Search Results (483)

Accurate stability assessment of potentially unstable rock blocks is essential for the safe construction and operation of hydropower infrastructure. This study applies a comparative limit-equilibrium workflow to a single, well-characterized sliding-type rock block (156.7 m³) bounded by three discontinuities (J1 250°/35°, J2 305°/75°, J3 215°/80°) on a hydropower station slope; discontinuity attitudes were measured with a geological compass and a terrestrial three-dimensional laser scanner, and the slope surface was reconstructed by UAV photogrammetry. A common, fully documented parameter set is used by a conventional two-dimensional method, a block-dividing limit-equilibrium method, and a three-dimensional residual-thrust method with moment equilibrium. With the site-suggested shear strengths and the lower-bound cohesion as the representative value, the block-dividing method gives factors of safety of 1.169, 1.063 and 0.903 under natural, heavy-rainfall and seismic conditions at optimal azimuths of 265.8°, 266.3° and 269.1°; the corresponding two-dimensional values are 1.037, 0.904 and 0.729, and the residual-thrust values are 1.497, 1.324 and 1.049. The block-dividing factors are 12.7–23.9% above the two-dimensional profile, whereas the residual-thrust result lies a further 16–28% higher (44–47% above the two-dimensional profile); this over-estimate is traced to the steep (75°, 80°) lateral release planes and to a mesh- and lambda-sensitive column solution, and is therefore non-conservative. A cohesion sensitivity analysis with fixed friction angle shows that the factor varies by 51–57% across the suggested cohesion interval. Under code-specified targets of 1.30/1.20/1.05, horizontal anchorage requires 366/427/566 kN versus 1150/1470/1413 kN for a perpendicular-to-slope layout, so a horizontal scheme of about 0.6 MN is adopted. Finite-element validation and field piezometric/displacement monitoring data, unavailable for this block, are identified as required future work. Full article

Breast cancer remains a major global health concern, making accurate lesion assessment essential for effective clinical decision making. Deep learning has shown promising performance in breast ultrasound analysis, yet models evaluated on data from the same source may not generalize reliably to images acquired using different scanners and acquisition settings. This study therefore examines cross-dataset generalization and the factors that can improve it. Five lightweight YOLO instance-segmentation architectures (YOLOv8n, YOLO11n, YOLO11s, YOLO26n, and YOLO26s) were trained using a patient-grouped BUS-BRA protocol and a single training seed, and evaluated internally on held-out data and externally on BUS-UCLM, which served as the single target dataset. Ultrasound-specific preprocessing, test-time augmentation (TTA), and Optuna-selected configurations were assessed across 40 paired internal–external comparisons. Cross-dataset evaluation demonstrated consistent lesion delineation on BUS-UCLM, with matched Dice ranging from 0.861 to 0.875 and matched IoU from 0.764 to 0.786 across the five architectures. Preprocessing improved mask mAP@50–95 across all ten checkpoints on both datasets, while TTA improved detection-adjusted Dice despite having little effect on mAP@50–95. Optuna tuning improved external mAP@50–95 across all five architectures despite limited internal gains. Grad-CAM++ showed predominantly lesion-centered attention across both datasets, while ONNX Runtime deployment achieved 5.65–13.62 FPS on CPU. These findings highlight the importance of external validation, detection-aware evaluation, and efficient deployment for reliable breast ultrasound segmentation. Full article

