Search Results (2,548)

Background: Allergic contact dermatitis (ACD) is a common inflammatory skin disease and patch testing (PT) remains the gold standard for its diagnosis; however, PT interpretation is time-consuming and prone to inter-observer variability. Growing advances in digital imaging and artificial intelligence (AI) have encouraged the development of automated PT evaluation systems. This review aimed to summarize the use of deep learning networks (DNNs) and image-preprocessing techniques for PT classification. Methods: A literature review was conducted to identify original research published between 2020 and 2025 that applied deep learning algorithms to PT image analysis. Included studies were assessed with respect to model architecture, dataset characteristics, preprocessing strategies, and diagnostic performance. Results: Six original studies employing deep learning for PT image classification met the inclusion criteria. They employed a range of architectures, including YOLOv5x, EfficientNetB0, Xception, and custom CNN models. Reported diagnostic performance varied, with accuracy values ranging from 90% to 99.5%, F1-scores from 0.37 to 0.98, and AUROC values up to 0.94. Despite promising results, models remain unreliable for ICDRG grading, especially for severe reactions, and methodological variability in dataset composition, imaging conditions, preprocessing pipelines, and classification tasks limits comparability across studies. Conclusions: Deep learning shows promise for automated PT interpretation, but further standardized and multicenter studies with detailed preprocessing protocols and comprehensive ICDRG grading are required for clinical implementation. Full article

Methane source-rate inversion from remote-sensing plume imagery is essential for emissions monitoring, but its accuracy is often limited by uncertainty in ancillary wind information. This study examines how physical knowledge can be integrated into a deep-learning inversion model when the available wind input is imperfect. Using a controlled large-eddy-simulation (LES) benchmark designed for EnMAP/PRISMA-style imaging-spectrometer methane quantification, we compare six models that span image-only regression, flexible wind conditioning, simplified hard integrated-mass-enhancement (IME) coupling, and soft physics-guided learning under clean inputs, deterministic wind bias, stochastic Gaussian wind noise, and source-rate-stratified tests. Under clean benchmark conditions, flexible wind conditioning provides the best scalar accuracy, with FiLM reaching a mean absolute percentage error (MAPE) of 6.19% and a root mean squared error (RMSE) of 1323.36, followed closely by Concat (MAPE 6.37%, RMSE 1325.69). The simplified hard-coupling model is sensitive to wind perturbations: DIN-hard rises from MAPE 8.44% under clean inputs to 31.39% and 26.89% under deterministic wind-bias multipliers α = 0.7 and α = 1.3, respectively, and becomes unstable under stronger Gaussian wind noise in the tested protocol. By contrast, DIN-soft-v2 remains competitive under clean conditions (MAPE 6.39%, RMSE 1360.94), follows smoother degradation under biased or noisy wind, and improves plume spatial diagnostics relative to DIN-soft (center-of-mass shift 3.92 versus 4.07 pixels; plume alignment degree 2.60 versus 2.72 degrees). The calibrated IME-style physical baseline reaches a clean MAPE 24.45%, indicating that the learning-based models substantially outperform this benchmark physical proxy. Within this LES-based benchmark and the tested wind-perturbation protocols, the results suggest that IME-inspired physical knowledge is more robustly incorporated as a calibratable soft prior than as the simplified hard log-additive forward coupling considered here; however, transfer to real satellite scenes still requires validation. Full article

Background/Objectives: Triple-rule-out CT angiography (TRO-CTA) enables simultaneous evaluation of coronary, pulmonary, and aortic causes of acute chest pain, but conventional single-acquisition protocols may compromise vascular enhancement because of conflicting contrast timing requirements. This study evaluated whether a physiology-based sequential helical–axial–helical acquisition strategy could provide consistent tri-territory enhancement in emergency settings. Methods: In this retrospective single-center study, 71 consecutive evaluable emergency department patients (mean age, 66.6 ± 17.0 years; 33 women) with undifferentiated acute chest pain underwent TRO-CTA using a structured sequential protocol (pulmonary, coronary, and aortic phases) guided by individualized test-bolus timing. Objective image quality was assessed using vascular attenuation, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR); subjective image quality was independently graded by two radiologists. Results: Mean vascular attenuation exceeded predefined diagnostic thresholds in all territories (pulmonary 546.7 ± 237.8 HU [95% CI, 490.4–603.0]; coronary 438.8 ± 113.9 HU [95% CI, 411.9–465.8]; aortic 604.3 ± 190.9 HU [95% CI, 559.2–649.5]). Diagnostic interpretability was achieved in all three territories in every technically analyzable examination without repeat contrast-enhanced imaging. Median subjective image-quality scores were 5 (IQR, 4–5) for pulmonary, 4.5 (IQR, 4–5) for coronary, and 4 (IQR, 4–5) for aortic phases; interobserver agreement was good to excellent. Mean total DLP was 461.5 ± 122.5 mGy·cm. Conclusions: A sequential physiology-based TRO-CTA strategy is technically feasible in a tertiary emergency setting and provides consistent tri-territory enhancement. Because this was a single-arm technical validation study, prospective comparative and outcome-based studies are required to confirm its clinical impact. Full article

