Browse Articles

THSD7B promotes tumor progression and is associated with prognosis in gastric adenocarcinoma

THSD7B (thrombospondin type-1 domain-containing 7B) has been implicated in several malignancies; however, its role in gastric adenocarcinoma remains unclear. This study aimed to investigate the expression pattern, clinical significance, and biological function of THSD7B in gastric adenocarcinoma. Public datasets from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) were analyzed to evaluate THSD7B expression and its association with clinical outcomes. Functional enrichment analysis was performed to explore potential biological processes. In vitro assays, including cell proliferation, colony formation, wound healing, and Transwell invasion, were conducted following THSD7B knockdown or overexpression in gastric cancer cell lines. In addition, a xenograft model was established to assess tumor growth in vivo. THSD7B expression was significantly elevated in gastric adenocarcinoma tissues compared with normal controls and was associated with patient survival. Functional analyses suggested that THSD7B-related genes were mainly enriched in cell adhesion and cytoskeleton-associated processes. In vitro experiments showed that THSD7B knockdown suppressed cell proliferation, migration, and invasion, whereas overexpression produced the opposite effects. Consistent with these findings, THSD7B modulation was accompanied by alterations in adhesion-related signaling molecules and phenotype-associated protein expression. In vivo, THSD7B promoted tumor growth in xenograft models. In conclusion, THSD7B is associated with tumor progression and clinical outcomes in gastric adenocarcinoma and may be involved in the regulation of cell motility-related processes. These findings suggest that THSD7B may serve as a potential biomarker in gastric cancer.

Editorial Note: Soil quality indicator-based land productivity modelling for agricultural sustainability

State and federal policies and school meal participation: A descriptive analysis from Arizona

Starting in 2023, the state of Arizona implemented a series of policies to improve access to school meals: (1) A state policy to eliminate co-pays for reduced-price meals, implemented in January, 2023; (2) the federal Medicaid Direct Certification demonstration project, implemented in August, 2023; and (3) a federal policy to expand eligibility for Community Eligibility Provision (CEP), implemented in October, 2023. We track changes in school meal participation rates and number of meals served for breakfast and lunch after implementation of these policies using longitudinal data from 1,730 public and charter schools in Arizona. Outcomes were compared across different policy periods with baseline- the period prior to implementation of the three policies. Following the co-pay policy elimination, compared to baseline, participation among students eligible for reduced-price category increased significantly by about 10%, for both breakfast and lunch. After the implementation of all three policies, compared to baseline, student participation across all eligibility categories increased by 4.8% and 7.2% for breakfast and lunch, respectively. Similarly, compared to baseline, with all three policies in place, the average number of free meals served daily increased by 27.9% and 26.8% for breakfast and lunch, respectively. School CEP participation expanded from 451 to 737 schools- a 63% increase following the implementation of the lower eligibility threshold for CEP participation. As a result of increased school CEP participation, an additional 147,125 children had access to free school meals. Participation trends across eligibility categories varied by school urbanicity, grade level, and demographic characteristics of enrolled students. These findings suggest that implementing state and federal policies can enhance access to and participation in school meal programs.

KYNU in macrophages contributes to the unique immune feature of LUAD via integrating single-cell and bulk RNA sequencing data: an exploratory analysis

Background Lung adenocarcinoma (LUAD) is a predominant subtype of lung cancer associated with an unfavorable prognosis. However, the roles of the tumor microenvironment (TME) and Kynureninase (KYNU) in LUAD remain largely unclear. This study aimed to investigate the potential role of KYNU in macrophages and LUAD. Methods All LUAD related data were downloaded from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. The expression of KYNU was analyzed across different cell types following dimensionality reduction analysis. Immune cell infiltration and immunotherapy response prediction were performed using CIBERSORT and TIMER, respectively. Gene set variation analysis (GSVA) was employed for functional enrichment. Results Among all immune cells in LUAD, KYNU was primarily expressed in monocytes and macrophages. The upregulated genes in KYNU+macrophages group were significantly enriched in in gene ontology (GO) terms related to antigen processing and presentation. There were increased MHC-I/ MHC-II signal interactions between KYNU+macrophages and B cells as well as T cells. In LUAD patients with higher proportions of KYNU+macrophages, a significantly greater number of patients benefited from immunotherapy ( p  = 0.033). GSVA results indicated that the MHC pathway was significantly activated in high KYNU+macrophage group. Conclusions KYNU is primarily in LUAD macrophages, contributing to the distinct immune features and correlating with the enhanced antigen presentation in LUAD. This study preliminarily confirms that KYNU may serve as a potential biomarker for immunotherapy.

