Search Results (16)

The automated detection of deep-sea polymetallic nodules is critical for processing large volumes of benthic imagery. However, its scalability faces challenges from cross-expedition covariate shifts, such as changes in lighting, altitude, and camera payloads, which lower zero-shot model performance. While semi-supervised pseudo-labelling presents a potential alternative to time-consuming re-annotation, simple implementations can quickly lead to confirmation bias. This study identifies two primary sources of this degradation: spatial noise from tiling fragmentation at tile borders and an architecture-agnostic interior false positive floor caused by semantic domain shift. This work proposes using a multi-model ensemble for pseudo-labelling to reduce the noise impact. Using a spatial border filter and confidence stratification, three architecturally distinct teacher models (YOLOv8, Faster R-CNN, and DINO) are employed to determine a reliable and domain-invariant subspace. Under a strict anti-leakage Leave-One-Partition-Out protocol, the proposed approach surpasses the supervised fine-tuning baseline at 100-tile pseudo-label budget across four random seeds (macro mAP_50:95 of $0.4745\pm 0.0042$ versus $0.4467\pm 0.0079$), with gains concentrated in the most domain-shifted fold. Beyond this budget, our findings highlight two important adaptation trends: a pool-size degradation trend where excessive pseudo-label volume actively degrades generalisation, and the observation that the fine-tuned models reduce pseudo-label fidelity despite higher precision, providing evidence for the advantage of using frozen source checkpoints for cross-domain adaptation. Full article

A nondestructive, rapid, and portable detection method for moisture content (MC) in living tree trunks remains unavailable. Tree radar, developed based on ground-penetrating radar (GPR) technology, represents a promising approach for tree trunk MC detection owing to its high penetration depth and low susceptibility to environmental interference. However, its application to living tree MC detection is constrained by curvature-induced wave propagation complexity, interspecific structural heterogeneity and the limited availability of labeled MC samples obtained through destructive coring, collectively resulting in poor model performance. The study proposed a novel GPR-based MC detection method employing a multi-scale one-dimensional convolutional neural network integrated with an attention mechanism and mixed data augmentation (mixed-MS1DCNNAM). GPR amplitude data extracted from the first 6.5 ns of B-scan signals were used to capture MC-related features via a custom program developed in MATGPR. A mixed model for four tree species with 15–30 cm diameters at breast height (DBH) achieved an R² of 0.7908 and an RMSE value of 0.1059, outperforming traditional models, with test metrics calculated at the tree level by averaging predictions from five directional GPR scans per tree. Furthermore, three DBH-specific sub-models (15–20 cm, 20–25 cm, and 25–30 cm) and four single-species sub-models were developed, yielding improved performance (R² ≥ 0.7246, RMSE ≤ 0.1033; RMSE ≤ 0.0959, MAE ≤ 0.0626, except for European white birch). These results highlighted the effectiveness of stratification by DBH class and tree species. Overall, this study effectively addresses aforementioned challenges and establishes a generalizable nondestructive approach for living trees under field conditions, facilitating sustainable forest management in tree growth monitoring, forest disaster monitoring, harvested timber storage and wood quality assessment. Full article