Whole-slide image (WSI) compression is a fundamental requirement in digital pathology. Yet, current validation metrics, such as PSNR and pathologist agreement, were developed before the emergence of pathology foundation models (PFMs). PFMs capture fine-grained tissue details in high-dimensional spaces that can be disrupted by compression artifacts invisible to the human eye, potentially affecting downstream tasks. To the best of our knowledge, we present a systematic study of model- and dataset-specific embedding-drift indicators using representation drift ($\Delta E$), a cosine-based similarity measure, across clinically motivated compression ratios. We evaluate three vision transformer models of identical ViT-Large architecture but different training data: DINOv2 (general vision, 142 million natural images), UNI (pathology, over 100,000 clinical WSIs), and Phikon-v2 (a second PFM). We use 4000 TCGA-BRCA H&E tumor tiles from two WSIs at six compression ratios, ranging from lossless to 80:1. The results demonstrate that UNI first exceeds the predefined $\Delta E>0.01$ sentinel criterion at the tested CR 10:1 operating point (PSNR $=37.76$ dB, usually considered excellent). DINOv2 first exceeds the same sentinel criterion at CR 20:1. At CR 20:1, UNI’s $\Delta E=0.109$ while DINOv2’s is $0.012$, a nearly tenfold difference at the same image quality. At CR 80:1, UNI’s cosine similarity drops to 0.316 while DINOv2 retains 0.895. Phikon-v2 first exceeds the same sentinel criterion at CR 10:1 and tracks UNI far more closely than DINOv2 ($\Delta E=0.311$ versus $0.037$ at CR 40:1). The two pathology-pretrained models exhibit substantially greater drift than DINOv2, a pattern consistent with an association between pathology-domain training and increased compression sensitivity. However, checkpoint-specific factors such as training objectives, preprocessing, learned invariances, embedding normalization, and training data may also contribute. Additional tests on a balanced 500-tile TCGA-LUAD pilot set, denser compression ratios, and an alternative JPEG2000 encoder show similar model-dependent drift patterns; however, the empirical drift values and operating criteria should be interpreted as model- and dataset-specific rather than as generalizable thresholds across tissues, institutions, scanners, or staining protocols. We introduce the Rate Distortion Representation (RDR) curve as a model-aware evaluation tool that reveals this representation blind spot: a compression range where PSNR remains conventionally acceptable while embedding geometry is substantially altered. These results are model- and dataset-specific and should not be interpreted as universal compression thresholds or clinical safety boundaries. Overall, the findings support the use of representation-level robustness assessment as a complement to image-level fidelity metrics in AI-oriented digital pathology workflows. Full article

Background/Objectives: Rubidium-82 (⁸²Rb) PET myocardial perfusion imaging (MPI) yields, in a single study, quantitative absolute myocardial blood flow (MBF, mL/min/g), myocardial flow reserve (MFR), and gated left ventricular (LV) function (LVEF, EDV, ESV, SV). These quantitative values depend on the tracer, scanner, kinetic model, software, and underlying population and have been characterised almost exclusively in North American and European cohorts. The Arabian Gulf, where Kuwait has among the highest age-standardised diabetes prevalence worldwide (25.6%), is essentially unstudied, so the distribution and behaviour of these parameters in such a real-world cardiometabolic population are unknown. To evaluate the distribution of the quantitative ⁸²Rb PET parameter set in a real-world Arabian Gulf cohort and, within a small clinically defined normal subgroup, to describe sex- and age-related patterns in absolute MBF, MFR, and LV function. Given the limited size of the normal subgroup, these values are presented as exploratory, hypothesis-generating observations rather than definitive population reference norms. Methods: Retrospective single-centre study of 330 consecutive ⁸²Rb PET/CT studies (analytic cohort n = 292 after exclusion of repeat studies and those with incomplete quantitative output; mean age 63.1 ± 12.1 years) who underwent adenosine-stress ⁸²Rb PET/CT MPI. The clinically defined normal subgroup, normal perfusion (C1), normal global MFR (≥2.0), normal resting LVEF, and no documented cardiac history, comprised 44 patients (25 female, 19 male). Reference values are reported as sex- and age-stratified centiles (5th, 25th, median, 95th percentiles), with the 5th percentile reported for each parameter. Sex differences used Mann–Whitney U with rank-biserial r and Cohen’s d (95% CI); age was examined across broad bands. All analyses followed APA 7 standards with Bonferroni correction. Results: In the clinically defined normal subgroup, median global stress MBF was 2.94 mL/min/g (5th percentile 2.03) and median global MFR was 2.72 (5th percentile 2.06); every value satisfied the C1 definition (MFR ≥ 2.0) by construction, so these limits describe the preselected subgroup and cannot independently validate the 2.0 threshold. Women had higher resting MBF than men (median 1.07 vs. 0.90 mL/min/g; p = 0.007), with numerically lower global MFR that did not survive correction for multiple comparisons (2.57 vs. 2.98; p = 0.022); stress MBF did not differ by sex (3.00 vs. 2.91; p = 0.522). LV volumes were smaller in women, while LVEF was similar between sexes. The 5th percentile for stress LVEF was 54% (women) and 53% (men). These absolute-flow reference values are lower than those reported for Western low-risk cohorts (stress MBF ~3.25 mL/min/g; MFR ~3.18), consistent with the higher cardiometabolic burden of this population. Conclusions: In this exploratory single-centre evaluation, quantitative ⁸²Rb PET flow values in a small clinically defined normal Arabian Gulf subgroup were lower than Caucasian-derived values, while sex-related differences in resting flow were evident and age-related differences were numerically consistent with previously reported trends. Because the normal subgroup is small, these findings are hypothesis-generating and require confirmation in larger, prospectively screened cohorts before use as population reference values; they nonetheless indicate that population- and pipeline-specific calibration is needed when quantitative ⁸²Rb thresholds derived elsewhere are applied locally. Full article