Identification of AC025811.3 and AC012354.6 as two critical survival-related lncRNAs for uterine corpus cancer

Uterine corpus endometrial carcinoma (UCEC) ranks as the most frequently diagnosed gynecologic malignancy and the second leading cause of gynecologic cancer-related mortality. Long non-coding RNAs (lncRNAs) have emerged as critical regulators of gene expression and tumor biology; however, their prognostic significance in UCEC remains largely unexplored. To systematically identify survival-associated lncRNA biomarkers, we integrated clinical and transcriptomic data from two independent cohorts: TCGA-UCEC (548 tumor, 35 normal) and CPTAC-Uterus (102 tumor, 15 normal). Following upper quartile normalization and ComBat-based batch effect correction, differentially expressed lncRNAs (FDR < 0.01, |log2FC| > 2) were identified using Student’s t-test. The intersecting set of consistently dysregulated lncRNAs from both cohorts was subjected to Cox proportional hazards regression to identify survival-associated candidates. Functional inference was performed through Spearman correlation with protein-coding genes and Gene Set Enrichment Analysis (GSEA) of KEGG pathways. A total of 550 lncRNAs were consistently downregulated and 148 were upregulated in UCEC across both cohorts. Cox regression identified 30 survival-associated lncRNAs (FDR < 0.01), all with elevated expression correlating with worse overall survival. The top two candidates, AC025811.3 and AC012354.6, showed significant stage-dependent expression patterns across FIGO stages I–IV and were functionally enriched in immune regulation and carbohydrate metabolism pathways. In conclusions, AC025811.3 and AC012354.6 represent novel candidate prognostic lncRNA biomarkers in UCEC. Experimental validation, including FISH-based tissue localization and staged quantification, and functional assays, is warranted to confirm their biological roles.

Motherhood choice in multiple sclerosis (MoMS) – Pilot trial of web-based decision support

Backgroud Uncertainty concerning motherhood is common among women with multiple sclerosis (wwMS). Therefore, we developed and pre-tested a patient decision aid (PtDA) and a nurse-led decision coaching intervention (DC) to support motherhood choice. The DC includes the PtDA, a decision guide on motherhood choice, and decision coaching. Methods We conducted a randomised pilot trial across Germany to test feasibility, with decisional conflict (Decisional Conflict Scale, DCS) as an exploratory endpoint. Initially, we planned a 3:1 randomisation ratio (planned; PtDA = 48, DC = 16). Due to recruitment difficulties, we switched to a 1:1 randomisation ratio to ensure an appropriate sample size (PtDA > 20, DC > 10). Women between 18 and 45 years old with relapsing-remitting MS or clinically isolated syndrome, and who had not yet decided about motherhood, were eligible. We recruited two nurses for our decision coaching training course. We used questionnaires to measure decisional conflict, programme feasibility, knowledge, and worries regarding pregnancy in MS. Interviews were conducted to gain in-depth information on the potential feasibility of the programmes. Interviews were analysed thematically and questionnaires descriptively. We conducted explorative group comparisons and merged findings using joint display analysis. Results We trained two decision coaches in the DC group and recruited 35 wwMS (PtDA = 22; DC = 13) over five months. Median DCS scores in the DC group were 49 at baseline and 15 at follow-up (range 0–100; higher scores indicate greater decisional conflict). In the PtDA group, scores were 54 (baseline) and 31 (follow-up). Explorative group comparison indicated lower decisional conflict at follow-up in the DC group than in the PtDA group (p = 0.035). Interviewees (PtDA = 5; DC = 6; nurses = 2) described both interventions as helpful for decision-making. Qualitative findings indicate greater satisfaction levels in the DC group. Conclusion Both interventions appear useful for wwMS in making motherhood choices. The DC programme seems more promising in supporting decision-making. Trial registration German Clinical Trials Register (DRKS); DRKS00038534.

