Dense prediction is a fundamental problem in medical image analysis. As Convolutional Neural Networks (CNNs) are limited by the intrinsic locality of convolution operations, transformers with the ability to capture long-range visual dependency have been widely adopted for dense prediction. However, due to the high computation and memory loads of self-attention operations, transformers are typically applied at downsampled resolutions (e.g., after patch embedding), which cannot effectively leverage the tissue-level textural information that is recognizable only at high-resolution image features (e.g., full/half of the image resolution). Unfortunately, this textural information is crucial for differentiating subtle human anatomy/pathology in medical images. In this study, we hypothesize that Multi-Layer Perceptrons (MLPs) are superior alternatives to transformers for medical dense prediction, as they can capture finer-grained long-range dependency at higher-resolution features under equal computation/ memory constraints. To validate this, we conducted a comprehensive empirical investigation of MLPs in various medical scenarios. We built a hierarchical MLP framework that applying MLPs to extract image feature pyramids beginning from the full image resolution, and then evaluated it with various MLP blocks on diverse dense prediction tasks, including medical image restoration, registration, and segmentation. Extensive experiments on six public datasets show that applying MLPs at higher resolutions yielded superior performance over CNN- and transformer-based counterparts across all evaluation tasks. Our findings suggest that MLPs can serve as superior medical vision backbones over CNNs and transformers, with significant potential to influence future model designs for medical dense prediction.
As a key diagnostic and monitoring tool, medical imaging has drawn considerable attention in tumor microenvironment (TME) research, emphasizing the need to understand current research trends, key focus areas, and emerging developments. To explore this field comprehensively, this study utilizes bibliometric methods to analyze research on the TME through medical imaging. The relevant research publications were retrieved from the Web of Science Core Collection (WoSCC) database. CiteSpace was utilized to visualize the co-occurrence of institutions and keywords and create dual-map overlay of journals. VOSviewer generated visual maps for authors, citations, and references. Furthermore, some analyzed data, including publication trends and cooperative relationships, were plotted using R. The final study included 1,256 articles for further analysis. The number of publications rose steadily from 5 in 2008 to 215 in 2025, with citations also showing sustained growth from 4 to 5,628 over the years. The China (n=446, 35.51%) was the most prominent contributor. Krishna Murali C. (n=16) had the highest publication count. The USA (total link strength =201), the University of California System (n=42), and Ellingson Benjamin M. (total link strength =82, collaboration =26) were the respective leaders in the field. Cancer Research [impact factor (IF) =16.6, average citations =76.85] led the field in both IF and average citations. This study highlights the current research status and hotspots in TME through medical imaging and points to the feasibility of using medical imaging to evaluate the TME components more accurately. The potential for monitoring the response to immunotherapy and image-guided individualized treatment will be key future research trends.
Breast cancer is the most common malignant tumor affecting women, and pathology serves as the primary method for its diagnosis. In recent years, artificial intelligence (AI) has increasingly been applied to the pathological diagnosis of breast cancer. This study, therefore, aims to conduct a bibliometric analysis of the literature on artificial intelligence in breast cancer pathology (AIBCP) to map the research landscape, identify key trends, and highlight emerging directions. A total of 1089 relevant publications from 2011 to 2025 were retrieved from the Web of Science (WOS) database. Bibliometric analysis was conducted using bibliometric VOSviewer, and CiteSpace, while BERTopic modeling was applied to extract semantic topics from abstracts. The study examined publication trends, geographical and institutional contributions, author and journal profiles, citation patterns, references, as well as keywords and abstract-based topics. Research on AIBCP has grown substantially since 2017, with 2025 recording the highest number of publications. The majority of them are original articles (70.25%). The leading contributing countries and regions are the USA, China, India, and the European Union, while the University of Toronto emerged as the most productive institution. Key influential authors include Jeroen van der Laak and Nasir Rajpoot. The most cited study focused on deep learning for lymph node metastasis detection. Fourteen distinct research themes were identified, with Topic 0, centered on deep learning for histopathology image analysis, being the dominant topic (53.54%). Other topics included mitosis detection and analysis, tumor microenvironment and prognosis, spatial transcriptomics and gene expression, natural language processing, and multimodal and explainable AI models. Topic analysis reveals a shift from unimodal image analysis towards multimodal "pathomics" and emerging interdisciplinary areas. The field of AIBCP is rapidly evolving, driven by deep learning-based pathological image analysis and increasingly expanding into comprehensive multimodal and clinical research. Future directions should focus on addressing the AI "black box" problem, to improve interpretability, advancing multimodal approaches, and optimizing human-AI collaboration frameworks. This bibliometric analysis provides a foundational resource to guide researchers and clinicians in navigating and advancing the field.