Background/Objectives: Population aging necessitates a shift from disease-focused paradigms to a holistic characterization of biological aging processes. While chronological age remains the primary metric, it poorly captures inter-individual variability in physiological resilience and health trajectories. This study aimed to identify robust, multidimensional aging phenotypes independent of chronological age and sex using integrative factor analysis of heterogeneous biomedical data from a Russian cohort—a population underrepresented in aging research. Methods: We analyzed data from 1201 conditionally healthy adults (aged 18–99 years) enrolled in the RUSS AGE study. A comprehensive dataset comprising 118 variables across 11 modalities—including biochemical markers, anthropometry, physical function, cognitive-emotional assessments, lifestyle factors, and psychosocial indicators—was integrated using Multi-Omics Factor Analysis v2 (MOFA2). Following the extraction of 16 latent factors and residualization for demographic confounders, consensus clustering was performed to identify distinct aging phenotypes. Phenotype stability was internally recapitulated using gradient-boosting classifiers (XGBoost, CatBoost) in a stratified five-fold cross-validation and on a held-out test set. Results: MOFA2 identified 16 stable latent factors, explaining 21.3% of the total variance and capturing coordinated variation across metabolic, inflammatory, cardiovascular, cognitive, and behavioral domains. Consensus clustering revealed five reproducible phenotypes—Anemic (n = 82), Metabolically Subcompensated (n = 99), Metabolically Decompensated (n = 304), Overloaded (n = 302), and Balanced (n = 414)—characterized by distinct multisystem profiles independent of age (p > 0.05 after FDR correction) and sex. Supervised classification achieved high discriminative performance (macro F1-score = 0.75, OvR ROC-AUC = 0.93 on the held-out test set), quantifying the internal reconstructability of the phenotype labels from the original feature space rather than external generalization to an independent cohort. Conclusions: This study demonstrates the feasibility of data-driven, biologically coherent phenotyping of healthy aging using integrative factor analysis. The identified phenotypes represent stable configurations of physiological, functional, and psychosocial characteristics that transcend chronological age, providing a foundation for the future development of risk-stratification tools, preventive interventions, and biological-age calculators, subject to subsequent validation in longitudinal and independent external cohorts. Full article

Accurate distinction between non-muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC) remains the key local staging problem in bladder cancer because treatment intensity, timing of radical therapy, and suitability for bladder-preserving strategies all depend on it. Multiparametric magnetic resonance imaging (mpMRI) and the Vesical Imaging-Reporting and Data System (VI-RADS) now provide a standardized imaging framework for local staging and increasingly support MRI-first clinical pathways. Artificial intelligence (AI) has emerged as an additional decision-support layer, but the evidence base remains methodologically uneven. In this structured narrative review, we synthesized peer-reviewed literature from January 2020 to March 2026, while retaining foundational VI-RADS studies from 2018 to 2019, and prioritized guideline documents, meta-analyses, prospective cohorts, multicenter and externally validated AI studies, response-assessment studies, and papers addressing implementation and reporting quality. Current evidence shows that radiomics and deep learning models can achieve high discrimination for MIBC detection on MRI, and that the most plausible incremental value of AI lies in equivocal VI-RADS lesions, reader support outside high-volume expert settings, and multimodal risk stratification. However, most studies remain retrospective, highly selected, segmentation-dependent, and vulnerable to reference-standard bias, domain shift, and poor calibration. This review therefore emphasizes several translational issues that are often underreported: lesion-level versus patient-level inference, the distortive effect of TURBT-based labels, the need to evaluate false-negative consequences in VI-RADS 3 tumors, and the distinction between diagnostic support and broader pathway redesign. We also discuss response assessment, nacVI-RADS, segmentation automation, multicenter and federated infrastructure, workflow ownership, and the limits of imaging-only models in a biologically heterogeneous disease. The most credible near-term role of AI is not autonomous diagnosis, but augmentation of standardized mpMRI and VI-RADS within multidisciplinary care. Future progress will depend on prospective utility studies, site-held-out validation, transparent reporting, and the integration of imaging with molecular and cellular heterogeneity through radiogenomic and multi-omics approaches. Full article