Delays in diagnosis and treatment of depressive disorder among young adults: A national online survey-based cross-sectional study

Background Depression is highly prevalent among U.S. young adults and associated with long-term functional impairment and increased suicide risk. While delays in diagnosis and treatment of depression are well documented among older adults, the magnitude and predictors of such delays in the younger population are poorly understood. Objective To characterize the time to diagnosis and treatment of depressive disorder and predictors of diagnostic and treatment delay among young adults. Methods This cross-sectional study used a self-reported survey conducted via the online research platform Prolific in May 2025. Eligible participants were U.S. adults aged 18–35 years with a history of at least one depressive episode. Sociodemographic, clinical, and psychosocial characteristics, including age of depressive symptom onset, degree of social support, and frequency of social group engagement, were assessed. Primary outcomes were probability of not receiving a depressive disorder diagnosis despite symptoms, time from symptom onset to diagnosis, probability of not seeking treatment, and time from symptom onset to treatment. Secondary outcomes were perceived treatment effectiveness and current symptom control. Results In total, 871 respondents met inclusion criteria. Of those with one or more lifetime depressive episodes, 46.2% reported never receiving a depressive disorder diagnosis. Median time from symptom onset to diagnosis was 3 years (IQR: 0–7). Over a quarter (27.4%) never sought treatment; among those who did, 93.5% received care, but 31.4% experienced a delay of 1–4 years, and 28.8% experienced a delay of 5 + years. Symptom onset in childhood (ages 0–12) or adolescence (ages 13–17) was associated with longer time to diagnosis and treatment and lower perceived treatment effectiveness. Greater social support was associated with shorter time to diagnosis; lower probability of never receiving a diagnosis, never seeking treatment, or experiencing prolonged treatment delay; and higher perceived treatment effectiveness and current symptom control. Frequent engagement in social groups was also associated with greater perceived treatment effectiveness. Conclusions Among U.S. young adults, prolonged delays in depression diagnosis and treatment are common. Early symptom onset is associated with longer delays and worse outcomes, whereas greater social support is associated with shorter delays and more favorable outcomes. These findings highlight the need for further research to clarify causal mechanisms and for interventions to promote timely diagnosis and treatment among young adults at risk for depression.

Correction: Predicting poor functional outcomes for patients with large computed tomography perfusion core infarctions treated with endovascular thrombectomy

A reliability assessment of the basic erosive wear examination and the tooth wear evaluation system 2.0 utilizing intraoral scan data

Objectives This study assessed the reliability and clinical applicability of two tooth wear screening indices—Basic Erosive Wear Examination (BEWE) and the Tooth Wear Screening module of the Tooth Wear Evaluation System 2.0 (TWES 2.0)—using intraoral scans. Materials and methods A total of 246 anonymized intraoral scans from adult patients were independently evaluated by two calibrated examiners. Examiner calibration was performed prior to the study using a representative set of intraoral scans. Calibration was repeated until consensus regarding the application of the scoring criteria was achieved before formal data collection. Scores for all sextants were recorded for BEWE and TWES 2.0. Inter-rater agreement was primarily assessed using weighted kappa coefficients, as BEWE and TWES 2.0 are ordinal, numerically coded indices. Wilcoxon signed-rank tests were additionally used to assess systematic directional differences between paired scores. Statistical significance was set at p < 0.05. Results BEWE demonstrated good reliability, with weighted kappa values ranging from 0.760 to 0.851 across sextants, 0.868 for the total BEWE score, and an overall weighted kappa of 0.841. TWES 2.0 showed moderate to good reliability, with weighted kappa values ranging from 0.543 to 0.761 across sextants and an overall weighted kappa of 0.715. Conclusions Both BEWE and TWES 2.0 are reliable and practical for screening noncarious tooth wear via intraoral scans. BEWE showed slightly higher inter-rater consistency, whereas TWES 2.0 allows more detailed evaluation of occlusal and palatal surfaces. These indices can support standardized monitoring, early detection, and clinical management of tooth wear. Examiner calibration remains essential, particularly for TWES 2.0. Clinical trial registration This was a retrospective reliability study and did not involve an interventional clinical trial; therefore, registration was not applicable.