Conventional cytogenetic analysis remains central to the diagnosis and risk stratification of hematological malignancies but is constrained by labor-intensive workflows, inter-observer variability, and sensitivity to image quality. Although artificial intelligence (AI) approaches have been proposed for individual analytical tasks, clinically integrated, end-to-end pipelines aligned with reporting standards remain limited. We developed and evaluated a clinically oriented, AI-assisted cytogenetic analysis pipeline integrating image enhancement, chromosome detection, numerical and targeted structural abnormality assessment, and standardized reporting. Image quality was enhanced using a Pix2Pix-based generative model, followed by chromosome localization with a YOLOv8 detector and structural classification using a Siamese ResNet-18 architecture. Outputs were translated into ISCN-formatted reports aligned with College of American Pathologists (CAP) requirements. The system was evaluated retrospectively on clinical bone marrow karyogram datasets, with performance assessed using image-level fidelity metrics, analytical performance, and system-level feasibility under expert oversight. Image enhancement demonstrated high structural fidelity (mean SSIM >0.98). Within a gated, quality-controlled pipeline, chromosome detection achieved robust performance, and the structural classifier showed strong discriminative ability for the targeted abnormality t(9;22). At the case level, primary concordance with expert interpretation was 85.8%, increasing to 92.35% following adjudication of clinically acceptable reporting differences. Automated generation of CAP-aligned ISCN reports was achieved under predefined quality-control constraints with mandatory expert validation. This study presents a clinically grounded, proof-of-concept AI-assisted cytogenetic decision-support pipeline integrating enhancement, gated analysis, abnormality assessment, and structured reporting. While demonstrating encouraging system-level performance, the framework is intended as a decision-support tool and remains at the stage of controlled feasibility evaluation. Further prospective validation, expanded structural coverage, and deployment-level governance will be required before clinical implementation.
Automated whole-slide image (WSI) analysis, specifically applications of deep learning (DL)-based algorithms, has been enabling automated detection, classification, segmentation, and prognosis for various diseases. Performance evaluation plays an important role in the success of these complex big-data-based technologies. Our purpose is to conduct a performance evaluation of DL segmentation models applied to a breast cancer WSI dataset provided by the Tumor InfiltratinG lymphocytes in breast cancER challenge and investigate methodological issues in the assessment of WSI segmentation models. We evaluated the performance of DL models in the segmentation of tumoral and stromal regions and the effect of color normalization on improving the performance of these models when the training and testing data are from different sources. One important issue is the aggregation of image segmentation performance when the reference standard includes annotations only from selected regions of interest (ROIs). We introduced three different methods for aggregating performance based on different units of analysis (pixels, ROIs, and slides) and a bootstrap method to estimate the variance of the performance results at the slide level. We found that using different units of analysis can produce not just different mean performance estimates but also different levels of uncertainty. Our results also showed that color normalization significantly improved DL model performance when the training and testing data are from different sources. Our study demonstrates the importance of image acquisition, study design, and statistical analysis methods used in the performance evaluation of computational pathology applications.
