Multi-structure segmentation in brachial plexus ultrasound requires delineating small nerves alongside larger muscles and vessels despite ambiguous tissue boundaries. We propose MOSBNet, which combines category-conditioned feature processing, multi-scale fusion and auxiliary point supervision. The Target Equalized Attention Module (TEAM) uses coarse class masks for branch-specific refinement and region normalization. Upsampling Multi-Scale Fusion (UMF) aggregates decoder features, and Category-Supervised Point Reclassification (CPR) provides training-only supervision at uncertain locations. On a 200-image exploratory evaluation subset of the UBPD 2023 dataset, MOSBNet achieves an mDSC of 73.18 ± 0.69% and nerve DSC of 52.87 ± 0.54% across three training runs compared with the baseline nerve DSC of 50.35 ± 2.48%. Patient-level analysis yields a mean nerve DSC gain of 2.10 percentage points (95% confidence interval 0.87–3.45). Adding TEAM to UMF + CPR improves nerve DSC in all three runs, with a paired gain of 2.34 ± 0.92 percentage points, suggesting a contribution of TEAM within the evaluated configuration.
Background: Orthodontic treatment is widely performed in young populations, yet its post-treatment functional associations beyond dental occlusion remain poorly understood. The cranio-cervico-hyoid complex plays a central role in posture, swallowing, and airway function, and may be sensitive to developmental or treatment-related influences. This study investigated whether, among otherwise healthy young women with a diagnosis of Class II division 1 malocclusion, a history of orthodontic treatment with mandibular advancement is associated with altered passive mobility of this system. Methods: A cross-sectional observational study was conducted on 170 women aged 18–30 years with a diagnosis of Class II division 1 malocclusion (104 treated with mandibular advancement for 6–12 months (ortho group); 66 with the same diagnosis who had never undergone orthodontic treatment (ctrl group)). Anthropometric, epidemiological, clinical, and psychosocial variables were collected, including VAS, FIQR, CSI, and DASS-42. Passive mobility of the hyoid bone, C0–C1, and C2–C3 segments (primary outcome) was assessed by an experienced blinded examiner. Group differences were analyzed using chi-square, Fisher’s exact test, Welch’s t test, or Wilcoxon tests, with effect sizes and multiple-comparison corrections (Benjamini–Hochberg and Bonferroni). At the time of examination, all treated participants presented Class I occlusion, whereas all controls retained Class II division 1 malocclusion. Results: Groups were comparable in baseline anthropometric and epidemiological characteristics. The orthodontic group reported higher FIQR scores and greater pain intensity on the VAS, although these differences did not remain significant after multiple-comparison correction. Marked biomechanical differences emerged: reduced passive mobility of the hyoid bone, C0–C1, and C2–C3 was significantly more frequent in orthodontically treated subjects, with moderate to strong effect sizes. The composite variable “at least one dysfunction” was present in 74% of orthodontic subjects versus half of controls. These associations remained significant after multiple-comparison correction. Because all treated participants had been corrected to Class I while all controls remained in Class II division 1, the effect of treatment could not be separated from that of the corrected occlusion. Conclusions: Young women with Class II division 1 malocclusion previously treated with mandibular advancement showed a higher prevalence of altered passive mobility of the cranio-cervico-hyoid complex, and differences in cervical pain intensity and functional impact were small and did not remain significant after multiple-comparison correction. Because treatment history and current occlusion were completely confounded, the findings cannot establish whether these differences are attributable to mandibular advancement, occlusal correction, or pre-existing factors. Given the cross-sectional design and the lack of consistent data on retention period and exact time since treatment completion, these findings should be interpreted as post-treatment associations rather than causal or long-term effects. Further longitudinal studies are needed to clarify the temporal relationship and clinical relevance of these alterations.