Recent advances in deep learning (DL) have enabled the integration of diverse biomedical data for disease prediction and risk stratification. Building on this progress, the overall objective of this study was to develop and evaluate a multimodal DL framework for robust multi-label classification (MLC) of major comorbidities in patients with obstructive sleep apnea (OSA) using physiological time series signals and clinical data. This study proposes a robust framework for multi-label classification (MLC) of comorbidities in patients with OSA using diverse physiological and clinical data sources. We conducted a retrospective observational study including a convenience sample of 144 patients referred for overnight polysomnography at the Sleep Medicine Center (SleepLab Split), University Hospital Centre Split (KBC Split), Split, Croatia. Patients were selected based on predefined inclusion criteria and data availability. A one-dimensional Convolutional Neural Network (1D-CNN) was developed to process and fuse time series signals, oxygen saturation ($Sp{O}_2$), derived $Sp{O}_2$ features, and nasal airflow (FP0), with demographic and physiological parameters, enabling the identification of key comorbidities such as arterial hypertension, diabetes mellitus, and asthma/COPD. The instruments included polysomnography-derived signals (SpO₂ and FP0 airflow) and structured demographic/physiological parameters. Signals were preprocessed and used as inputs to the proposed fusion model. The proposed model was trained and fine-tuned using the Optuna hyperparameter optimization framework, addressing class imbalance through weighted loss adjustments. Its performance was comprehensively assessed using multi-label evaluation metrics, including macro/micro F1-score, AUC-ROC, AUC-PR, subset and partial accuracy, Hamming loss, and multi-label confusion matrix (MLCM). The study protocol was approved by the Ethics Committee of the School of Medicine, University of Split (Approval No. 003-08/23-03/0015, Date: 17 October 2023). The 1D-CNN achieved superior predictive performance compared to traditional machine learning (ML) classifiers with macro AUC-ROC = 0.731 and AUC-PR = 0.750. The model demonstrated consistent behavior across age, gender, and BMI groups, indicating strong generalization and minimal demographic bias. In conclusion, the results confirm that $Sp{O}_2$ and airflow signals inherently encode comorbidity-specific physiological patterns, enabling efficient and scalable screening of OSA-related comorbidities without the need for full polysomnography. Although the study is limited by data set size, it provides a methodological basis for the application of multi-label DL models in clinical decision support systems. Future research should focus on the expansion of multi-center datasets, thereby improving model interpretability and potential clinical adoption. Full article

Accurate credit risk assessment for small firms and farmers is crucial for financial stability and inclusion; however, many models still rely on binary default labels, overlooking the continuum of borrower vulnerability. To address this, we propose Transformer–LightGBM–Stacked Logistic Regression (TL-StackLR), a hybrid stacking framework for multi-class loan default prediction. The framework combines three learners: a Feature Tokenizer Transformer (FT-Transformer) for feature interactions, LightGBM for non-linear pattern recognition, and a stacked LR meta-learner for calibrated probability fusion. We transform binary labels into three risk tiers, Low, Medium, and High, based on quantile-based stratification of default probabilities, aligning the model with real-world risk management. Evaluated on datasets from 3045 firms and 2044 farmers in China, TL-StackLR achieves state-of-the-art ROC-AUC scores of 0.986 (firms) and 0.972 (farmers), with superior calibration and discrimination across all risk classes, outperforming all standalone and partial-hybrid benchmarks. The framework provides SHapley Additive exPlanations (SHAP) interpretability, showing how key risk drivers, such as income, industry experience, and mortgage score for firms and loan purpose, Engel coefficient, and income for farmers, influence risk tiers. This transparency transforms TL-StackLR into a decision-support tool, enabling targeted interventions for inclusive lending, thus offering a practical foundation for equitable credit risk management. Full article

Background: Traumatic brain injury (TBI) remains a significant contributor to global mortality and long-term neurological disability. Accurate prognostic biomarkers are crucial for enhancing prognostic accuracy and guiding personalized clinical management. Objective: This review assesses the prognostic value of arterial spin labeling (ASL), a non-invasive MRI technique, in adult and pediatric TBI, with a focus on quantitative cerebral blood flow (CBF) and arterial transit time (ATT) measures. A comprehensive literature search was conducted across PubMed, Embase, Scopus, and IEEE databases, including observational studies and clinical trials that applied ASL techniques (pCASL, PASL, VSASL, multi-PLD) in TBI patients with functional or cognitive outcomes, with outcome assessments conducted at least 3 months post-injury. Results: ASL-derived CBF and ATT parameters demonstrate potential as prognostic indicators across both acute and chronic stages of TBI. Hypoperfusion patterns correlate with worse neurocognitive outcomes, while region-specific perfusion alterations are associated with affective symptoms. Multi-delay and velocity-selective ASL sequences enhance diagnostic sensitivity in TBI with heterogeneous perfusion dynamics. Compared to conventional perfusion imaging, ASL provides absolute quantification without contrast agents, making it suitable for repeated monitoring in vulnerable populations. ASL emerges as a promising prognostic biomarker for clinical use in TBI. Conclusion: Integrating ASL into multiparametric models may improve risk stratification and guide individualized therapeutic strategies. Full article