Eating disorders are serious, multi-systemic and chronic disturbances in eating behavior among young people. Body image dissatisfaction is a known risk for developing eating disorders; therefore, this study aims to explore the proportion of participants screening-positive for elevated eating disorder risk among medical students in Jordan and their relationship with associated factors including body shape concerns. A cross-sectional study that used an online survey to collect the data for 402 undergraduate medical students (117 males; 285 females) from six public universities in Jordan. The surveying tool included a sociodemographic section, Eating Attitudes Test (EAT-26), and Body Shape Questionnaire (BSQ-16B). IBM SPSS Statistics (version 27) was used for descriptive statistics, multivariate analysis and logistic regression to identify the factors associated with the increased risk of eating disorders. Female students had higher mean EAT-26 scores compared to males (p = 0.03), while BSQ-16B scores did not differ by gender. BMI was significantly associated with both eating attitudes and body shape concerns, with obese participants demonstrating higher EAT-26 scores than those with normal BMI (p = 0.029) and a graded increase in BSQ-16B scores across the BMI categories (all p < 0.001). Students living alone reported higher BSQ-16B scores compared to those living with their families (p = 0.029). Participants with a history of psychiatric illness, use of diet pills or laxatives, or recent weight loss greater than 10 kg had significantly higher EAT-26 and BSQ-16B scores (all p ≤ 0.047). Females were more likely to fall into the high-risk group compared to males (36.1% vs. 29.9%, p = 0.03). This study revealed that a considerable proportion of medical students in Jordan were screening-positive for elevated eating disorder risk. The strongest associated factors were female gender, higher BMI, rapid weight loss, use of diet pills or laxatives, and a history of psychiatric illness, with a strong correlation observed between eating-disorder risk and body shape concerns. These findings highlight the need for further research and targeted preventive strategies to support the mental and physical wellness of medical students.
To investigate patient priorities that may inform the choice of laryngeal cancer treatment. Multi-institutional, survey-based, conjoint analysis study focusing on seven attributes: lifespan, treatment type, cancer cure, self-image, mode of breathing, voicing, and swallowing. Patients with a history of treated laryngeal cancer (>6 months from treatment completion with no evidence of recurrent disease) were recruited. Tertiary care medical centers. Conjoint analysis yields utility scores, a quantitative measure of preference for an attribute. Higher utility scores indicate greater preference. Chi-squared, univariate logistic regression, and univariate linear regression analyses were used to evaluate associations between patient demographic and medical features with relative attribute preference. This study included 151 patients with previously treated laryngeal cancer. For the cohort, the mean importance scores (±standard deviation) were swallowing 25.7% (±8.4%), lifespan 21.5% (±9.3%), cancer cure 14.0% (±6.4%), mode of breathing 12.8% (±4.8%), voicing 9.2% (±3.5%), treatment type 9.1% (±5.0%), and self-image 7.7% (±4.4%). Patients who required salvage surgery after upfront chemoradiotherapy placed more value on cancer cure compared to the other treatment groups (coefficient 2.76, 95% CI 0.33-5.19). In patients with a history of treated laryngeal cancer, swallowing is the most important treatment priority, followed closely by lifespan. However, patients who underwent salvage surgery placed more value on cancer cure during decision-making. These findings demonstrate that patient treatment preferences are diverse and may change throughout the cancer care journey, with some patients placing a higher value on quality of life than lifespan and cancer cure.
Brain tumour detection and analysis using medical imaging requires the extraction of both local spatial features and global contextual representations. Although convolutional neural networks (CNNs) excel at capturing local spatial patterns and Transformer-based architectures model long-range dependencies effectively, the optimal architectural paradigm for clinical deployment remains unresolved. This systematic review and meta-analysis evaluates hybrid CNN-Transformer architectures for brain tumour detection, focusing on the integration of local and global feature learning, diagnostic accuracy and computational efficiency. The roles of generative adversarial networks (GANs) for addressing data scarcity and multimodal imaging fusion for diagnostic completeness are also critically examined. A systematic search was conducted across IEEE Xplore, PubMed, Scopus and Google Scholar for studies published between January 2021 and May 2025. From 1876 initially identified articles, 94 met the prespecified inclusion criteria following quality assessment using the QUADAS-2 and ROBINS-I frameworks. A random-effects meta-analysis of diagnostic accuracy was performed using the DerSimonian-Laird estimator, with statistical heterogeneity quantified using I2 and publication bias assessed using funnel plot asymmetry and Egger's test. Computational efficiency was standardised to GigaFLOPs using a reference input of 240 × 240 × 155 voxels (BraTS benchmark), with FLOP estimates derived from primary publications where available and bounded by theoretical complexity formulas otherwise, with estimated