Acid phosphatase (ACP) is a clinically important enzyme whose early-stage detection is hindered by its extremely low abundance, nonspecific tissue distribution, and rapid loss of activity under conventional analytical conditions. Herein, we present a target-driven in situ nanozyme synthesis strategy that enables rapid and ultrasensitive point-of-care testing (POCT) of ACP. In this approach, ACP catalyzes the hydrolysis of L-ascorbic acid 2-phosphate sesquimagnesium (AAPS), producing ascorbic acid (AA). The generated AA partially reduces Fe³⁺ ions to Fe²⁺, thereby initiating alkaline co-precipitation and in situ formation of Fe₃O₄ nanoparticles. Polyvinylpyrrolidone (PVP) stabilizes the nanoparticles and preserves catalytic accessibility, while their intrinsic magnetism allows for efficient magnetic separation to eliminate matrix interference. The resulting Fe₃O₄@PVP nanozymes display pronounced peroxidase-like activity, catalyzing hydrogen-peroxide-mediated oxidation of 3,3′,5,5′-tetramethylbenzidine (TMB). Quantitative readout can be achieved using either spectrophotometric analysis or smartphone imaging. The sensing platform achieves a detection limit of 0.021 U/L within 40 min and demonstrates excellent sensitivity, selectivity, and operational robustness. Successful validation in human serum confirms its clinical feasibility, while smartphone-based imaging enables portable and low-cost quantification suitable for decentralized diagnostics. Collectively, this work establishes a generalizable paradigm for target-triggered nanozyme generation aimed at detecting low-abundance and labile biomarkers. Full article

Background: Cluster epilepsy related to PCDH19 is a rare X-linked disorder that mainly affects females. Atypical presentations, such as infantile spasms, are exceptionally rare, leading to diagnostic and therapeutic challenges. Case Summary: An 8-month-old girl presented with early-onset, drug-resistant epilepsy from 4 months of age, displaying tonic and focal seizures, infantile spasms with hypsarrhythmia, and neurodevelopmental regression. Whole exome sequencing identified a novel heterozygous mutation, c.1072del (p.Val358SerfsTer10), in the PCDH19 gene. Following vigabatrin therapy for infantile spasms, the patient subacutely developed a movement disorders: dystonic movements and action tremor in forearm and hands. Brain Magnetic Resonance Imaging (MRI) revealed symmetrical restricted diffusion and cytotoxic edema in both the thalami and the internal capsules, confirming Vigabatrin-Associated Brain Abnormalities on MRI (VABAMR). Concurrently, a systemic carnitine deficiency was identified, which could additionally have compromised mitochondrial bioenergetics and intensified the tendency to cytotoxic cerebral edema. A strategic reduction in vigabatrin led to complete resolution of movement disorder and neuroimaging abnormalities. Conclusions: This case underscores the high phenotypic variability of PCDH19 mutations, and also the importance of early advanced genetic testing and the consideration of even rare side effects of drugs in differential diagnosis. It is unclear whether that baseline carnitine deficiency could potentially increase the risk of VABAMR. Full article

Endometriosis is a chronic, debilitating condition affecting approximately 10–15% of reproductive-aged women and it is often associated with significant diagnostic delays due to its heterogeneity and unreliable non-invasive tests. Artificial intelligence (AI) offers innovative methods for improving endometriosis diagnosis, prognosis and research via advanced pattern recognition and data analysis capabilities. The integration of AI in diagnostic workflow has the potential to improve efficiency, accuracy, and patient outcomes. This review summarises current developments of AI—including machine learning, deep learning, and natural language processing—in the diagnostic workflow of endometriosis. It analyses different fields of diagnostics ranging from AI-assisted imaging in detection of pouch of Douglas to multi-omics biomarkers assisting the clinical decision process. AI can enhance accuracy, reducing diagnostic delays and supporting personalised treatment planning. However, there are multiple limitations, such as small datasets, overfitting, and lack of external validation and variability. Further research and evaluation are required before it can be implemented into healthcare systems. AI holds promise as a non-invasive, scalable adjunct to current diagnostics, potentially reducing the economic and personal burden endometriosis carries. Full article