Background: The aim was to analyze the current state of deep learning (DL)-based prostate cancer (PCa) diagnosis with a focus on magnetic resonance (MR) prostate reconstruction; PCa detection/stratification/reconstruction; positron emission tomography/computed tomography (PET/CT); androgen deprivation therapy (ADT); prostate biopsy; associated challenges and their clinical implications. Methods: A search of the PubMed database was conducted based on the inclusion and exclusion criteria for the use of DL methods within the abovementioned areas. Results: A total of 784 articles were found, of which, 64 were included. Reconstruction of the prostate, the detection and stratification of prostate cancer, the reconstruction of prostate cancer, and diagnosis on PET/CT, ADT, and biopsy were analyzed in 21, 22, 6, 7, 2, and 6 studies, respectively. Among studies describing DL use for MR-based purposes, datasets with magnetic field power of 3 T, 1.5 T, and 3/1.5 T were used in 18/19/5, 0/1/0, and 3/2/1 studies, respectively, of 6/7 studies analyzing DL for PET/CT diagnosis which used data from a single institution. Among the radiotracers, [⁶⁸Ga]Ga-PSMA-11, [¹⁸F]DCFPyl, and [¹⁸F]PSMA-1007 were used in 5, 1, and 1 study, respectively. Only two studies that analyzed DL in the context of DT met the inclusion criteria. Both were performed with a single-institution dataset with only manual labeling of training data. Three studies, each analyzing DL for prostate biopsy, were performed with single- and multi-institutional datasets. TeUS, TRUS, and MRI were used as input modalities in two, three, and one study, respectively. Conclusion: DL models in prostate cancer diagnosis show promise but are not yet ready for clinical use due to variability in methods, labels, and evaluation criteria. Conducting additional research while acknowledging all the limitations outlined is crucial for reinforcing the utility and effectiveness of DL-based models in clinical settings. Full article

Object detection is a task that performs position identification and label classification of objects in images or videos. The information obtained through this process plays an essential role in various tasks in the field of computer vision. In object detection, the data utilized for training and validation typically originate from public datasets that are well-balanced in terms of the number of objects ascribed to each class in an image. However, in real-world scenarios, handling datasets with much greater class imbalance, i.e., very different numbers of objects for each class, is much more common, and this imbalance may reduce the performance of object detection when predicting unseen test images. In our study, thus, we propose a method that evenly distributes the classes in an image for training and validation, solving the class imbalance problem in object detection. Our proposed method aims to maintain a uniform class distribution through multi-label stratification. We tested our proposed method not only on public datasets that typically exhibit balanced class distribution but also on private datasets that may have imbalanced class distribution. We found that our proposed method was more effective on datasets containing severe imbalance and less data. Our findings indicate that the proposed method can be effectively used on datasets with substantially imbalanced class distribution. Full article

The new type of multi-temporal global land use data with multiple classes is able to provide information on both the different land covers and their temporal changes; furthermore, it is able to contribute to many applications, such as those involving global climate and Earth ecosystem analyses. However, the current accuracy assessment methods have two limitations regarding multi-temporal land cover data that have multiple classes. First, multi-temporal land cover uses data from multiple phases, which is time-consuming and inefficient if evaluated one by one. Secondly, the conversion between different land cover classes increases the complexity of the sample stratification, and the assessments with different types of land cover suffer from inefficient sample stratification. In this paper, we propose a spatiotemporal stratified sampling method for stratifying the multi-temporal GlobeLand30 products for China. The changed and unchanged types of each class of data in the three periods are used to obtain a reasonable stratification. Then, the strata labels are simplified by using binary coding, i.e., a 1 or 0 representing a specified class or a nonspecified class, to improve the efficiency of the stratification. Additionally, the stratified sample size is determined by the combination of proportional allocation and empirical evaluation. The experimental results show that spatiotemporal stratified sampling is beneficial for increasing the sample size of the “change” strata for multi-temporal data and can evaluate not only the accuracy and area of the data in a single data but also the accuracy and area of the data in a multi-period change type and an unchanged type. This work also provides a good reference for the assessment of multi-temporal data with multiple classes. Full article