values explicitly distinguished throughout. Across all 94 included studies, the pooled diagnostic accuracy was 93.5% (95% CI: 92.7%-94.4%); however, confirmed publication bias (Egger's p = 0.043) indicates this represents an upper-bound approximation rather than an unbiased population estimate. Because subgroup study counts were insufficient for formal random-effects pooling (CNN-only: n = 3; Transformer-only: n = 2; CNN-Transformer hybrid: n = 4; minimum recommended n = 10 per subgroup), no subgroup meta-analysis was performed. Instead, descriptive mean accuracies are reported as hypothesis-generating observations only: CNN-only models 91.7%, Transformer-only models 93.6% and CNN-Transformer hybrid models 94.6%. These figures must not be interpreted as pooled meta-analytic estimates; they reflect mean observed accuracy across a small number of included studies and are reported solely to illustrate directional trends consistent with the mechanistic rationale for hybridisation. Substantial heterogeneity was observed (I2 = 78.3%; p < 0.001). Three integration paradigms were identified: sequential (45% of models; 93.8% accuracy; 1.8 GFLOPs), parallel (32%; 94.3%; 2.8 GFLOPs) and hierarchical (23%; 94.9%; 3.5 GFLOPs). Parallel architectures demonstrated optimal clinical viability, balancing accuracy with a mean inference time of 2.1 s. GAN-based augmentation improved rare tumour class detection by 7%-10%, with conditional GANs outperforming vanilla architectures. Multimodal MRI + PET fusion achieved 94.2% accuracy at 2.8 GFLOPs, whereas triple-modality integration yielded marginal additional gains (95.1%) at substantially elevated computational cost (9.1 GFLOPs). Notably, 65% of included studies used the BraTS benchmark exclusively, and hybrid model accuracy declined from 94.6% on high-grade gliomas to 88.3% on low-grade gliomas, with hybrid architectures exhibiting 2.3× greater susceptibility to Gaussian noise than CNN-only equivalents, limitations that constrain generalisation to real-world clinical settings. Descriptive comparison of mean observed accuracies based on study counts is insufficient for confirmatory meta-analysis, suggesting hybrid CNN-Transformer architectures may offer diagnostic accuracy advantages over CNN- and Transformer-only approaches; this observation is hypothesis-generating only and requires validation in a larger, more balanced evidence base. Among integration strategies, parallel architectures demonstrated the most favourable accuracy efficiency balance in the reviewed evidence. GANs and multimodal imaging function as essential architectural enablers, addressing data scarcity and diagnostic incompleteness, respectively. Significant challenges remain in computational efficiency, noise robustness and generalisation to rare tumour subtypes, representing priority directions for future research.
This study introduces a novel geometric marker, derived from CT image data analysis via grey theory and prototyping verification, to enhance the planning of sacroiliac joint screw placement, aiming at optimizing safety. A dataset comprising 107 adult cases with hip thin-slice CT scans in DICOM format was collected for statistical analysis. The sacral bone model was segmented from the scanning data sets to facilitate planning for sacroiliac screw surgery. The saddle point (SP), located at the intersection of the contralateral sacral promontory and the sacral wing, was identified as the target direction for needle insertion. The entry point on the anterior convexity and posterior concavity of the ear-shaped surface was connected to the posterior superior aspect to determine the retrograde positioning of the screw channel. Utilizing a home-made software, E3D, the safe channel was visualized using three cross-sectional views with the screw channel as the axis. Adjustment of the central screw channel's position in the transverse and coronal planes facilitated its establishment. The distances between the highest (up, u) and lowest (down, d) entry points on the ear-shaped surface (ud, mm), as well as the distances between the most anterior (ahead, a) and most posterior (back, b) points (ba, mm), were measured to approximate the area (surface, s, mm2) within the four points, aiding in assessing the size of the safe range. Out of the 107 cases, two were unable to undergo sacroiliac screw placement due to sacral morphological variations (1 male and 1 female). In the remaining 105 cases (61 male and 44 female), 98.13% successfully underwent bilateral safe screw placement (53 cases on the left side and 52 on the right side). Statistical tools and grey relation analysis are applied to extract templates for data reuse and validation. The utilization of the central screw channel as the axis in the three-section view facilitated a direct three-dimensional assessment of the screw channel's safety, ensuring its containment within the bone tissue. Consequently, planning the sacroiliac screw channel with the contralateral saddle point as the target direction and the posterior superior aspect of the ear-shaped surface as the entry point is deemed safe and feasible, notwithstanding individual variations. To validate the findings, a printed sacroiliac joint prototype and a real human sacroiliac joint specimen were employed to simulate puncture insertion surgery. The results demonstrate the practicality of the newly identified marker as a crucial reference point for surgical planning. Subsequent steps will involve the statistical analysis of additional medical specimens and the execution of clinical surgeries to demonstrate its application.
