搜索结果：Physical and engineering sciences in medicine

To evaluate the dosimetric impact of depth-dependent ion recombination and empirical effective point of measurement (EPOM) positioning in megavoltage photon beams, with particular focus on flattening filter-free (FFF) beams. Ion recombination correction factors ([Formula: see text]) were characterised as a function of depth and field size for three ionisation chambers (Roos, SNC125, CC13) using the two-voltage method under reference conditions (SSD 100 cm and 10 cm field size) and for additional MLC-defined 5 cm and 2 cm square field sizes, on a point-by-point basis across multiple beam energies. Empirical EPOMs were derived by aligning percentage depth ionisation (PDI) curves to a reference plane-parallel chamber. The dosimetric consequences of using generic [Formula: see text] and EPOM assumptions were assessed, and scan-derived [Formula: see text] values were validated against point dose measurements. A marked depth dependence in [Formula: see text] was observed for all chambers, most notably in FFF beams. The CC13 exhibited the greatest depth-related variation, resulting in recombination-related PDD deviations of up to 1.3% at extended depths. Empirically determined EPOMs were consistently smaller than the conventional 0.6[Formula: see text] shift, with normalised values of 0.42 and 0.38 for SNC125 and CC13, respectively. Using the conventional shift would introduce a residual dose deviation of approximately - 0.5%. The combined influence of uncorrected ion recombination and the generic EPOM produced a PDD bias of 0.8% at 10 cm depth for the 10 FFF beam, which is relevant both for reference dosimetry and for depth-sensitive treatment sites. This study demonstrates that empirical, chamber-specific EPOM and [Formula: see text] correction factors improve dosimetric accuracy and PDD measurements, and consequently, reference dosimetry and TPS beam modelling for which PDD10 cm is a key parameter. With the increasing adoption of FFF beams, reliance on generic assumptions for [Formula: see text] and EPOM introduces clinically relevant systematic deviations, approaching 1.0% at the calibration depth and becoming larger at greater depths. These corrections should be considered an essential component of linac and chamber commissioning to ensure robust reference dosimetry and accurate beam modelling. Given their measurable impact, such practices warrant inclusion in ACPSEM guidelines, in alignment with emerging best-practice frameworks and the evolving precision requirements of modern radiotherapy.

Dementia is a progressive neurodegenerative disorder that severely impacts cognitive functions and daily living, especially in aging populations. Among its subtypes, Alzheimer's disease (AD) and frontotemporal dementia (FTD) exhibit overlapping clinical symptoms, making early and accurate differentiation a critical challenge. Electroencephalography (EEG), as a non-invasive and cost-effective modality, provides valuable insights into the neurophysiological disruptions associated with these conditions. This study aims to develop a robust EEG-based diagnostic framework capable of accurately classifying AD, FTD, and healthy controls (HC) by integrating domain-specific signal processing with advanced deep learning techniques. This study employed a publicly accessible dataset consisting of resting-state EEG recordings from a total of 88 participants, comprising 29 individuals with AD, 23 diagnosed with FTD, and 36 age-matched HC. The proposed model integrates Common Spatial Pattern (CSP) filtering with a sequential modified hybrid architecture that combines Convolutional Neural Networks (CNNs) and a Vision Transformer (ViT). By fusing domain-informed spatial filtering with deep hierarchical feature learning, the model captures both local signal characteristics and global contextual dependencies. A 10-fold cross-validation approach was employed to assess model performance and generalizability. The proposed model achieved notable classification accuracies of 95.86%, 94.76%, 94%, and 92.14% for the AD/HC, FTD/HC, AD/FTD, and AD/FTD/HC classification tasks, respectively. These results underscore the diagnostic potential of EEG-based deep learning frameworks in distinguishing among neurodegenerative conditions and highlight their promise in supporting more precise and individualized clinical interventions. This study presents a novel end-to-end EEG classification pipeline that fuses domain-guided spatial filtering with deep neural feature learning. The promising results suggest that the proposed method could serve as a valuable component in future clinical decision support systems for dementia, contingent upon further validation in real-world clinical settings.