The Sinian Dengying Formation reservoirs in the Anyue Gas Field, Sichuan Basin, exhibit extreme heterogeneity. The existing karst geomorphological classification is too coarse to account for the significant differences in high-quality reservoir scale and productivity observed among adjacent wells within the same third- to fourth-order karst geomorphic units. To accurately predict the distribution of high-quality reservoirs, it is essential to reveal the precise mechanisms by which low-order micro-geomorphology controls paleokarstification and reservoir development. Based on 3D seismic data, core samples, thin sections, Formation MicroScanner (FMI) images, and geochemical analysis of data from the Gaoshiti–Moxi area, this study investigates how various micro-geomorphic units control karstification from both macroscopic and microscopic perspectives and subsequently analyzes their reservoir-controlling mechanisms. The results indicate the following: (1) The paleogeomorphology at the top of the Dengying Formation was restored using the impression method with a decompaction correction. A classification standard for micro-geomorphology at the top of the fourth member of the Dengying Formation (Z₂dn⁴) dolostone was established, identifying ten micro-geomorphic units, including coastal plains, river channels, and terraces. (2) Micro-geomorphic morphology dictates the distribution of the paleokarst hydrodynamic field, thereby controlling the intensity of karstification. River channels, terraces, and coastal plains, characterized by strong hydrodynamics, are the most favorable micro-geomorphic units. Inclined plains, with moderate hydrodynamics, are secondary favorable units. Conversely, concave plains and karst hills, with weak hydrodynamics and slow water alternation, are unfavorable units. (3) Six models of high-quality reservoir development, dominated by micro-geomorphic units in both intra-platform and platform-margin settings, were constructed, emphasizing the multi-element coupling of “strong karstification–40%–60% microfacies–faults”. This study reveals that micro-geomorphology is the key factor controlling the development of Z₂dn⁴ karst reservoirs. Predicting sweet spots by superimposing sedimentary facies and fault distributions onto micro-geomorphic maps can effectively improve the drilling success rate and development efficiency of deep carbonate reservoirs. Full article