Polycystic Ovary Syndrome (PCOS) is a multifaceted disorder with physical, emotional, and social implications. Culturally adapted tools for assessing Health-Related Quality of Life (HRQoL) in Indian women, particularly those under 18 and over 45 years of age, need to be developed. This study aimed to culturally adapt the Nasiri-Amiri PCOS-specific HRQoL questionnaire for Malayalam-speaking women, evaluate its reliability and factor structure using exploratory factor analysis (EFA), and identify key HRQoL domains in this population. Between July 2024 and December 2025, a 52-item Malayalam-adapted questionnaire was distributed to women aged 13-52 years, of whom 201 met the inclusion criteria. Reliability assessments were performed using Cronbach's alpha. The ability of the data to be factored was evaluated using the Kaiser-Meyer-Olkin test and Bartlett's Test of Sphericity. Maximum likelihood extraction with Equamax rotation was used for exploratory factor analysis, with factor retention guided by the root-mean-square residual. Strong internal consistency for the instrument (Cronbach's α = 0.818) and adequate sampling adequacy (KMO = 0.794; Bartlett's p < 0.001) were assessed for exploratory factor analysis (EFA), which provided eight clinically relevant HRQoL domains (29 items total). The domains identified in this study included hirsutism, skin disease, menstrual irregularities, physical/emotional burden, emotional inconsistency, body image, problems with self-care, and social support. The major contributors to impaired HRQoL were menstrual dysfunction and other visible symptoms. Within the study sample, women reported greater distress associated with their weight than their appearance. Paracetamol use was comparatively higher for self-medicated pain relief during menstruation. Extending the assessment beyond the conventional 18-45 year reproductive window reveals that the PCOS-related burden persists into adolescence and adulthood. The adapted Malayalam PCOS-HRQoL tool showed acceptable reliability, making it a useful measure of quality of life in women with PCOS in this population. Further confirmatory factor analysis and longitudinal studies are required to confirm these findings.
Malaria is a potentially fatal illness caused by a parasite of the genus Plasmodium that humans get by being bitten by female Anopheles mosquitoes carrying the infection. The incidence of malaria worldwide is disproportionately high in the African continent. Automated systems and cognitive analysis of digitized images of blood smears were used to diagnose Plasmodium malaria. This method is implemented in the Aidos intelligent system, which is easily accessible online. For the study, the database included images of 191 blood smears of patients infected with malaria and 227 images of blood samples from healthy patients. The images were digitized using the method developed by Professor Lutsenko E.V. The images were digitized for 12 light spectra. Then, spectral analysis of the blood smear images was carried out only for 18 new patients, and the duration was 10 seconds. The average similarity value of Plasmodium malaria recognition in patients was achieved at 66.965%. No false positive decisions were obtained for digitalized blood smears from healthy patients. The automated system-cognitive analysis of digitized blood smears provides instant diagnostic support. It allows medical workers with limited knowledge in microscopy and artificial intelligence to perform diagnostics.
Reliable optical coherence tomography angiography (OCTA) requires not only high-resolution acquisition but also standardized image quality control to ensure accurate vascular quantification. However, existing quality assessment approaches largely rely on subjective grading or global image descriptors and do not account for the region-dependent characteristics of OCTA decorrelation signals. Here, we propose an anatomically informed OCTA quality assessment framework that integrates multi-region segmentation with statistically calibrated semi-supervised learning. The segment anything model is employed to partition each image into large vessels, capillary networks, and the foveal avascular zone (FAZ), enabling that region-specific evaluation accounts for heterogeneous artifact sensitivity and optical signal formation mechanisms. Physically interpretable metrics, including vessel edge sharpness, contrast-to-noise ratio, and signal-to-noise ratio, are extracted to construct a decorrelation-grounded feature space. A distribution-calibrated grading strategy with a modulation factor of 0.9 is introduced to support stable grading under limited annotations. Evaluated on the public OCTA-500 dataset and an independent clinical dataset, the framework achieves an average Dice coefficient of 81.21 percent for region segmentation and a grading accuracy of 90.0 percent with a Cohen kappa of 0.864. By transforming global heuristic scoring into anatomically resolved and measurement-consistent evaluation, the framework supports automated data filtering for AI training pipelines and can be integrated into OCT acquisition workflows for device-level performance calibration and standardized quality control.