Background/Objectives: Accurate evaluation of knee osteoarthritis (KOA) severity is critical for optimal patient care, yet manual radiographic grading remains subject to observer variability. This study aims to evaluate the performance of a fine-tuned contrastive language–image pre-training (CLIP) framework designed to assist clinicians in grading KOA severity in plain radiographs using the Kellgren–Lawrence (KL) classification system (Grades 0–4). Methods: The model operates by projecting visual features from radiographs and standard textual clinical descriptions into a shared embedding space. Training was conducted using 8260 posterior–anterior (PA) fixed-flexion X-ray images from the Osteoarthritis Initiative (OAI) dataset. For robust external evaluation across distinct data distributions, the model was tested on an independent dataset consisting of 1650 plain radiographs. Results: When evaluated on the external validation dataset, the fine-tuned CLIP model achieved an accuracy of 76.94% and an F1-score of 76.66%. Comparative analysis demonstrates that these aligned vision-language representations provide competitive, stable diagnostic capabilities even when applied to an entirely independent data distribution. Conclusions: Fine-tuned CLIP architectures offer a viable and valuable foundation for semantically transparent, computer-aided evaluation of KOA. Full article

Background/Objectives: Accurate bone lesion classification on radiographs is critical for clinical decision-making and forensic identification. Existing deep learning approaches treat radiographs as whole images, neglecting available spatial annotations and clinical metadata. To develop an ROI-guided deep learning framework integrating clinical metadata for three-class (Normal, Benign, Malignant) bone lesion classification and to assess its clinical safety profile. Methods: Using the BTXRD (3746 radiographs: 1879 Normal, 1525 Benign, 342 Malignant), an EfficientNetV2-S backbone was combined with an 11-dimensional metadata MLP trained on ROI-cropped regions. Training employed Focal Loss with adaptive class weighting, Mixup/CutMix augmentations, Stochastic Weight Averaging, and Test-Time Augmentation. Five-fold stratified cross-validation with bootstrap confidence intervals (n = 2000) and probability calibration metrics were used. Results: The framework achieved 96.05% accuracy (95% CI: 95.41–96.66%), 93.94% balanced accuracy, 92.62% macro F1-score, and 99.21% macro-AUC (95% CI: 98.89–99.42%). Critically, near-zero Malignant-to-Normal misclassifications occurred (1/342, 0.29%; 95% Clopper–Pearson CI: 0.01–1.62%) across all 3746 predictions. The minority Malignant class attained F1 = 83.53% despite comprising only 9.1% of the dataset. Conclusions: ROI-guided deep learning with metadata fusion achieves state-of-the-art bone lesion classification with clinically safe error patterns and probability outputs whose calibration was explicitly quantified, supporting its potential as a decision support tool in diagnostic radiology and forensic anthropology, pending external validation on independent cohorts. Full article

Background/Objectives: Shear wave elastography (SWE) may provide an objective method for quantifying piriformis muscle stiffness, but its clinical and research use requires evidence that the measurement procedure is reliable. This study aimed to determine the intra- and inter-examiner reliability of a standardized SWE protocol for assessing piriformis muscle stiffness and to provide measurement error thresholds for clinical interpretation. Methods: Twenty-one healthy volunteers were assessed bilaterally by two examiners with different levels of ultrasound experience. The piriformis muscle was identified in the long axis beneath the gluteus maximus, and SWE images were acquired using a standardized protocol. Each side was measured twice by each examiner, resulting in 168 ultrasound images. Reliability was analyzed using side-specific observations (n = 42). Intraclass correlation coefficients (ICCs), standard errors of measurement (SEMs) and minimal detectable changes (MDCs) were calculated for shear modulus and shear wave speed. Results: Inter-examiner reliability was good for single measurements, with ICCs of 0.872 for shear modulus and 0.813 for shear wave speed. When the average of two measurements was used, ICCs were similar, reaching 0.876 and 0.832, respectively. Inter-examiner MDC values ranged from 6.1 to 6.2 kPa for shear modulus and from 0.37 to 0.39 m/s for shear wave speed. Intra-examiner reliability was excellent for both examiners, with ICCs ranging from 0.938 to 0.979. Test–retest MDC values ranged from 2.7 to 3.3 kPa for shear modulus and from 0.19 to 0.22 m/s for shear wave speed. Conclusions: SWE provides good inter-examiner and excellent intra-examiner reliability for assessing piriformis muscle stiffness using a standardized acquisition protocol. Longitudinal assessments should preferably be performed by the same examiner, and changes should be interpreted in relation to SEM and MDC values, particularly in multi-examiner settings where absolute measurement error is larger. These thresholds reflect measurement error and should not be interpreted as evidence of diagnostic validity or responsiveness to treatment. Full article