Bluetooth Low Energy (BLE) resolvable private addresses (RPAs) reduce passive tracking, but RPA-resolving creates receiver-side work: a scanner whose bonded identity set exceeds controller resolving-list capacity which can be forced into repeated host-side Identity Resolving Key (IRK) searches by floods of syntactically valid unknown private-address candidates. We formulate this as capacity-constrained resolving scheduling and propose P3, our persistence-aware admission-control scheduler for BLE RPA resolution, evaluated in its main form P3-Persist. After cache and resolving-list fast paths miss and the unresolved-work budget is exhausted, P3 grants a small reserve only to over-the-air RPA values that reappear. This uses visible repetition rather than an identity-correlated pre-resolution label. In paired simulations, identical visible traces produced identical initial pre-resolution decisions, but later modeled delay/defer features were distinguishable in some tested conditions; accordingly, no end-to-end side-channel-free claim is made. In the evaluated 20-seed, 1800 s simulation matrix (N = 512, RL = 8), P3-Persist keeps modeled unique-flood work in the same bounded approximately 45k AES-equivalent/min regime as budget-only limiting while increasing modeled legitimate resolution to 0.9569 under medium flood and 0.9557 under heavy flood, compared to 0.57–0.62 for budget-only baselines. A 20-seed epoch–loss–density experiment shows that service recovery is conditional rather than universal, and a denser adaptive-address search identifies an overlapping near-cap region around m = 320 and r = 1, with a highest confirmed mean of 146,196.29 AES-equivalent/min (96.69% of the modeled cap). These results establish bounded-work and conditional service-recovery behavior within the simulator; they do not establish production-firmware effectiveness, end-to-end side-channel freedom, on-device timing, energy, current, latency, or queueing behavior. Full article

Lumbar foraminal stenosis grading contributes to surgical-level selection, but automated four-grade classification remains challenging. Published pipelines for this dataset reach approximately 65% four-class accuracy, and deep classifiers offer no anatomical rationale. We investigated whether interpretable, millimeter-scale morphometry from segmentation masks improves grading beyond a deep image model. We analyzed the LSS-MRI-AISSLab sagittal T2-weighted dataset (469 patients, 2979 expert-graded foramina spanning L1-L2 through L5-S1 bilaterally on a four-grade scale). Ten morphometric descriptors were computed from mid-sagittal polygon segmentations and scaled to millimeters using each patient’s recorded pixel spacing. A dual-branch network combined a fine-tuned ResNet-18 embedding of each foraminal region of interest with the morphometric vector through an ordinal regression head. Foraminal regions were supplied from expert bounding-box annotations; automated localization within the full sagittal examination was not evaluated. Four configurations (nominal softmax, appearance-only, anatomy-only, and fusion) were compared on a locked patient-level test set of 94 patients after five-fold cross-validation, with quadratic weighted kappa (QWK) as the primary endpoint and patient-clustered bootstrap inference. Feature-grade correlations were reported pooled and adjusted for lumbar level. Fusion achieved QWK 0.813 (95% confidence interval [CI] 0.769–0.847) and 75.3% four-class accuracy. Appearance-only was statistically indistinguishable (QWK 0.806; delta QWK +0.006, 95% CI −0.023 to 0.036, p = 0.68), whereas anatomy-only reached 0.444, and a level-and-side-only reference reached 0.314. Boundary discrimination was strong (area under the curve 0.92–0.99), 98.3% of predictions fell within one grade, and performance was consistent across scanner vendors. Morphometric associations were confounded by level: the apparent spondylolisthesis effect (rho −0.349) disappeared after adjustment (rho −0.000), while disc height, null when pooled (rho +0.021), emerged as a genuine within-level effect (rho −0.108). A fine-tuned ordinal image classifier achieved strong agreement for four-grade lumbar foraminal stenosis classification. The evaluated segmentation-derived morphometric features did not improve performance beyond imaging alone, and several apparent anatomic associations reflected confounding by lumbar level. External and prospective validation in complete clinical MRI workflows are needed before implementation. Full article