Cost-effectiveness analysis of mammography screening for early detection of breast cancer in Nigeria

Mammography still remains the gold standard for breast cancer screening, considering its impact on breast cancer mortality. However, it has a relatively low utilization rate in Nigeria. Although the National Strategic Cancer Control Plan (NSCCP) has a goal of making screening services and early detection of cancer available for all Nigerians, there is currently no national breast cancer screening program implemented in Nigeria. The modelling study aimed to evaluate the cost-effectiveness of mammography screening from the healthcare provider’s perspective and to determine the appropriate screening interval for Nigerian women, aiming to enhance the efficiency and effectiveness of breast cancer detection programs. A state-transition Markov model was adapted to simulate annual and biennial mammography, breast cancer diagnosis, and treatment in a cohort of cancer-free Nigerian women aged 40 years and followed them for a lifetime. The study was conducted from the healthcare provider’s perspective. Disability-adjusted life year (DALY) averted, representing the health outcomes, was used to estimate the incremental cost-effectiveness ratio (ICER). Costs and outcomes were discounted at an annual rate of 5%. Annual mammography screening costs US$238.60, averted a DALY of 1.060, and was the most cost-effective intervention with an ICER of US$207.24 (95% CI US$213.31 – US$216.88)/DALY averted, which was below the willingness-to-pay threshold of $1074. Mammography screening strategies were estimated to be cost-effective from the healthcare payer’s perspective under the model assumptions. Annual screening showed the most favorable cost-effectiveness profile among the strategies evaluated, but this finding is model-dependent and should be interpreted as comparative economic evidence rather than a definitive screening recommendation. These results can inform future research, policy discussions, and consideration of sustainable financing for breast cancer screening in Nigeria.

Gender differences in the relationship between adult attachment and self-identity: A network analysis research among Chinese college students

Objective This study aims to investigate the relationship between adult attachment and self-identity among Chinese college students using network analysis, with a specific focus on examining gender differences in network structure, global strength. and edge strength. Methods A convenience sampling method was employed, and a total of 624 university students from China were surveyed using the Experiences in Close Relationships Inventory and the Self-Identity Questionnaire developed by Chinese scholars. Network analysis was conducted to estimate the structure of adult attachment and self-identity, identify central and bridge symptoms, and examine gender differences in network structure, global strength, and edge strength. Results The self-identity items formed a tightly connected cluster. Temporal disintegration constituted the core node bridging adult attachment and self-identity networks, followed by identity diffusion. Attachment anxiety was the strongest bridge node connecting adult attachment to self-identity, primarily associating with temporal disintegration. Network invariance test indicated no significant gender differences. Global strength invariance test was significantly higher in females than in males. Edge strength invariance test revealed that there were significant differences in the strength of some edges between males and females. Conclusion In the adult attachment and self-identity network, temporal disintegration and identity diffusion serve as core nodes, around which close clusters form . Attachment anxiety serving as the key bridge. Although the network structure is similar across genders, females show significantly stronger overall network connectivity. These findings highlight the importance of examining both local and global metrics when studying group differences in psychological networks.