Rheumatoid arthritis (RA) is a chronic inflammatory disease characterized by pain, swelling, stiffness, and loss of joint function, making early diagnosis challenging. The study aims to assess the differences between RA patients (n = 70) and healthy individuals (n = 30) while classifying Ritchie Articular Index (RAI) values (0-3) based on inflammation levels using artificial intelligence algorithms. Metacarpophalangeal (MCP), and proximal-interphalangeal (PIP) joints were analyzed for the degree of inflammation. Static thermal data was collected from individuals at rest in a controlled environment. Then, alcohol was applied to the participants' hand regions, followed by a 180-second thermal video recording of the same region. In the pre-processing step, background noise cleaning and alignment were performed. Background was eliminated using Snake algorithm. Thermal video recordings were aligned using Scale Invariant Feature Transform (SIFT) algorithm. The Skeletonization algorithm was employed to detect fingers and joint regions in the images. For static thermal analysis, initial temperature ([Formula: see text]) values were extracted from the resting thermogram data. In dynamic thermal analysis, the temperature parameters [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text] were calculated. A statistical analysis of the four temperature parameters across different RAI values revealed that [Formula: see text] (p = 0.025) and [Formula: see text] (p = 0.042) exhibited statistically significant differences among the four RAI levels. Machine learning models were trained using the resting temperature values of patient and healthy groups, and the SVM achieved the highest success rate of 93%. It is believed that the proposed system may help diagnose RA in clinical settings and contribute to determining the severity of inflammation.

The demand for bedside radiography is increasing due to critical clinical needs, including infection control and the limited mobility of severely ill patients. However, radiation dose adjustment in these settings remains heavily reliant on the expertise and experience of radiographers. To address this issue, a novel flat panel detector (FPD) integrated with an automatic exposure control (AEC) system has been developed. This study aims to experimentally evaluate the fundamental performance of this system and clarify its clinical utility, including its potential limitations. The dependency of the AEC performance on object thickness and tube voltage was investigated using acrylic phantoms. To simulate clinical scenarios, the AEC response was examined using a chest phantom. Additionally, the effects of source-to-image distance and oblique X-ray incidence on the AEC performance were also evaluated using a quality-control test device. Our results elucidated the behavior of the exposure index (EI) and image quality under varying tube voltage and object thickness. In clinical conditions, the introduction of the AEC system significantly reduced EI, confirming its potential for effective dose management. Multiple factors were identified that influence both the AEC response and image quality, such as sensor positioning, imaging distance, and beam angle. These findings demonstrate that the AEC-equipped FPD system maintains consistent image quality while effectively reducing the radiation dose under various simulated imaging conditions. Our results also underscore the importance of accounting for environmental factors that affect dose control and image characteristics, highlighting the need for practical adjustment in routine clinical operation.

Image-guided radiotherapy (IGRT) has enhanced the precision of cancer treatment by integrating imaging modalities such as computed tomography (CT), magnetic resonance imaging (MRI) and cone-beam computed tomography (CBCT) into daily radiotherapy workflows. In head and neck cancer, where anatomical changes are common, accurate image registration between planning and treatment scans is essential to ensure dose accuracy. However, geometric distortions in CBCT (such as translation, rotation, and scaling resulting from patient positioning variations observed in daily CBCT images) can affect tumour targeting and dose delivery. This pilot study assesses a MATLAB-based image correction algorithm that uses rigid bony landmarks and point cloud registration together with spatial transformation to align CBCT with planning CT. Two head and neck cancer patients were retrospectively analysed, selected for their contrasting anatomical responses: one with substantial tumour regression and one with minimal change. Imaging was performed on the Halcyon V3.1 linear accelerator (Varian Medical Systems), with 25 daily CBCT scans per patient (85–96 slices per scan), resulting in 50 datasets for analysis. Spatial deviations were measured along the X, Y, and Z axes, and dose recalculations were performed for each treatment fraction. The correction method significantly improved spatial congruence and reduced geometric discrepancies caused by voxel spacing and acquisition parameters. Uncorrected scans showed dose deviations of up to ± 12% in organs at risk, notably the spinal cord and parotid glands. These findings demonstrate the feasibility and dosimetric relevance of automated CBCT correction in daily head and neck radiotherapy. Although limited in sample size, the study provides a detailed technical and dosimetric analysis of spatial distortions and supports future validation in larger patient cohorts.