Non-invasive colorectal cancer (CRC) screening offers an important opportunity to increase colonoscopy participation and reduce mortality. This study evaluates the potential of the gut-liver axis to predict colorectal neoplasia using artificial intelligence (AI)-based analysis of the liver in routine CT images as an opportunistic screening approach. In this retrospective study, data from 1,997 patients were analyzed, including 1,189 without neoplasia and 808 with colorectal neoplasia (423 adenomas, 385 CRC). Radiomic features were extracted from three-dimensional liver segmentations, and the dataset was split into training (n = 1,397) and test (n = 600) cohorts. Five machine learning models were trained using five-fold cross-validation on the 20 most informative features. The best-performing radiomics-based XGBoost model achieved a test AUROC of 0.810 (95% CI: 0.767-0.837), outperforming a clinical-only model (AUROC: 0.457). After threshold optimization, sensitivity reached 74.1% and specificity 72.3% for detecting colorectal neoplasia. Subclassification between CRC and adenoma was less accurate (AUROC: 0.674). These findings demonstrate that AI-based liver analysis from routine CT scans can predict colorectal neoplasia, supporting its potential as an accessible adjunct to CRC screening and highlighting the gut-liver axis as a novel biomarker source.
Colorectal polyps are key determinants of colorectal cancer. Their accurate detection during colonoscopy has been a technically challenging work due to differences in shape size imaging conditions and texture. Emerging advances in Artificial Intelligence predominantly in deep learning have been making significant changes in the automatic detection and classification of polyps. This review presents a systematic and in-depth analysis of artificial intelligence-based methods for colorectal polyp detection classification and segmentation. Publicly available datasets are extensively reviewed along with data pre-processing and augmentation techniques that highlights low contrast noise and class imbalance. The review also investigates about the present state-of-the-art models for all three tasks. It is based on architecture designs performance trends and relative strengths. A thorough assessment has been made for the standard performance metrics used in existing literature for fair and consistent benchmarking. Finally existing gaps and future research paths have been discussed with an objective to fill the performance-translation gaps between experimental performance and clinical deployment. This review gives a structured reference for AI-based colorectal polyp analysis.
Vision models for medical imaging often require tens of millions of parameters, raising questions about whether architectural efficiency can be achieved without sacrificing classification performance. We introduce MedLiT-seed (2.1 Million parameters) and MedLiT-nano (0.75 Million parameters), two ultra-lightweight vision transformers designed for efficient and scalable medical image analysis. MedLiT employs a streamlined Mixture-of-Experts (MoE) architecture with SwiGLU feedforward networks, grouped query attention, and depth-wise scaling. Models were pre-trained using masked autoencoding on ImageNet and MedMNIST, followed by fine-tuning on 12 MedMNIST 2D subsets. We evaluated performance across multiple configurations and compared against benchmark models including ResNet, MedViT, and AutoML systems. MedLiT-seed achieved the highest Area Under Curve (AUC) on 4 subsets and second-highest on 2 others, outperforming models with 10-20× more parameters. MedLiT-nano achieved results comparable to, and even exceeding, ResNet-18 and AutoML baselines in several subsets. Transfer learning from ImageNet significantly improved convergence and generalization. Increasing embedding size yielded greater performance gains than increasing expert count. MedLiT demonstrates that MoE-based token routing represents a viable architectural pathway for achieving competitive accuracy relative to its floating-point operations (FLOP) across diverse medical imaging modalities on the order of 2M parameters. These results suggest that selectively routing computation through specialised experts, rather than scaling model size, can serve as an effective design principle for more compact medical vision models. Such architecture can be utilised for low-resource clinical environments and scalable fine-tuning across diverse healthcare tasks, though limitations on multi-label tasks highlight clear directions for future architectural refinement.