Research on fine-tuning algorithms for Large Language Models integrating Uncertainty Modeling and External Memory Augmentation

This paper proposes a parameter-efficient fine-tuning framework that integrates uncertainty modeling with external memory augmentation, aiming to improve robustness, confidence calibration, and contextual completeness in downstream natural language processing tasks. From the methodological perspective, the uncertainty modeling module explicitly characterizes uncertainty in inputs and intermediate representations through feature-level estimation, cross-layer propagation, and confidence calibration, thereby enhancing training stability and reducing the influence of noisy signals. Meanwhile, the external memory augmentation module employs key-value retrieval and gated fusion mechanisms to provide reusable contextual support, alleviating information loss caused by limited contextual summarization and improving representation quality under heterogeneous evaluation settings. Extensive experiments and ablation studies were conducted on text classification and named entity recognition tasks across multiple public benchmark datasets, using GPT-2 Small, GPT-2 Medium, and LLaMA3-8B as backbone models. The results demonstrate that the proposed framework consistently outperforms several mainstream fine-tuning methods in terms of accuracy, F1 score, and robustness, while also showing stable behavior under learning-rate sensitivity and missing-information settings. Overall, this study provides a novel perspective for efficient and interpretable fine-tuning paradigms, achieving a favorable balance among performance improvement, parameter efficiency, and deployment feasibility, and offering a practical basis for future extensions to more complex downstream scenarios.

Quercetin suppresses the progression of HBV-associated hepatocellular carcinoma by modulating the EGFR signaling pathway

Background Quercetin, a bioactive flavonoid compound widely present in medicinal and edible plants, has demonstrated therapeutic potential against hepatocellular carcinoma (HCC). However, its specific mechanism in hepatitis B virus-associated hepatocellular carcinoma (HBV-HCC) remains unclear. This study aims to systematically elucidate the efficacy and molecular mechanisms of quercetin against HBV-HCC. Methods Integrated in vitro and in vivo experimental models were employed. The inhibitory effects of quercetin on HCC cell viability, proliferation, clonogenicity, and migration were assessed through CCK-8, EdU, colony formation, and scratch assays. Network pharmacology and molecular docking were integrated to identify potential targets of quercetin within HCC. Lentiviral transfection was used to construct HCC cell lines overexpressing HBx and EGFR, with key signaling pathways verified via Western blot. Additionally, a xenograft mouse model was established, and the EGFR inhibitor osimertinib was combined to evaluate quercetin’s therapeutic efficacy and underlying mechanisms in HBV-HCC. Results Through an extensive analysis of the target interaction network analysis of quercetin in HCC, this study identified and prioritized 29 potential therapeutic targets, with the EGFR recognized as the principal target molecule. The results of molecular docking experiments indicated that both EGFR and GSK3β exhibited good binding affinity. Subsequent in vitro studies revealed that quercetin substantially suppresses the growth and migration of HBV-HCC cells. It achieves this by dose-dependently suppressing EGFR, thereby attenuating the signaling of the downstream PI3K/AKT/GSK3β axis and concurrently reversing the epithelial-mesenchymal transition (EMT) process. In vivo investigations, complemented by control studies using EGFR inhibitors, further validate that quercetin exerts its anti-tumor effects against HBV-HCC through specific targeting of EGFR and the suppression of the EMT program. Conclusion This research validates the therapeutic effectiveness of quercetin in inhibiting HBV-HCC and elucidates the molecular mechanisms responsible for its action. Mechanistically, quercetin inhibits the PI3K/AKT/GSK3β signaling axis by targeting EGFR, thereby reversing the EMT process and ultimately impeding HBV-HCC progression. These critical outcomes provide a novel theoretical foundation for the targeted therapy of HBV-HCC using quercetin.