Knowledge-based planning (KBP) has emerged as a promising approach to improve the consistency and efficiency of radiotherapy by utilizing prior clinical data to guide treatment planning. This study aimed to evaluate the performance of various machine learning (ML) algorithms for predicting key dosimetric parameters in 3D conformal radiotherapy (3D-CRT) of left-sided breast cancer, with the goal of reducing manual trial-and-error during plan optimization. A retrospective dataset of 75 breast cancer patients treated with 3D-CRT was used. Geometric and dosimetric features, including beam and organ-at-risk (OAR) parameters, were extracted from clinically approved plans. Ten supervised regression algorithms, including SVR, LASSO, Ridge, AdaBoost, Gradient Boosting Regressor, Histogram-Based Gradient Boosting, KNN, MLP, SGD, and Kernel Ridge Regression, were trained to predict dosimetric outcomes such as D_mean and V_x for the heart and lung, and homogeneity index (HI). Data preprocessing included outlier capping, quantile-based oversampling, and MaxAbs scaling. Model performance was evaluated using fivefold cross-validation and an independent test set, employing RMSE, MAPE, and MedAE as metrics. Among the evaluated algorithms, the KNN model demonstrated the consistent predictive performance across all dosimetric endpoints, achieving the lowest RMSE and MAPE for D_mean Heart (2.09 Gy, 0.17) and D_mean Lung (3.12 Gy, 0.15). Feature importance analysis identified geometric parameters such as borders, wedge angles, and OAR volumes as the most influential predictors. ML-based KBP can accurately predict dosimetric outcomes prior to dose calculation, improving planning efficiency and consistency in breast 3D-CRT. The KNN algorithm showed the highest reliability, suggesting its suitability for integration into clinical decision-support systems.

Tricuspid regurgitation (TR) is a prevalent and progressive condition associated with heart failure symptoms; however, numerous patients are unsuitable for surgery, and achieving a long-term outcome is challenging. This renders TR an unmet clinical problem. Computational modeling facilitates the evaluation of the biomechanics associated with regurgitation, aiming to enhance therapeutic strategies as transcatheter edge-to-edge repair. This study sought to establish a computational framework integrating a tricuspid valve (TV) model with the right ventricle (RV) wall to evaluate native valve function, TR pathology, and repair using the MitraClip device. Computed tomography angiography (CTA) from three patients was utilized to reconstruct the tricuspid valve leaflets, annulus, and RV wall. A parametric CAD pipeline for modeling the leaflet surfaces and chordal architecture was built using ex-vivo data. Annular kinematics (systole-to-diastole) were computed and then applied as a boundary condition in the finite-element analyses, with tricuspid regurgitation induced by displacing papillary positions and augmenting transvalvular loading. The MitraClip deployment was subsequently simulated whereas the post-deformation hemodynamics were assessed using a lattice-Boltzmann solver to quantify intraventricular velocity fields and vorticity. All three models demonstrated physiological leaflet coaptation and stress distributions in a healthy patient condition, with maximum stresses at end-systole quantified in the leaflet belly and chordal insertions. Post-MitraClip simulations demonstrated restored valve closure without retrograde flow during systole; however, flow exhibited a double-orifice inflow and localized vorticity near the clip during diastole. The jet orientation and downstream patterns depended on patient anatomy. This computational framework replicates TV biomechanics and post-repair hemodynamics from standard imaging, facilitating quantitative, patient-specific evaluation. The current patient-specific, image-based computational method can predict TV-related performance and post-repair hemodynamic outcomes, facilitating personalized planning to minimize residual TR and enhance device placement.

