777% to 1.4499%, an absolute reduction of 0.5279 percentage points and a relative reduction of 26.69%, with a two-way bootstrap 95% confidence interval of −1.6220 to −0.0661 percentage points. The mean Brier score decreased from 0.03033 to 0.01688. RASMF produced lower mean error in 12 of 16 simulated-degradation and modality-loss conditions, with 6 condition-specific confidence intervals excluding zero in its favour. Increasing acoustic acquisition progressively reduced mean error and calibration error but increased processing demand; estimated pipeline time was 24.77 ms at the nominal 10% setting compared with 20.60 ms at the nominal 0% setting. A simpler logit stacker achieved a slightly lower mean error of 1.4295%, and the difference from RASMF was not statistically supported. The results therefore support RASMF as an explicit reliability-aware selective decision architecture relative to the designated baseline, without establishing universal predictive superiority over simpler fusion strategies. Full article

Reliable monitoring of reclaimed cropland is hindered when one optical acquisition is unavailable or degraded. We formulate missing-modality bi-temporal optical–SAR change detection and propose the Modality-Availability-Aware Robust Change Network (MARC-Net), a new architecture that combines explicit availability conditioning, condition-aware temporal proxy stabilization, a shared residual input adapter, multi-level signed temporal interaction, dilated context refinement, and hierarchical change decoding. A two-phase condition-balanced learning strategy jointly develops mixed-missing representations and optimizes Full, Missing-O, and Missing-S behavior without reconstructing the unavailable image. On four reclaimed-cropland scenes, the final model obtains IoUs of 0.7826, 0.7039, and 0.7780, respectively, with a three-condition mean of 0.7548. It exceeds the strongest evaluated external baseline mean (0.7381) while using one checkpoint and a fixed argmax decision rule. LOSO and controlled optical-degradation experiments further characterize robustness under geographic shift and progressive observation-quality degradation. These results demonstrate that availability-aware temporal stabilization and condition-balanced optimization provide an effective operating point for incomplete-input reclaimed-cropland monitoring while preserving complete-input performance. Full article

Computational drug repurposing increasingly integrates chemical, biological, omics, network, text, and clinical data through deep learning. This structured narrative review examines how such modalities are encoded, aligned, and fused. We organize representative studies into four mechanism-centered families: heterogeneous-graph neural networks, multimodal knowledge-graph embeddings, pretrained language/sequence model-based cross-modal alignment, and multi-view or reconstruction-based fusion. Direct drug–disease association and repurposing studies form the core evidence; drug–target interaction, drug–drug interaction, target-identification, molecular-pretraining, and drug–microbe studies are treated as adjacent methodological evidence. We compare architectures, evaluation settings, failure modes, and evidence levels across oncology, neurology, infectious, and rare diseases. Practical guidance covers leakage-aware random, cold-start, temporal, and cluster-based evaluation; an actionable reproducibility checklist; and a scenario-based model-selection framework. We distinguish computational prioritization, docking, preclinical, retrospective clinical, and prospective evidence, and examine data sparsity, uncertain negatives, missing or noisy modalities, interpretability, and translational limitations. Future priorities include temporal and causal evaluation, external and multi-center validation, federated learning, and emerging therapeutic modalities. Multimodal fusion can improve complementary representation, but its value depends on task definition, data quality, evaluation design, and independent validation. Full article

Underwater object detection is a core technology for environmental perception and autonomous operation of unmanned underwater vehicles (UUVs). However, optical and acoustic sensing alone suffer from physical limitations, leading to missed and false detections in turbid, low-light, or long-range conditions. To overcome these limitations, this paper develops an acoustic–optical multimodal fusion detection module (AOMFDM) tailored for UUV deployment. The module employs dual YOLOv5 models for separate processing of sonar and optical images. An interference source quantification estimation network is introduced to extract environmental degradation features, including noise, blur, illumination, contrast, and color cast. A heterogeneous feature map matching network and a deep sparse autoencoder are further designed to achieve cross-modal alignment and fusion of acoustic and optical features. Additionally, attention mechanisms, anchor-based box annotation, and weighted boxes fusion (WBF) are incorporated to enhance detection robustness. For model training and evaluation, we construct the Underwater Sonar Detection (USD) and Underwater Optical Detection (UOD) datasets, covering diverse water qualities, illumination levels, target materials, and interference scenarios. Experimental results demonstrate that, by exploiting the complementarity of acoustic and optical modalities together with adaptive alignment strategies, the proposed module significantly boosts both detection reliability and generalization capability for UUVs in challenging underwater environments. Full article

Water and fertilizer management in facility and precision agriculture increasingly depends on images, soil sensors, weather records, management logs, and economic data rather than on empirical operation alone. A cross-modal decision framework is developed to align crop visual traits with soil water–salt dynamics, meteorological variation, and management operations. Image preprocessing, abnormal sensor screening, missing value interpolation, temporal resampling, multi-window alignment, and temporal augmentation are used to form comparable inputs before multimodal fusion. Using plot-blocked evaluation that keeps complete plot trajectories together, AgriWFD-RLNet obtains 93.46% status-recognition accuracy and an $R^2$ of 0.923 for yield–resource–economic response prediction. In offline policy evaluation against the logged conventional management reference, it estimates 16.92% water saving, 14.37% fertilizer saving, and 15.84% Net Return improvement while maintaining 94.68% yield stability and 96.72% safety compliance. These decision outcomes are model-based estimates from the 2024 Wuyuan dataset rather than outcomes of a prospective field deployment. Full article

Background: Colonoscopy is the gold-standard screening modality for colorectal cancer (CRC) prevention, enabling detection and endoscopic resection of premalignant polyps and reducing CRC incidence and mortality by up to 77% and 53%, respectively. However, colonoscopy effectiveness is substantially dependent on endoscopist expertise, with significant inter-operator variability in adenoma detection rates (ADR) and, consequently, a risk of missed lesions, particularly diminutive and morphologically subtle adenomas. Recent advances in artificial intelligence (AI), specifically computer-aided detection (CADe) and computer-aided diagnosis (CADx) systems utilizing deep learning convolutional neural networks, have emerged as promising technologies to standardize lesion detection accuracy and reduce adenoma miss rates. Methods: A focused narrative literature review was conducted examining randomized controlled trials, meta-analyses, and implementation studies evaluating AI-assisted colonoscopy systems across diverse clinical populations and healthcare settings. Results: Evidence demonstrates that CADe systems consistently improve ADR, particularly for diminutive polyps and morphologically challenging lesions, though superiority over expert endoscopists remains inconsistent. CADx systems reliably meet ASGE-PIVI performance thresholds for diminutive polyp characterization, supporting implementation of resect-and-discard and diagnose-and-leave strategies. However, substantial heterogeneity exists regarding real-world effectiveness, cost-effectiveness, and optimal implementation frameworks across diverse settings. Conclusions: While AI-assisted colonoscopy demonstrates clinical promise in improving lesion detection and enabling optical diagnosis, realizing durable population-level benefit requires the establishment of standardized validation methodologies, large-scale pragmatic trials with patient-centered outcomes, robust regulatory frameworks, and equitable implementation strategies addressing health disparities globally. Full article