Comparing the uplink performance of 3D and 2D antenna models in THz networks in the presence of joint human and wall blockages

Terahertz (THz) communication is considered as a key technology enabler for realizing Sixth Generation (6G) network. THz band communication offers several promising advantages, but numerous challenges are expected due to the inherent limitations of propagation at THz frequencies in the 6G network, such as path loss, interference, human and wall blockages, etc. In retrospect, THz band communication finds its use in indoor network deployments. In this paper, a framework is developed to analyze the impact of the uplink performance of a single-tier THz network, incorporating the impact of wall and human blockages in the indoor environment. To model a practical system, 3D antenna model have been employed, which accounts for both horizontal and vertical radiation patterns, whose performance have been benchmarked against 2D antenna model that accounts only for horizontal direction. This evaluation has enabled us to highlight the impact of practical antenna models on THz communication performance. Using the developed system model, generalized expressions for uplink mean interference, uplink coverage probability, and area spectral efficiency have been derived. The impact of THz uplink network performance has been analyzed using an antenna model with varying user equipment heights and different main lobe beam widths, as well as considering different path loss exponents for Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) conditions. The analytical results obtained against different network conditions have been compared and validated against Monte Carlo simulations and both have been found in agreement.

Representing national images of self and others through China’s diplomatic discourse: A corpus-based study

Drawing on van Dijk’s Ideological Square framework, this paper adopts a corpus-based method to examine the discursive strategies in their responses by the spokespersons for China’s Ministry of Foreign Affairs during regular press conferences amid a public health crisis. The analysis focuses on how these discursive strategies shape the national images of China and the other four permanent members of the United Nations Security Council. The results show that (1) the spokespersons actively employed communicative discursive strategy to clarify China’s stance and international cooperation initiatives while also using offensive discourse strategy to counter criticisms from US-led Western nations and media regarding the virus and the pandemic; (2) although the spokespersons’ discourse generally aligns with van Dijk’s Ideological Square of positive self-presentation and negative other-presentation, this model is not fixed but subject to dynamic changes driven by the self-serving principle. It is argued that factors such as diplomatic ideology, geopolitical relations, and traditional Chinese culture underlie the spokespersons’ use of discursive strategies and national images representations. This study contributes to reconceptualizing an existing discourse model by offering data-driven insights into the operational mechanisms of ideological discourse in the contexts of global political communication and national image construction.

Joint optimization of task offloading and energy trading in edge-enabled smart grids using deep reinforcement learning

The proliferation of distributed energy resources (DERs) and the ubiquity of Internet of Things (IoT) devices are driving the integration of mobile edge computing (MEC) into smart grids. This convergence enables real-time data processing for prosumers but introduces a complex cyber-physical coupling: computational offloading decisions directly impact local energy consumption, thereby altering the prosumer’s status in the peer-to-peer (P2P) energy market. Conversely, dynamic market prices influence the economic viability of offloading. This paper addresses the joint optimization of computational task offloading and P2P energy trading in an edge-assisted smart grid ecosystem. We formulate the problem as a mixed-integer nonlinear programming (MINLP) model aimed at maximizing long-term system utility, balancing throughput, latency, and economic incentives under strict edge server capacity and community energy neutrality constraints. To tackle the curse of dimensionality and system stochasticity, we propose a hybrid framework combining Deep Q-Networks (DQN) with a constraint-aware heuristic mechanism. The DQN agent learns adaptive offloading policies from high-dimensional states, while a deterministic rule-based layer ensures strict adherence to community energy balance. Simulation results based on real-world solar generation and market data demonstrate that our proposed method outperforms baseline strategies—including local-only execution and greedy heuristics—improving average utility by 12.3% and reducing task delay by 16.5%, while maintaining robust operational feasibility.