The early diagnosis of Parkinson's disease (PD) using SPECT imaging continues to be challenging due to the subtle dopaminergic deficits present in the early stages of the disease. This study proposes a novel hybrid approach combining conventional and deep learning features to improve PD classification, and applies it to reclassify scans without evidence of dopaminergic deficit (SWEDD) cases. We used SPECT images from early PD patients and healthy controls (HC) and extracted SBR metrics, morphometric, and deep learning features. Our multi-stage feature selection pipeline employed near-zero variance filtering, ANOVA F-test analysis, correlation-based feature elimination, and Random Forest importance scoring. We evaluated multiple machine learning algorithms and selected Linear Discriminant Analysis as the optimal classifier, then applied this model to reclassify 79 SWEDD cases. Feature selection reduced 79 significant features to 15 optimal features: 1 SBR metric (6.7%), 7 morphometric (46.7%), and 7 deep features (46.7%). The hybrid Linear Discriminant model achieved the best performance, outperforming individual feature approaches with 97.40% test accuracy, 96.25% sensitivity, 98.65% specificity, and 99.59% AUC. Statistical analysis revealed morphometric features had the highest mean importance (0.0699 ± 0.0539), followed by deep (0.0400 ± 0.0570) and SBR features (0.0206 ± 0.019). SWEDD reclassification identified 5 cases (6.3%) with imaging patterns consistent with early PD, while 74 cases (93.7%) maintained HC characteristics. This study presents a proof-of-concept demonstration of the effectiveness of integrating conventional measures with deep learning techniques for improving the early diagnosis of PD, while offering new insights into SWEDD case reclassification.

A treatment planning system (TPS) is responsible for calculating the radiation dose for patients undergoing brachytherapy. However, to verify TPS dose accuracy of intracavitary brachytherapy, which feature particularly steep and complex dose gradients, 3D-printed phantoms made of polylactic acid (PLA) can be used. A study was designed to create an in-house phantom for verification of gynecological brachytherapy measurement using a radiophotoluminescent glass dosimeters (RPLGDs) and to evaluate the dosimetric differences between measurement and calculation by the treatment planning system under clinical conditions.An in-house phantom holder was designed to move the axis of the holder to the rectum point that differs according to the patient's anatomy. The holder of the applicator was designed for various types of applicators in intracavitary brachytherapy. This clinical study was used to quantify variations between the calculated and measured dose for 6 plans at various points in the phantom, which included point A, point B, the bladder point, and the rectum points.The RPLGDs demonstrated a linear dose response up to 10 Gy, excellent angular dependence, and an associated uncertainty of 3.3% (k  =  1). In the clinical case, the dose differences between the measured and calculated values at Point A, Point B, bladder, and rectum were +1.99  ±  1.11%, 1.01  ±  0.02 Gy, and 0.10 Gy, +4.42 ± 2.56%. and + 3.53  ± 1.44%, respectively.Dosimetry with RPLGDs using the 3D printed in-house phantom can accurately verify delivered dose in intracavitary brachytherapy for quality assurance purposes.

The complex dose delivery mechanisms of TomoTherapy (TOMO) demand rigorous patient-specific quality assurance (PSQA). This study systematically evaluates the relationship between ArcCHECK measurements and GPU-accelerated Monte Carlo (GPU-MC) calculations for nasopharyngeal carcinoma (NPC) TOMO plans across multiple gamma criteria, aiming to delineate their respective strengths and inform an optimized verification strategy. A retrospective analysis was conducted on 317 TOMO plans for NPC, each optimized using the Accuray Precision Treatment Planning System. Patient-specific dose verification was performed using ArcCHECK measurements and independent MC calculations implemented through PlanQA. Gamma passing rates (GPRs) were evaluated under nine different conditions, including both same-criterion and cross-criterion comparisons. To assess differences, agreement, and correlations between methods, statistical analyses were conducted using the Wilcoxon signed-rank test, Bland-Altman analysis, and Spearman correlation. Under identical Gamma criteria, there was no statistically significant difference in GPR between ArcCHECK and MC. However, cross-criterion comparisons revealed marked discrepancies, highlighting the criterion-dependent nature of GPR outcomes. Lenient standards typically exhibit good consistency and relatively minor deviations. Furthermore, the correlations among all combinations of these standards can be considered negligible. The comparable performance of GPU-MC and ArcCHECK under conventional criteria (3%/3 mm, 3%/2 mm) validates ArcCHECK's established role in verifying delivery fidelity. Crucially, under the stringent 2%/2 mm criterion where ArcCHECK passing rates decline, MC provides critical diagnostic power to differentiate between discrepancies originating from the dose calculation algorithm and those arising from the physical delivery process. For TOMO PSQA, GPU-MC and ArcCHECK are complementary. An integrated approach leveraging both methods is therefore recommended.