Precision medicine in prostate cancer (PCa) is increasingly best understood as a continuum linking risk-adapted detection, multimodal diagnosis and phenotyping, and implementation-ready decision pathways. Contemporary clinical guidelines emphasize structured diagnostic strategies, appropriate use of advanced imaging, and selective deployment of biomarkers when results can alter management. Upstream risk enrichment using polygenic risk scores and multivariable prediction models may improve the yield of clinically significant disease while mitigating harms related to overdiagnosis. At the point of suspicion, magnetic resonance imaging-first pathways and reflex biomarker testing provide practical tools to reduce unnecessary biopsy while maintaining safeguards for the detection of clinically important disease. Beyond diagnosis, prostate-specific membrane antigen positron emission tomography refines disease-state phenotyping in initial staging, biochemical recurrence, and limited-burden presentations, while standardized acquisition and reporting improve reproducibility and multidisciplinary communication. Germline and tumor-based molecular profiling should be operationalized as a longitudinal care process with clear consent, turnaround targets, and test-to-action rules that define what each result enables at specific decision nodes. Finally, longitudinal monitoring approaches, including liquid biopsy and artificial intelligence-enabled pathology, are evolving rapidly and require transparent reporting and rigorous risk-of-bias appraisal before broad clinical adoption. This narrative review synthesizes key evidence across the precision continuum and outlines a decision-node-based, test-to-action framework for maximizing clinical benefit, maintaining quality, and ensuring equitable access. Full article

Background/Objectives: Brain magnetic resonance imaging (MRI) is an important imaging modality for assessing neurological disorders. However, automatic multi-class MRI classification remains challenging because of visual similarity between disease categories, heterogeneous pathological patterns, class imbalance, and the need for reliable confidence estimation. This study aims to develop a comprehensive and well-calibrated deep learning framework for image-level brain MRI classification across multiple neurological categories. Methods: This paper introduces a new deep learning framework, MCND-ComputeNet++, for brain MRI classification into eight image-level categories using the MCND dataset, which comprises 16,400 two-dimensional brain MRI images belonging to eight diagnostic categories: AD-MildDemented, AD-ModerateDemented, AD-VeryMildDemented, BT-glioma, BT-meningioma, BT-pituitary, MS, and Normal. The proposed model uses a single pretrained EfficientNetV2-S backbone to extract hierarchical feature maps from three intermediate stages. These multi-level features are projected into a common latent space, spatially aligned, adaptively fused through learnable gated multi-scale fusion, further refined using convolutional processing, and aggregated using spatial attention pooling before classification. The training strategy combines class-balanced focal loss with label smoothing, MixUp/CutMix regularization, exponential moving average weight smoothing, warmup cosine learning-rate scheduling, temperature scaling, and test-time augmentation to improve generalization and calibration. The framework was evaluated using accuracy, precision, recall, macro-F1, macro-AUC, macro-average precision, expected calibration error, Brier score, bootstrap confidence intervals, ablation analysis, McNemar testing, and comparisons against standard pretrained baseline models. Results: MCND-ComputeNet++ achieved mean accuracy, macro-F1, macro-AUC, and macro-average precision values of 0.9738, 0.9771, 0.9993, and 0.9971, respectively, with narrow bootstrap confidence intervals indicating stable image-level performance. These findings should be interpreted as image-level/slice-level performance on MCND, because patient-level identifiers and subject-wise splitting were not available. These results outperformed most evaluated baselines, including ResNet50, DenseNet121, EfficientNetB0, EfficientNetV2-S with a standard classifier, Swin-Tiny, and ConvNeXt-Tiny, across several discrimination and calibration metrics. Compared with ConvNeXt-Tiny, the proposed model achieved higher macro-AUC and macro-average precision, together with a lower ECE and Brier score, suggesting improved image-level discrimination and confidence reliability. Compared with the EfficientNetV2-S standard classifier, accuracy increased from 0.9308 to 0.9738, while the Brier score decreased from 0.1045 to 0.0400. Conclusions: The results suggest that MCND-ComputeNet++ is a promising image-level brain MRI classification framework for the eight MCND categories. The proposed model integrates hierarchical feature extraction, shared latent projection, gated multi-scale fusion, convolutional refinement, spatial attention pooling, and calibrated inference within a unified architecture. However, because the current evaluation was conducted at the image/slice level without available patient-level identifiers, the findings should not be interpreted as patient-level clinical diagnostic validation. Further studies using subject-wise splitting, external multi-center datasets, 3D volumetric modeling, and multimodal clinical information are required to assess generalizability and potential clinical decision-support applicability. Full article