Purpose To quantify regional respiratory function using four-dimensional free-breathing dynamic MRI (dMRI) and evaluate vertical expandable prosthetic titanium rib surgery impact on diaphragm curvature in pediatric thoracic insufficiency syndrome (TIS) using pre- and postoperative and control comparisons. Materials and Methods Curvature was retrospectively analyzed in 149 pediatric patients with TIS from May 2010 to November 2022 (49 pre- and postoperative, 70 preoperative-only, 30 postoperative-only dMRI) and compared with 190 controls. Mean follow-up ± SD was 2.4 years ± 1.8. Diaphragm contours were delineated at end-expiration and end-inspiration, and curvature was quantified across 13 regions per hemidiaphragm. Analyses included paired t tests, one-way analysis of variance to compare with controls, and correlation analyses relating postoperative curvature to ventilatory status and thoracic Cobb angle. Results Patients with TIS (mean age, 3.5 years ± 3.5; 27 male) demonstrated region-, plane-, and phase-dependent preoperative curvature differences compared with controls (mean age, 11.9 years ± 3.6; 92 male). Significant pre- to postoperative curvature changes were limited to five region-plane-phase combinations. The right hemidiaphragm anterior-lateral region at end-expiration showed the only sagittal-plane change (6.3 m-1 ± 0.7 to 8.2 m-1 ± 0.6, P = .02), approaching control values (9.3 m-1 ± 2.6). Several regions were no longer different from controls, most prominently in the right hemidiaphragm coronal plane at end-inspiration, whereas others, particularly sagittal end-inspiration regions, remained different (P < .05). Postoperative curvature correlated with ventilatory status, strongest in central sagittal regions (ρ ≤ 0.392, P < .001), and with thoracic Cobb angle in posterior sagittal regions (ρ ≤ 0.385, P < .001). Conclusion Surgery resulted in plane- and phase-specific improvements in diaphragm curvature, with partial normalization toward control values predominantly in coronal-plane regions. Keywords: Pediatrics, MR-Functional Lung Imaging, MR-Imaging, Pulmonary, Diaphragm, Anatomy, Treatment Effects, Outcomes Analysis, Comparative Studies, Curvature, Dynamic MRI, Quantitative Radiology, Shape, Thoracic Insufficiency Syndrome (TIS) Supplemental material is available for this article. © RSNA, 2026.
Bisphenol F (BPF), a primary substitute for bisphenol A (BPA), is widely utilized in industrial production and daily life. However, its widespread environmental presence has raised concerns regarding potential health risks. This study aims to investigate the potential toxic targets of BPF in the pathogenesis of non-alcoholic fatty liver disease (NAFLD). Initially, potential target genes of BPF were identified using the ChEMBL, STITCH, and SWISS databases. NAFLD-related genes were obtained from the OMIM and GeneCards databases, yielding a preliminary set of 28 overlapping candidate targets. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were subsequently performed to elucidate the biological processes and signaling pathways potentially affected by BPF. Differential expression analysis of transcriptomic data from NAFLD and normal liver tissues obtained from the GEO database (GSE260666) revealed that CYP2C19 and SHBG were significantly upregulated in NAFLD samples, suggesting their potential as key targets of BPF. Molecular docking simulations using AutoDock demonstrated stable binding conformations between BPF and both CYP2C19 and SHBG proteins, with favorable binding free energies indicating strong interactions. Furthermore, molecular dynamics simulations confirmed the structural stability of the protein-ligand complexes under simulated physiological conditions. These findings provide a theoretical basis for understanding the toxic targets and mechanisms of BPF in NAFLD pathogenesis and offer insights for the prevention and treatment of NAFLD associated with BPF exposure from plastic products.