In this study, transfer learning has been used to overcome multilabel classification tasks. As a case study, movie genre classification by using posters has been chosen. Six state-of-the-art pretrained models, VGG16, ResNet, DenseNet, Inception, MobileNet, and ConvNeXt, have been employed for this experiment. The movie posters have been obtained from Internet Movie Database (IMDB). The dataset has been divided using an iterative stratification technique. A sequence of dense layers has been added on top of each model and these models have been trained and fine-tuned. All the results of the models compared considered accuracy, loss, Hamming loss, F1-score, precision, and AUC metrics. When the metrics used were evaluated, the most successful result regarding accuracy has been obtained from the modified DenseNet architecture at 90%. Also, the ConvNeXt, which is the newest model among all, performed quite satisfactorily, reaching over 90% accuracy. This study uses an iterative stratification method to split an unbalanced dataset which provides more reliable results than the classical splitting method which is the common method in the literature. Also, the feature extraction capabilities of the six pretrained models have been compared. The outcome of this study shows promising results regarding multilabel classification. As for future work, it is planned to enhance this study by using natural language processing and ensemble methods. Full article

Malignant pleural mesothelioma (MPM) has limited treatment options and poor prognosis. Frequent inactivation of the tumour suppressors BAP1, NF2 and P16 may differentially sensitise tumours to treatments. We have established chick chorioallantoic membrane (CAM) xenograft models of low-passage MPM cell lines and protocols for evaluating drug responses. Ten cell lines, representing the spectrum of histological subtypes and tumour suppressor status, were dual labelled for fluorescence/bioluminescence imaging and implanted on the CAM at E7. Bioluminescence was used to assess viability of primary tumours, which were excised at E14 for immunohistological staining or real-time PCR. All MPM cell lines engrafted efficiently forming vascularised nodules, however their size, morphology and interaction with chick cells varied. MPM phenotypes including local invasion, fibroblast recruitment, tumour angiogenesis and vascular remodelling were evident. Bioluminescence imaging could be used to reliably estimate tumour burden pre- and post-treatment, correlating with tumour weight and Ki-67 staining. In conclusion, MPM-CAM models recapitulate important features of the disease and are suitable to assess drug targets using a broad range of MPM cell lines that allow histological or genetic stratification. They are amenable to multi-modal imaging, potentially offering a time and cost-efficient, 3Rs-compliant alternative to rodent xenograft models to prioritise candidate compounds from in vitro studies. Full article

Introduction: There is no standard treatment after resection of colorectal liver metastases and the role of systemic therapy remains controversial. To avoid over- or undertreatment, proper risk stratification with regard to postoperative treatment strategy is highly needed. We recently demonstrated the prognostic relevance of EMT-related (epithelial-mesenchymal transition) genes in stage II/III CRC. As EMT is a major step in CRC progression, we now aimed to analyse the prognostic relevance of EMT-related genes in stage IV CRC using the study cohort of the FIRE-3 trial, an open-label multi-centre randomised controlled phase III trial of patients with metastatic CRC. Methods: Overall and progression free survival were considered as endpoints (n = 350). To investigate the prognostic relevance of EMT-related genes on either endpoint, we compared predictive performance of different models using clinical data only to models using gene data in addition to clinical data, expecting better predictive performance if EMT-related genes have prognostic value. In addition to baseline models (Kaplan Meier (KM), (regularised) Cox), Random Survival Forest (RSF), and gradient boosted trees (GBT) were fit to the data. Repeated, nested five-fold cross-validation was used for hyperparameter optimisation and performance evaluation. Predictive performance was measured by the integrated Brier score (IBS). Results: The baseline KM model showed the best performance (OS: 0.250, PFS: 0.251). None of the other models were able to outperform the KM when using clinical data only according to the IBS scores (OS: 0.253 (Cox), 0.256 (RSF), 0.284 (GBT); PFS: 0.254 (Cox), 0.256 (RSF), 0.276 (GBT)). When adding gene data, performance of GBT improved slightly (OS: 0.262 vs. 0.284; PFS: 0.268 vs. 0.276), however, none of the models performed better than the KM baseline. Conclusion: Overall, the results suggest that the prognostic relevance of EMT-related genes may be stage-dependent and that EMT-related genes have no prognostic relevance in stage IV CRC. Full article