Correction: A new criterion for defining tunnel portal failure using the strength reduction method

Development and validation of search hedges for Transgender and Gender Diverse (TGD) populations in Ovid MEDLINE and Ovid APA PsycInfo

Introduction This paper describes the development and validation of highly sensitive search hedges for Ovid MEDLINE and Ovid APA PsycInfo that effectively identify literature on transgender and gender diverse (TGD) populations. Methods Two librarians developed the search hedges using relevant keywords and controlled vocabulary terms, building on previous work on identifying transgender populations in evidence synthesis. The hedges were tested and refined to capture diverse and expansive gender identities across cultures and disciplines. The hedges were validated for sensitivity using a gold standard set of 144 articles from the Knowsy portal of evidence syntheses tagged as Two-Spirit, transgender, or gender non-binary. To assess precision an international research team of subject experts independently screened a randomized sample of search results in a two-stage screening process with an additional screener resolving disputes. Results The final search hedges demonstrated 100% sensitivity in both MEDLINE and APA PsycInfo, identifying all 144 relevant articles from the Knowsy gold standard set. The MEDLINE search hedge achieved a 71% precision, and the APA PsycInfo hedge achieved a 67% precision. These results balance comprehensive retrieval while minimizing non-relevant articles for an efficient screening process. Conclusions These search hedges in MEDLINE and APA PsycInfo are valuable tools for researchers and librarians to more effectively identify literature on TGD populations. These tools will be crucial for ongoing work in addressing gaps in research and health disparities faced by TGD populations and will be particularly valuable for researchers conducting evidence synthesis projects related to this population.

Real-time detection of rare roadside obstacles using YOLOv8-n in autonomous vehicles

Rare road obstacles, including traffic cones, fallen trees, debris, barrels, and rocks, pose significant safety risks to autonomous vehicles. This paper presents a lightweight real-time detection framework using YOLOv8-n to accurately identify such obstacles on resource-constrained hardware. Multiple open source datasets containing annotated images of rare objects were combined and curated into a unified dataset. The model was refined using transfer learning, and its resilience to changing illumination and partial occlusion was enhanced by data augmentation techniques such brightness fluctuation, rotation, flipping, and geometric distortion. On a mid-range NVIDIA P100 GPU, the model maintained an inference speed of 68 frames per second while achieving a precision of 95.4%, recall of 93.9%, F1-score of 94.6%, and mean average precision (mAP@0.5) of 98.1%. These findings show that the framework is appropriate for edge-based autonomous driving systems where low latency and computational efficiency are crucial since it provides precise real-time detection without the need for expensive hardware.

Unimodal vs. multimodal deep learning for non-invasive MGMT promoter methylation prediction in glioblastoma: A systematic evaluation on the BraTS 2021 dataset

Glioblastoma multiforme (GBM) is the most aggressive primary brain tumor in adults, with a median survival of 14.6 months under standard radiotherapy and temozolomide (TMZ) chemotherapy. The methylation status of the O⁶-methylguanine-DNA methyltransferase (MGMT) promoter is a critical biomarker predicting TMZ response; however, its determination currently requires invasive tissue sampling. Non-invasive prediction of MGMT promoter methylation from multiparametric MRI (mpMRI) through deep learning represents a compelling alternative, yet its clinical feasibility remains unresolved. Using the BraTS 2021 dataset (582 patients, four MRI sequences: FLAIR, T1w, T1wCE, T2w), we conducted a systematic comparative study of unimodal and multimodal deep learning approaches based on VGG-16, exploring 1,380 experimental configurations (unimodal: 192; multimodal: 1,188) across three imaging planes, eight slice counts, and three multimodal fusion strategies (early, intermediate, and late fusion). In the unimodal setting, the best model trained on T2w coronal images (32 slices, no transfer learning) achieved an accuracy of 0.6458 and an AUC of 0.6422 on the validation set, but dropped to 0.5586 and 0.5533 on the independent test set, revealing substantial overfitting attributable to limited dataset size. Strikingly, multimodal fusion consistently failed to outperform the best unimodal model, with all three fusion strategies plateauing at ~0.64 accuracy and ~0.64 AUC on validation data. Transfer learning improved generalization across train/test distributions at the cost of peak performance. These findings suggest, for the tested framework in this study, that MGMT methylation status prediction from mpMRI remains fundamentally constrained by dataset heterogeneity and size, irrespective of modality combination strategy, and that T2w coronal acquisitions could be more interesting in future data collection efforts.