Background: Large language models (LLMs) have shown potential for clinical decision support. However, the diagnostic performance of newly released frontier models—including GPT-5 and Gemini 2.5 Pro—in gastroenterology remains insufficiently evaluated. This study evaluated the diagnostic performance and repeated-run consistency of four frontier LLMs—DeepSeek-V3.1, GPT-5, GPT-4.1, and Gemini 2.5 Pro—in challenging gastroenterology cases. Methods: This cross-sectional comparative study evaluated four LLMs (DeepSeek-V3.1, GPT-5, GPT-4.1, and Gemini 2.5 Pro) using 50 challenging gastroenterology cases retrieved from PubMed. Cases published after February 2025 were selected to minimize the potential risk of training-data contamination and prior model exposure. Each case was structured into a standardized prompt comprising clinical history, examination findings, laboratory results, and imaging data. Three independent runs were performed for each case. First-run outputs were used to assess ordinal primary diagnostic scores and differential-diagnosis quality, whereas all three runs were used to assess repeated-run consistency. Outputs were anonymized and scored by a gastroenterologist blinded to model identity. Results: Gemini 2.5 Pro achieved significantly higher ordinal primary diagnostic scores than the other three models (all Holm-adjusted pairwise p < 0.01). Exact-match rates were 46% for Gemini 2.5 Pro, 22% for GPT-5, and 18% each for GPT-4.1 and DeepSeek-V3.1. All four models had a median differential-diagnosis quality score of 4; the Friedman test did not detect a statistically significant difference among the models (p = 0.725). Complete agreement in the primary diagnosis across all three runs was observed in 68–78% of cases, depending on the model. No statistically significant differences in repeated-run consistency scores were detected among the models. Conclusions: In this investigator-selected benchmark of 50 published gastroenterology cases, Gemini 2.5 Pro achieved higher ordinal primary diagnostic scores than the other three models. No statistically significant differences were detected in differential-diagnosis quality or repeated-run consistency. These nonsignificant findings do not establish equivalence. Prospective evaluation in clinical settings is required before drawing conclusions about clinical effectiveness.
Early and accurate skin cancer diagnosis is critical for improving patient outcomes, yet current deep learning classifiers often suffer from limited convergence stability, underutilize patient-level metadata, and lack traceable diagnostic reasoning. To overcome these challenges, this study presents an explainable multimodal deep learning framework optimized via a hybrid Adam–Gray Wolf Optimization (Adam–GWO) strategy. The framework integrates dermoscopic image representations with structured patient metadata (age, sex, anatomical site) through feature-level fusion. The hybrid optimizer combines Adam mini-batch gradient updates with epoch-wise GWO validation-guided swarm exploration for dynamic feature mask gating and hyperparameter tuning. Explainability is established using Grad-CAM++ visual localization maps alongside SHAP metadata feature attributions. To prevent data leakage, evaluation was conducted using group-stratified splits (70% training, 15% validation, 15% testing) based on patient and lesion identifiers across five fixed random seeds on the HAM10000 and SIIM–ISIC 2020 benchmark datasets. Across VGG16, ResNet-50, InceptionV3, and Vision Transformer (ViT-B/16) backbones, the proposed Adam–GWO strategy consistently outperformed baseline optimizers and alternative hybrid methods. On SIIM–ISIC 2020, InceptionV3 with Adam–GWO achieved 89.4 ± 0.6% classification accuracy and 0.945 ± 0.007 AUC, outperforming image-only (87.1%, 0.920 AUC) and metadata-only (72.0%, 0.776 AUC) baselines. On HAM10000, ViT-B/16 achieved 94.0 ± 0.5% accuracy and 0.973 ± 0.004 AUC. Computational profiling indicated a modest training overhead of 9.7–10.7% per epoch, while embedded feature selection reduced single-image real-time inference latency to 11.8 ms/image. Overall, the proposed framework demonstrates a robust balance of predictive performance, empirical reproducibility, computational efficiency, and clinical interpretability.
This single-case report examines dermoscopy-indexed peripheral line-field confocal optical coherence tomography (LC-OCT) reassessment during carbon dioxide (CO2) laser unroofing of a clinically suspected basal cell carcinoma (BCC). The proposed imaging contribution is the use of phase-related peripheral morphology to inform treatment-field reassessment. A 72-year-old woman with a 6 × 5 mm lesion on the right side of her back declined surgery and topical therapy. Three non-fractionated passes were performed. Manual planimetry showed a visible altered surface field that was 30.3% larger after the first pass and 14.3% larger after the second one relative to the baseline. BCC-compatible morphology was not identified in four untreated peripheral sectors approximately 1 mm beyond the visible margin at baseline, but it was detected in all four clinically corresponding sectors after the second pass. This prompted field enlargement before the planned final pass. No compatible morphology was identified in the selected final fields. Visual follow-up showed a resolving eschar at two weeks and a scar at six months. These observations do not validate tumor identity, histological margins or eradication; optical, thermal and sampling-plane effects remain possible explanations. Prospective case series with histological correlation should validate reproducibility and any clinical benefit of this proposed imaging application.