Brain clocks are promising tools for evaluating brain health. However, most current methods rely on structural neuroimaging. Functionally based approaches remain scarce, especially for assessing age-related neurodegenerative diseases. This study examines whether the brain age gap (BAG), the difference between chronological and predicted brain age, reflects neurodegeneration when estimated from electroencephalographic resting-state (rsEEG) α-oscillations, a well-established marker of brain functional aging. It also explores whether α-based brain clocks reflect sociodemographic diversity and structural inequality. The BAG was computed using spectral descriptors of α-activity in the rsEEG source space of 1228 healthy participants, individuals with mild cognitive impairment (MCI), and patients with Alzheimer's disease or behavioral variant frontotemporal dementia, residing in 10 countries with varying levels of structural inequality. BAGs are increased in MCI and dementia groups, particularly in posterior cortical regions. Structural inequality emerges as the strongest predictor of BAG, surpassing cognition, education, and sex. The findings indicate that an α-oscillation-based brain clock provides a sensitive functional marker of brain aging, capable of capturing neurodegenerative processes as well as the impact of social disparities. This scalable, accessible approach to brain health shows promise for translational use and population-wide screening in underserved, resource-limited settings.

Breast cancer is a significant global health issue, demanding early identification to provide appropriate therapy and satisfactory survival results. This work used two independent imbalanced datasets (EIS-BT and WBCD). CNN1D with Long Short-Term Memory (LSTM) was integrated to acquire features from these datasets to identify breast cancer. Three scenarios for breast cancer detection were investigated based on CNN1D-LSTM derived characteristics from Dataset-1, Dataset-2, and their combination. The Synthetic Minority Over-sampling Technique (SMOTE) was used to balance the collected features in all three scenarios. The suggested CNN1D-LSTM-SMOTE approach, in conjunction with Support Vector Classification (SVC), yields impressive results with a Matthews Correlation Coefficient (MCC) of 97.2 % on Dataset-1 and 100.0% on Dataset-2. Random Forest Classifiers (RFC) perform better, achieving an MCC of 98.4% on the combined features. The K-fold approach was used, yielding average MCCs of 91.7%, 74.1%, and 96.9% on Dataset-1, Dataset-2, and the combined features, respectively. Statistical analysis revealed a p-value of 0.01, signifying the significance of the findings, and a standard error of 0.006 for the combined features. Bootstrapping was employed to calculate confidence intervals, resulting in Lower Confidence Intervals (LCI) of 95.6% and Higher Confidence Intervals (HCI) of 97.9% for the combined features. These findings highlight the model's potential clinical application in supporting oncologists with early, real-time, reliable, and automated breast cancer diagnosis, leading to improved diagnostic procedures and patient outcomes.

Radiotherapy dosimetry in composite and modulated fields remains complex, especially when using small field ionization chambers in the second part of the Alfonso et al. formalism. This study investigates the response of the IBA- CC 01 ionization chamber in machine-specific reference, one static field of, and clinical IMRT step-and-shoot composite fields for a 6 MV flattening-filter-free (FFF) TrueBeam STx ® photon beam. A previously validated BEAMnrc model of the TrueBeam linac was used to generate high-statistics phase-space files, which were then combined with an egs_chamber model of the IBA-CC01 to calculate absorbed dose to water and detector response in static and composite fields. Latent variance was evaluated at three fixed points (central, off-axis, and peripheral) across several IMRT step-and-shoot fields, showing that the detector's latent variance remains below 0.2% and is largely independent of the detector's position. Radiochromic film measurements using Gafchromic EBT4 in a solid water phantom, following AAPM TG-235, validated the Monte Carlo simulation of the plan-class-specific reference fields. For these, off-axis ratio profiles from film and Monte Carlo agree within a few percent in the high-dose region, and a gamma analysis with 3.5%/2.5 mm criteria (global) yielded passing rates of 97% and 95% for cross-planes and in-plane profiles, respectively. Monte Carlo-derived correction factors for the IBA-CC01 in IMRT step-and-shoot composite fields are close to one and, within a mean absolute difference of less than 1.5%, align with the small field correction factors reported in IAEA-AAPM TRS 483 for static fields of similar size. These findings suggest that, for the 6 MV FFF TrueBeam beam and the IMRT step-and-shoot fields examined here, the IBA-CC01 functions effectively as a practical field detector for relative dosimetry and for calculating detector-specific correction factors in composite fields. In contrast, absolute reference dosimetry should still rely on reference-class ionization chambers under conventional reference conditions.