The digitalization of sustainable energy systems increasingly relies on containerized Internet of Things services deployed across cloud–edge architectures. These services introduce software supply-chain risks associated with public container images and their dependencies. This study evaluates 22 container images representing 13 official or vendor-maintained technologies used for data storage and processing, communication, proxy and API services, and application runtime environments. Each image was analysed using Trivy and Grype, resulting in 44 vulnerability scans performed using vulnerability databases available on 14 June 2026. The effect of image minimization was assessed using five strictly matched standard–minimized pairs, while scanner agreement was evaluated for all images using unique CVE sets, the Jaccard coefficient, and symmetrical and directional Tversky indices. Across the complete sample, Trivy reported 7006 vulnerability findings and Grype reported 2299. Within the strictly matched sample, findings decreased from 5181 to 193 for Trivy and from 982 to 290 for Grype. However, these reductions were strongly influenced by the Ruby image, and the exact Wilcoxon signed-rank test did not confirm a statistically significant general minimization effect ($p=0.250$). Redis, HAProxy, and Ruby showed substantial reductions, whereas Caddy remained unchanged and both .NET SDK variants produced zero findings. The set-based analysis revealed incomplete and asymmetric agreement between the scanners, demonstrating that similar aggregate counts may represent different CVE profiles. Operational prioritization of six selected image variants further showed differences in remediation availability, EPSS scores, and CISA KEV inclusion. The results indicate that image minimization can reduce scanner findings but does not independently confirm container security. A multi-tool DevSecOps process combining immutable digest verification, Software Bills of Materials, vulnerability prioritization, image rebuilding, and continuous rescanning is therefore recommended. Full article

Histopathology foundation models (FMs) have become widely used as patch-level feature extractors in computational pathology (CPath), where domain shift is a central challenge, yet their generalization ability across acquisition centers and scanning platforms remains insufficiently studied. In this work, we evaluate the domain generalization of ten state-of-the-art FMs on two multi-source datasets with different supervision settings: SemiCOL, a colorectal cancer cohort of 499 whole-slide images (WSIs) for weakly labeled slide-level binary tumor classification, and BEETLE, a breast cancer cohort of 583 WSIs for patch-level four-class tissue classification. In an ablation-style setting, FMs are used as patch-level feature extractors, with patch embeddings mean-pooled into slide-level representations for SemiCOL, and a lightweight multi-layer perceptron trained for slide-level and patch-level classification on SemiCOL and BEETLE, respectively. To test FM domain generalization, we use three evaluation protocols: a Baseline source-mixed 5-fold cross-validation (CV) and two leave-source-out CV settings that assess cross-center and cross-scanner performance. On SemiCOL, all FMs achieve near-saturated performance, indicating stable performance under acquisition-source domain shift for slide-level Tumor vs. Benign classification. In contrast, BEETLE reveals clear generalization gaps, with scanner-induced domain shift more challenging than center-induced domain shift, and class-wise results showing that performance losses concentrate in epithelial discrimination. Overall, Virchow2 shows the strongest robustness across all evaluated protocols. These findings show that standard source-mixed CV can overestimate domain generalization across centers and scanners, and that FM choice matters in multi-source cohorts, especially for more challenging CPath tasks, where cross-domain failures are more visible. Full article