Somatoform disorders (SDs), including somatic symptom disorder and body dysmorphic disorder, are increasingly recognized for their impact on surgical decision-making. Their influence on breast reconstruction (BR) timing and type following mastectomy remains underexplored. Using the TriNetX National Health Research Network, we conducted a retrospective cohort study of 192,618 breast cancer patients who underwent mastectomy between 2005 and 2023. Patients were stratified into cohorts based on the presence or absence of a documented SD diagnosis (ICD-10: F45). Propensity score matching was applied to balance baseline characteristics. Primary outcomes included BR timing (immediate and delayed) and method (autologous and implant-based). Among the cohort, 3067 (1.6%) had an SD diagnosis. Immediate BR rates were not significantly different between groups. However, patients with SDs were significantly more likely to undergo delayed BR: HR = 1.59 (95% CI 1.21-2.09) for autologous and HR = 1.24 (95% CI 1.06-1.45) for implant-based reconstruction. Overall BR rates within two years were modestly higher in the SD group. While SDs do not deter patients from choosing immediate BR, they are associated with delayed reconstruction, likely reflecting psychological factors, such as decisional conflict, risk sensitivity, and body image concerns. Integrating mental health assessment into preoperative counseling may facilitate more timely and individualized BR decisions. Further research should examine long-term satisfaction and outcomes in this population. This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266 .
Vitiligo is a common depigmenting disorder frequently misdiagnosed due to its visual similarity to other hypopigmentary conditions. While artificial intelligence (AI) has shown promise in dermatological image analysis, most models lack interpretability and fail to provide actionable clinical recommendations. To develop and validate an AI-assisted diagnostic system that integrates a large language model (LLM) for differentiating vitiligo from ten other hypopigmentary disorders, while providing interpretable characteristics and structured clinical reports. We retrospectively collected clinical images from a multicenter cohort, including patients diagnosed with vitiligo or one of ten other hypopigmentary disorders across five hospitals in China. A multi-task Vision Transformer model was trained to classify eight key clinical characteristics and output diagnostic probabilities. The model's structured predictions were then fed into the DeepSeek LLM to generate comprehensive clinical reports. Diagnostic performance was evaluated on an independent set of 175 images (87 vitiligo and 88 non-vitiligo) and compared with 43 dermatologists. The study included 13,322 clinical images from 2,974 patients. For distinguishing vitiligo, the model achieved an AUC of 0.9906 (95% CI: 0.9844-0.9968), with sensitivity of 98.29% and specificity of 93.73%. Through multi-task learning, the model demonstrated accurate classification of eight key clinical characteristics, notably achieving 88.12% accuracy in identifying typical location and 86.78% in recognizing edge morphology. In a comparative test using 175 independent images, the AI model (AUC = 0.98) significantly outperformed a panel of dermatologists, particularly in diagnostic sensitivity. Moreover, the system successfully generated clinical reports via the DeepSeek LLM, providing structured diagnostic suggestions, differential diagnoses, and treatment plans. This multicenter study presents a clinically interpretable AI system for accurately discriminating vitiligo from similar disorders. By generating LLM-based reports, it enhances diagnostic transparency and supports clinical decision-making, offering potential assistance especially in resource-limited settings and advancing intelligent tools in dermatology.
Cellular signals are essential for sensing the microenvironment and coordinating physiological functions. Their multimodal nature, encompassing electrophysiological, chemical, mechanical, and optical components, helps define cellular functional states and fate. High-resolution spatiotemporal analysis of these weak, dynamic, and heterogeneous signals is critical for elucidating fundamental life processes, uncovering disease mechanisms, and advancing precision medicine. Recent advances in integrated circuit (IC) technology, particularly the co-integration of complementary metal-oxide-semiconductor (CMOS) and micro-electro-mechanical systems (MEMS), have enabled unprecedented capabilities for cellular signal analysis, driving a transition from conventional instruments and microfluidic platforms to chip-level cellular analysis. This review summarizes the key technological foundations, multimodal sensing mechanisms, and emerging applications of IC technology for cellular signal analysis. It explores the core principles of high-sensitivity, long-term stable signal acquisition, focusing on bioelectronic interfaces, biocompatible packaging, and low-noise signal processing. It also reviews breakthroughs in microelectrode arrays, field-effect transistors, CMOS image sensors, MEMS sensors, and multi-parameter chemical sensing chips for single-cell and population-level detection. The review highlights applications in drug screening, clinical diagnostics, single-cell analysis, and brain-computer interfaces. Finally, it addresses challenges such as biocompatibility, crosstalk suppression, and energy efficiency, while outlining future directions in material innovation, three-dimensional integration, and brain-inspired computing.