Hepatocellular carcinoma (HCC) remains a major global health burden, characterized by late diagnosis, limited therapeutic options, and high mortality rates. Conventional diagnostic tools such as serum α-fetoprotein testing and imaging lack sufficient sensitivity for early detection. In recent years, liquid biopsy has emerged as a minimally invasive approach that enables real-time molecular profiling of tumors through the analysis of circulating biomarkers such as nucleic acids, proteins, and extracellular vesicles. Recent advances have underscored exosomes—nano-sized extracellular vesicles (EVs) secreted by nearly all cell types—as pivotal mediators of intercellular communication and dynamic carriers of tumor-derived molecular information, offering exciting prospects for early cancer detection and personalized therapy. In HCC, EV microRNAs (miRNAs) participate in multiple oncogenic processes, including proliferation, angiogenesis, epithelial–mesenchymal transition, and immune modulation. Specific EV-associated miRNAs, such as miR-21, miR-122, miR-224, and miR-221, show distinctive expression profiles in HCC and correlate with tumor stage, metastasis, and patient prognosis. Moreover, panels of circulating EV-associated miRNAs demonstrate superior diagnostic accuracy compared with traditional biomarkers, underscoring their potential as non-invasive tools for early detection and disease monitoring. Their inherent stability in biofluids and resistance to enzymatic degradation further support their application in liquid biopsy approaches. Despite promising results, continued research is essential to validate EV-associated miRNA signatures and to integrate these “silent messengers” into routine clinical practice for precision management of hepatocellular carcinoma. Full article

Background: Cataract diagnosis and management pose a significant global health challenge, contributing to 17 million cases of blindness and over 83 million cases of vision impairment worldwide in 2020. This issue is particularly acute in regions lacking adequate ophthalmological services, where a shortage of eye care clinicians and specialized equipment like slit-lamp cameras leads to late diagnoses. To address this accessibility gap, we developed a computer-assisted cataract grading system using smartphone-acquired external eye photographs. This approach utilizes image processing and deep learning on a standard, hardware-free smartphone, offering a low-cost and portable alternative to traditional equipment. Methods: The study introduces a new advanced algorithm to stratify cataract severity into three distinct stages: normal, pre-mature, and mature. The methodology was developed using a combined dataset of 799 images sourced from the Cataract v01 Computer Vision Project and the Indian Institute of Technology, Delhi. A key step is isolating the iris and lens using a region of interest (ROI) extraction procedure powered by the open-source MediaPipe framework. Key to the algorithm’s efficacy is the use of transfer learning, adapting four customized ResNet architectures (ResNet-18, ResNet-34, ResNet-50, and ResNet-101) to address medical image analysis intricacies. These models were fine-tuned with specific modifications, including dropout layers and the Adam optimizer, for analyzing the digital periocular images. Results: Evaluation of the models shows varied performance across the various architectures when classifying cataract stages. While the simpler ResNet-18 model exhibited the lowest performance, the deeper models showed significant improvement. The ResNet-50 architecture achieved the highest accuracy of 94%. This model also demonstrated excellent precision (94%), recall (95%), and an F1-score of 95% in multi-class classification, outperforming the other tested models. Its depth enables precise cataract classification, positioning it as a robust and reliable tool for potential medical diagnostic deployment. Conclusions: Deep learning-based analysis of smartphone-acquired external eye images demonstrated feasibility for cataract detection in this study. This method could be a scalable and easy-to-use addition to screening, especially in places where resources are limited. Further work is needed to expand the dataset and to validate the algorithm against established clinical grading systems before broader clinical implementation. Full article