Ischaemic stroke constitutes a leading cause of global disability. Rapid extraction of the infarct core from multimodal computed tomography perfusion (CTP) imaging guides reperfusion therapy and clinical decision-making. Deep learning algorithms automate this delineation, yet hospital translation is hindered by high-dimensional data, inter-scanner variability, and the low contrast of early ischaemia. Architectural comparisons in the literature frequently carry methodological biases originating from disparate preprocessing protocols and data partitions. This study reduces these variables by evaluating three segmentation strategies under a shared preprocessing pipeline and an identical data partition using the ISLES 2024 dataset. Three models were trained on the same 133-patient partition using a shared preprocessing pipeline based on morphological skull-stripping and modality-specific clinical intensity ranges. The data, preprocessing, and partitions are held constant across models, while framework-dependent factors (optimiser, patch size, physical field of view, spatial resampling, augmentation policy, and model capacity) remain coupled to each architecture and are therefore treated as part of the compared strategy rather than as fully isolated variables. The first of these is a single-stage 5-channel nnU-Net, followed by a two-stage cascaded nnU-Net (2 and 7 channels) and a lightweight Transformer (SegFormer3D). Evaluation on a fixed 15-patient held-out test set isolated the architectural performance. The cascade model achieved the highest Dice Similarity Coefficient (0.224). The single-stage nnU-Net provided the most precise volumetric estimation, recording an Absolute Volume Difference (AVD) of 23.70 mL and a lesion-wise F1-score of 7.60%. On the other hand, SegFormer3D returned the lowest overall metrics (DSC 0.163, AVD 27.28 mL, F1 2.30%). In the small held-out cohort, paired statistical testing did not reveal significant differences between models, so the reported orderings describe the present dataset and experimental configuration rather than a general architectural law. Within these limits, the local inductive bias of the convolutional models retained an empirical advantage over the single Transformer evaluated when processing this moderately sized neuroimaging dataset, and complex cascade topologies offered only marginal gains compared with a well-calibrated single-stage network. Although the predictive segmentation of infarcted tissue at acute stages still demands computational improvements, these results suggest that preprocessing quality is at least as decisive for clinical impact as increasing the complexity of neural architectures. Full article

The Yellow-legged Gull (Larus michahellis) is one of the most frequently admitted seabirds to wildlife rehabilitation centers in the Mediterranean and Atlantic regions. Despite its clinical relevance, detailed computed tomographic references of the coelomic cavity are lacking. The aim of this study was to describe the normal cross-sectional anatomy and computed tomographic appearance of the coelomic cavity in this species and to develop an anatomical atlas to facilitate diagnostic image interpretation. Eight young adult female Yellow-legged Gull were examined post-mortem. Six specimens underwent computed tomography using a 16-slice helical scanner, followed by frozen transverse sectioning, while two specimens were dissected to provide detailed anatomical correlation. CT datasets were evaluated using bone, soft tissue, and pulmonary window settings, and three-dimensional volume-rendered reconstructions were generated. Anatomical dissections, cross-sectional slices, and CT images enabled identification and characterization of the principal structures of the respiratory, cardiovascular, digestive, urinary, and reproductive systems, as well as their spatial relationships within the coelomic cavity. The combined evaluation of gross anatomy and CT imaging allowed accurate recognition of major organs, air sacs, large vessels, and skeletal landmarks throughout the examined sections. The atlas provides a comprehensive reference for normal coelomic anatomy in Larus michahellis and establishes a baseline for the interpretation of CT examinations in clinical, rehabilitation, and research settings. These findings may improve diagnostic accuracy and support future investigations of coelomic disorders in this species. Full article

Aiming at addressing the problems of complex geometric features, high inter-class similarity, and insufficient single-scale information in semantic segmentation of point clouds for ancient building interior components, this paper takes the Ba Wang Academy of Shenyang Jianzhu University as the research object and proposes a Geometric Feature Multi-scale Network (GFMN). First, a point cloud dataset containing five types of components—windows, beams, walls, roofs, and columns—was collected and constructed using a FARO Focus3D X330 terrestrial laser scanner (FARO Technologies, Lake Mary, Florida, USA). Second, 46-dimensional handcrafted geometric descriptors were extracted for each discrete point from four aspects: basic point attributes, local geometric features, density and scale features, and multi-scale fusion. On this basis, features were grouped according to semantics and fed into independent encoding branches, where a gated adaptive fusion mechanism was employed to dynamically adjust the contribution of each branch, and optimization was performed in combination with a prototype classification head and a joint loss function. Experimental results show that the proposed method achieved an overall accuracy of 93.17% on the test set, significantly outperforming state-of-the-art methods such as PointNet, PointNet++, Point Transformer, and Point Cloud Transformer. This study provides an effective solution for high-precision semantic segmentation of ancient building interior components. Full article