Children are more radiosensitive than adults, making dose optimisation in paediatric computed tomography (CT) essential. Although Diagnostic Reference Levels (DRLs) are internationally recommended as optimisation tools, national DRLs for paediatric CT in Jordan remain limited and outdated. This study aimed to establish national DRLs for six paediatric CT protocols, evaluate inter-hospital and age-related dose variations, and compare results with international benchmarks. A retrospective multicentre study was conducted across six Jordanian hospitals between December 2023 and September 2025, including 3794 paediatric patients stratified into four age groups (<&#x2009;1, 1-4, 5-10, and 11-18&#xa0;years). Volumetric CT dose index (CTDIvol) and dose-length product (DLP) data were collected for six protocols: brain, chest, abdomen-pelvis, chest-abdomen-pelvis, sinuses, and contrast-enhanced neck soft tissue. DRLs were defined as the 75th percentile of institutional median CTDIvol and DLP values. Inter-hospital and age group variations were analysed, and univariable regression analyses assessed acquisition parameters associated with dose variation. Dose increased with patient age for trunk protocols, whereas brain, sinuses, and neck CT showed comparatively stable patterns. Substantial inter-hospital variability was observed across protocols, with institutional median CTDIvol differing markedly between centres, particularly for trunk examinations in younger age groups. In univariable regression analyses, all four acquisition parameters (kVp, mAs, pitch, and slice thickness) were significantly associated with CTDIvol and DLP (p&#x2009;&#x2264;&#x2009;0.001), with kVp demonstrating the strongest association (R2&#x2009;=&#x2009;0.603 for CTDIvol; R2&#x2009;=&#x2009;0.630 for DLP). Compared with published international DRLs, Jordanian brain CT dose metrics were higher in multiple age groups; chest comparisons should be interpreted cautiously where international benchmarks are weight-banded. This study established national paediatric CT DRLs for Jordan, highlighting the need for standardised protocols and periodic DRL review to enhance radiation protection.

This study presents the first head-to-head comparison of Manual optimization, knowledge-based planning (KBP; RapidPlan), and feasibility DVH-guided planning (FDVH; PlanIQ) for prostate volumetric-modulated arc therapy (VMAT). While previous reports have evaluated knowledge-based or feasibility-based planning separately, no prior work has directly compared all three approaches within the same patient cohort. In this investigation, treatment plans were generated for 12 patients using each strategy (36 nominal plans in total), and plan quality, stability under nominal conditions, and robustness under simulated isocenter shifts were systematically evaluated. A total of 7020 re-calculated plans were generated, reflecting the scope of the study. The results demonstrated that Manual planning achieved superior target homogeneity, with significantly lower D2% and higher D98% compared with RapidPlan, while D50% values were also lower, indicating a modest reduction in mid-dose levels. PlanIQ exhibited intermediate characteristics, offering modest sparing compared with Manual but lower target homogeneity. Importantly, this work explicitly distinguished stability, defined as low interpatient variability under shifted conditions, from robustness, defined as plan performance against geometric uncertainties for individual patients. Evaluating these dual aspects simultaneously for the first time provides novel insights beyond previous KBP or FDVH studies. The findings suggest that RapidPlan is advantageous in anatomically challenging cases with close PTV&#x2013;rectum proximity, whereas Manual optimization may remain preferable when dose homogeneity is prioritized. PlanIQ may also serve as a supplementary tool for planner training and standardization. Overall, this study provides new evidence and practical guidance for balancing plan quality, stability, and robustness in prostate VMAT, with clinical and educational implications.