Purpose: Variation in knee MRI diagnostic yield is often interpreted as reflecting imaging effectiveness. However, in real-world healthcare systems, diagnostic yield may instead be driven by referral behavior and patient selection. Understanding this distinction is essential for evaluating imaging utilization, healthcare efficiency, and potential overuse of advanced imaging. This study examines whether differences in MRI yield reflect imaging pathways or underlying referral patterns in routine clinical practice. Materials and Methods: This retrospective cohort study included consecutive patients undergoing knee MRI between January 2020 and December 2024. Patients with red flag indications were excluded to focus on discretionary imaging. The primary outcome was clinically relevant MRI findings based on final report impressions. The primary exposure was prior X-ray before MRI. Multivariable logistic regression was used for adjusted analysis, including age, sex, trauma status, mechanical symptoms, and symptom duration. Results: Among 486 patients, 59.5% had prior X-ray. Clinically relevant MRI findings were less frequent among patients with prior X-ray (40.1%) than among those without (49.7%), corresponding to an absolute difference of 9.6%. After adjustment for sex and clinical covariates, prior X-ray showed lower odds of clinically relevant findings, although this association was attenuated and no longer statistically significant (aOR 0.74, 95% CI 0.50–1.10; p = 0.138). Male sex was independently associated with higher odds of clinically relevant MRI findings (aOR 2.48, 95% CI 1.61–3.83; p < 0.001). Formal interaction testing did not demonstrate significant effect modification by trauma status (p = 0.317). These findings suggest that variation in MRI yield may reflect differences in referral pathways, patient selection, and healthcare utilization patterns. Conclusions: MRI yield in routine practice may be influenced by differences in clinical context and referral-related patient selection. Further studies are needed to better understand the contribution of imaging pathways to observed variation in diagnostic yield. Full article

Background: Cage subsidence after minimally invasive transforaminal lumbar interbody fusion raises revision risk and costs. Intraoperative computed tomography (CT) provides high-resolution, three-dimensional visualization of the endplate–cage interface and serves as a practical—though itself imperfect—reference standard for early subsidence, but it is not available at all institutions. Plain X-ray is widely available and inexpensive, but lower in resolution. The clinically relevant question is therefore not whether CT and X-ray are equivalent, but rather which X-ray protrusion depth measurement most reliably identifies CT-confirmed subsidence, and whether a positive intraoperative CT meaningfully predicts later radiographic subsidence. Objective: Using intraoperative CT as reference, we aimed to (1) determine the optimal X-ray protrusion depth threshold for CT-confirmed early subsidence; (2) test whether intraoperative CT predicts late radiographic subsidence; and (3) examine how early X-ray depth relates to intervertebral disc height (IVDH) and segmental lordosis (SL) loss. Methods: In a retrospective single-surgeon cohort (March 2015–July 2023), subsidence was defined as ≥2.0 mm endplate penetration on CT and measured on X-ray by parallax technique. Sensitivity, specificity, accuracy, and Cohen’s κ were calculated. Receiver operating characteristic (ROC) analysis evaluated X-ray depth as a continuous predictor and identified the Youden-optimal cutoff. Intraoperative CT was tested against late radiographic subsidence; no-intercept linear models estimated per-millimeter IVDH and SL loss. Results: Of 100 patients, 93 had paired imaging (mean age 66.7 years; body mass index 26.8 kg/m²). Subsidence appeared on CT in 16.1% and on X-ray in 15.1%. X-ray showed 80.0% sensitivity, 97.4% specificity, 94.6% accuracy, and κ = 0.80; ROC analysis demonstrated strong discrimination (area under the curve 0.91; 95% confidence interval 0.81–1.00), Youden-optimal cutoff 1.90 mm. Intraoperative CT predicted late subsidence (n = 76) with only 45.8% sensitivity and 96.2% specificity; missed cases had penetration depths indistinguishable from non-subsiders. Each 1 mm of early X-ray depth corresponded to 0.45 mm IVDH and 0.37° SL loss. Conclusions: An X-ray protrusion depth of 2.0 mm reliably identifies CT-confirmed early subsidence, providing a preliminary diagnostic cutoff for use when CT is unavailable. Intraoperative CT is highly specific but insensitive for late subsidence; meaningful risk stratification will require additional inputs. These hypothesis-generating findings warrant prospective validation. Full article