This study addresses the influence of process conditions on the dimensional accuracy, geometric deviations, and surface quality of PLA parts manufactured using high-dynamics material extrusion (MEX) technology. The aim was to identify the dominant sources of variability and to assess within-condition manufacturing consistency and inter-machine consistency. The investigation considered two 3D printers, nine build locations on the working platform, two printing strategies (layer-by-layer and model-by-model), and model face orientation. Additionally, an exploratory comparison of aligned and random seam configurations and an analysis of local temperature variations within the build chamber were performed. Regular dodecahedron geometries were manufactured using a Bambu Lab P1S system and processed under identical high-quality printing parameters. Dimensional measurements were performed using a Linear 100 universal length measuring machine, while full-field geometric deviations were acquired using a GOM Scan 1 structured-light 3D scanner. Surface roughness (Ra) was measured with a MarSurf XR 20 profilometer. Part orientation is the dominant source of dimensional variability, representing the largest relative contribution to linear deviation in the mixed-effects model (ΔR² = 0.776), while local temperature variations near the printing zone were associated with location-dependent dimensional deviations. Within the supplementary temperature dataset, the regression model including temperature and printers explained 66% of the variability in mean linear dimension. This association provides indirect evidence of a thermal contribution but does not establish direct causality. The layer-by-layer strategy provided better dimensional stability than the model-by-model approach. In the exploratory seam comparison, seam configuration did not explain the orientation-dependent LD pattern. Surface roughness variability was primarily geometry-driven (ΔR² = 0.852). Variability between independent manufacturing series and specimens for linear deviation and Ra was low after accounting for the investigated factors, indicating consistent process performance under constant settings; however, the present design did not allow measurement repeatability and reproducibility to be separated. In conclusion, dimensional accuracy in high-dynamics MEX is strongly associated with part orientation, while thermal variations may represent an additional contributing factor; however, the observed correlation between thermal conditions and dimensional variability does not establish direct causality. Full article