Background and Motivation: Cardiovascular disease (CVD) causes the highest mortality globally. With escalating healthcare costs, early non-invasive CVD risk assessment is vital. Conventional methods have shown poor performance compared to more recent and fast-evolving Artificial Intelligence (AI) methods. The proposed study reviews the three most recent paradigms for CVD risk assessment, namely multiclass, multi-label, and ensemble-based methods in (i) office-based and (ii) stress-test laboratories. Methods: A total of 265 CVD-based studies were selected using the preferred reporting items for systematic reviews and meta-analyses (PRISMA) model. Due to its popularity and recent development, the study analyzed the above three paradigms using machine learning (ML) frameworks. We review comprehensively these three methods using attributes, such as architecture, applications, pro-and-cons, scientific validation, clinical evaluation, and AI risk-of-bias (RoB) in the CVD framework. These ML techniques were then extended under mobile and cloud-based infrastructure. Findings: Most popular biomarkers used were office-based, laboratory-based, image-based phenotypes, and medication usage. Surrogate carotid scanning for coronary artery risk prediction had shown promising results. Ground truth (GT) selection for AI-based training along with scientific and clinical validation is very important for CVD stratification to avoid RoB. It was observed that the most popular classification paradigm is multiclass followed by the ensemble, and multi-label. The use of deep learning techniques in CVD risk stratification is in a very early stage of development. Mobile and cloud-based AI technologies are more likely to be the future. Conclusions: AI-based methods for CVD risk assessment are most promising and successful. Choice of GT is most vital in AI-based models to prevent the RoB. The amalgamation of image-based strategies with conventional risk factors provides the highest stability when using the three CVD paradigms in non-cloud and cloud-based frameworks. Full article

Objectives: To explore the potential of Radiomics alone and in combination with a diffusion-weighted derived quantitative parameter, namely the apparent diffusion co-efficient (ADC), using supervised classification algorithms in the prediction of outcomes and prognosis. Materials and Methods: Retrospective evaluation of the imaging was conducted for a study cohort of uterine cervical cancer, candidates for radical treatment with chemo radiation. ADC values were calculated from the darkest part of the tumor, both before (labeled preADC) and post treatment (labeled postADC) with chemo radiation. Post extraction of 851 Radiomics features and feature selection analysis—by taking the union of the features that had Pearson correlation >0.35 for recurrence, >0.49 for lymph node and >0.40 for metastasis—was performed to predict clinical outcomes. Results: The study enrolled 52 patients who presented with variable FIGO stages in the age range of 28–79 (Median = 53 years) with a median follow-up of 26.5 months (range: 7–76 months). Disease recurrence occurred in 12 patients (23%). Metastasis occurred in 15 patients (28%). A model generated with 24 radiomics features and preADC using a monotone multi-layer perceptron neural network to predict the recurrence yields an AUC of 0.80 and a Kappa value of 0.55 and shows that the addition of radiomics features to ADC values improves the statistical metrics by approximately 40% for AUC and approximately 223% for Kappa. Similarly, the neural network model for prediction of metastasis returns an AUC value of 0.84 and a Kappa value of 0.65, thus exceeding performance expectations by approximately 25% for AUC and approximately 140% for Kappa. There was a significant input of GLSZM features (SALGLE and LGLZE) and GLDM features (SDLGLE and DE) in correlation with clinical outcomes of recurrence and metastasis. Conclusions: The study is an effort to bridge the unmet need of translational predictive biomarkers in the stratification of uterine cervical cancer patients based on prognosis. Full article