Alzheimer’s disease (AD) is the most prevalent neurodegenerative disorder worldwide, affecting approximately 56.9 million people in 2021 and projected to reach 152 million by 2050. Its defining pathological features, amyloid-beta plaques and neurofibrillary tangles, accumulate for up to two decades before cognitive symptoms emerge, placing the preclinical and mild cognitive impairment (MCI) stages at the centre of the early detection problem. Despite this, current diagnostic practice in routine clinical settings remains unreliable, with post-mortem studies placing the specificity of clinical AD diagnosis between 44.3 and 70.8% even in specialist memory clinics. Machine learning (ML) and deep learning (DL) applied to neuroimaging and electrophysiological data have emerged as candidate tools for closing this diagnostic gap, yet whether the accuracy figures reported in published studies translate into clinically useful performance on independent data remains unresolved. This study presents a structured comparative review of machine learning and deep learning methods reported across four publicly available Alzheimer’s disease datasets, namely the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the Open Access Series of Imaging Studies (OASIS), the OpenNeuro ds004504 electroencephalography (EEG) dataset, and the Kaggle Alzheimer’s magnetic resonance imaging (MRI) dataset. Thirteen model families are examined through the published literature rather than through new experiments, and for each model and dataset combination, the best accuracy reported in the source study is recorded alongside the model’s mathematical formulation. All performance figures reported in this abstract and throughout the paper are taken from the published studies reviewed, not from new experiments conducted by the authors. Across the reviewed studies, deep learning architectures pre-trained on ImageNet and fine-tuned on neuroimaging data are reported to produce the highest accuracy on MRI classification tasks. Residual Network (ResNet)-101 is reported at 98.21 percent on ADNI and 97.45 percent on OASIS, while the IncepRes fusion architecture reaches 98.35% on OASIS by combining multi-scale feature extraction from InceptionV3 with residual connectivity from ResNet152V2. Traditional machine learning classifiers remain competitive on tabular clinical and biomarker data, with Extreme Gradient Boosting (XGBoost) reaching 91% on ADNI multiclass features. Logistic Regression achieves 82 to 85% on binary ADNI classification and is the only classifier in this review that provides explicit per-feature prediction contributions without post hoc tooling. Gaussian Naïve Bayes achieves 80 to 83% on the same task. On the OpenNeuro EEG dataset, K-nearest neighbours (KNN) with singular value decomposition (SVD) entropy features achieves 91% binary accuracy, with feature engineering quality determining performance more reliably than classifier architecture. Eight principal findings emerge from the cross-dataset analysis. Binary classification consistently outperforms multiclass by 10 to 30% across all datasets, reflecting the genuine biological ambiguity of the mild cognitive impairment category. Dataset size and augmentation predict reported accuracy more reliably than model architecture. Ensemble methods outperform individual classifiers by 5 to 8% in nearly every imaging study. Deeper architectures can overfit small clinical cohorts. EEG models trail MRI models by approximately 10 to 15% on comparable binary tasks. Cross-dataset generalisation has not been systematically evaluated in most studies, and the few that have tested it report accuracy drops of 5 to 10% or more when models encounter data from different scanners or cohorts. Eight recurring limitations constrain the clinical utility of these findings. Small sample sizes and limited demographic diversity, severe class imbalance inflating raw accuracy metrics, poor cross-dataset generalisation driven by scanner heterogeneity, limited deep learning interpretability, the dominance of binary over multiclass tasks, the absence of longitudinal modelling despite available datasets, inadequate standardisation of preprocessing and evaluation protocols, and the signal-to-noise ratio constraints specific to EEG recordings of elderly patients collectively define the gap between benchmark performance and clinical readiness. Future work must prioritise multi-centre training cohorts, multimodal fusion architectures, longitudinal progression modelling, and standardised interpretability evaluation as non-optional requirements for any system intended for clinical deployment. Full article

Background/Objectives: Prostate cancer (PCa) frequently metastasizes to bone, leading to severe clinical complications and reduced quality of life. Accurate and robust imaging-based characterization of bone lesions is therefore critical for diagnosis and treatment planning. Radiomics has emerged as a powerful tool for extracting quantitative information from medical images; however, classical radiomics features are often affected by inter-scanner variability, segmentation dependence, and limited ability to describe lesions with complex biological heterogeneity. This study aims to introduce a translational graph-based radiomics approach designed to extract novel quantitative descriptors with improved robustness and clinical reliability. Methods: A graph representation was derived from segmented Positron Emission Tomography/Computed Tomography (PET/CT) bone lesions by generating a point cloud followed by Delaunay triangulation to preserve geometric information. Graph signal processing techniques were applied to extract three classes of features: orientation, connectivity, and transform-based descriptors. The dataset included PET/CT scans from 50 PCa patients acquired using two different scanners, comprising 92 bone lesions classified as benign or malignant. Correlation analysis with classical radiomics features was performed to assess information redundancy. Robustness against batch effects and segmentation variability was evaluated. Classification performance was tested using Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) models based on proposed features, classical features, and their combination. Results: The proposed features captured non-redundant information compared to classical radiomics and demonstrated superior robustness to scanner-related batch effects and segmentation variability. In classification tasks, models using the proposed features consistently outperformed those based on classical radiomics. Using LDA, the proposed features achieved a mean balanced accuracy of 69.68% and a mean Area Under the Curve (AUC) of 72.14%. With SVM, they achieved a mean balanced accuracy of 65.16% and a mean AUC of 66.49%, exceeding the performance of classical and combined feature sets. Conclusions: This study presents a translational graph-based radiomics framework that extends beyond conventional methodologies, improving robustness and diagnostic performance. The proposed approach shows promise as an integrative tool for more reliable PET/CT-based characterization of bone lesions in prostate cancer. Full article