PURPOSE: High Stability of radiomic features is critical for developing reliable imaging biomarkers that can support risk stratification, treatment response assessment, and personalized therapy in lymphoma patients. To evaluate how partial volume correction (PVC) affects the Stability of 18F-FDG PET radiomic features in lymphoma lesions, with respect to lesion volume and tissue type. METHODS: This single-center retrospective study included 131 newly diagnosed lymphoma patients (2014&#x2013;2024) who underwent baseline 18F-FDG PET/CT. In total, 1,603 lesions (1,302 lymph nodes, 117 spleen/liver, 150 bone, and 34 bone and soft-tissue) were semi-automatically segmented and grouped by volume (<&#x2009;3, 3&#x2013;10, 10&#x2013;30, >&#x2009;30 mL) and tissue type. Ninety-three radiomic features were extracted from non-PVC and PVC images processed with the Richardson&#x2013;Lucy (RL) and Reblurred Van Cittert (RVC) algorithms after isotropic resampling (3&#xa0;mm) and discretization (0.25 SUV bin size), following IBSI guidelines. Stability was quantified using the coefficient of variation (CoV) and the intraclass correlation coefficient (ICC2, absolute agreement), with statistical comparisons performed via Mann&#x2013;Whitney U tests and false discovery rate (FDR) correction. RESULTS: PVC significantly improved feature Stability, particularly for large lesions (>&#x2009;30 mL), with median ICC2 > 0.90 across most feature categories (e.g., First-Order&#x2009;=&#x2009;0.99, GLSZM&#x2009;=&#x2009;0.97, NGTDM&#x2009;=&#x2009;0.97). Small lesions (<&#x2009;3 mL) showed lower stability (ICC2 = 0.84&#x2013;0.94) and higher CoV (0.09&#x2013;0.21), mainly in texture-based features. First-Order and GLCM features were the most robust overall (ICC2 = 0.92&#x2013;0.99; CoV&#x2009;=&#x2009;0.07&#x2013;0.11). Bone and spleen lesions exhibited the highest Stability (median ICC2 &#x2248; 0.95), whereas lymph node and liver features were more variable. All volume- and tissue-dependent differences remained significant after FDR correction (p&#x2009;<&#x2009;0.05). CONCLUSION: PVC using RL and RVC markedly enhances FDG-PET radiomic Stability in lymphoma, particularly for larger and structurally uniform lesions. Robust features such as First-Order and GLCM can support standardized radiomics workflows and the development of reliable biomarkers for prognosis and personalized therapy. Additionally, PVC reduces variability in texture features, especially in small or heterogeneous lesions. Multicenter validation would further strengthen generalizability beyond this single-center setting.

Brain tumors present significant health challenges and necessitate early diagnosis due to their high mortality rates. Diagnosis through Magnetic Resonance Imaging (MRI) requires specialized expertise and remains susceptible to error. Consequently, the demand for automated diagnostic systems continues to grow. In response, this study proposes a novel Deep Learning (DL) model for brain tumor classification. A publicly available Figshare dataset containing 3064 T1-weighted contrast-enhanced brain MRI images representing three tumor types was used. The classification performance of fifteen DL architectures was initially evaluated to determine the most effective backbone. EfficientNetV2 demonstrated superior results and was selected for further development. An attention-based MLP-Mixer architecture was then integrated with EfficientNetV2 to improve classification performance. The final model&#x2019;s performance was comprehensively compared with other DL models and established methods in the literature. Grad-CAM visualization was applied to interpret and validate the model&#x2019;s decision-making process. The proposed model was evaluated using stratified five-fold cross-validation, achieving 98.53% accuracy, 98.37% precision, 98.48% recall, and 98.42% F1-score. These results demonstrate superior performance relative to previous studies. Additionally, Grad-CAM visualizations show that the model consistently focuses on relevant regions within MRI images, thereby improving interpretability and clinical reliability. The integration of EfficientNetV2 with an attention-based MLP-Mixer resulted in a robust DL model for clinical decision support systems, providing high accuracy and interpretability in brain tumor classification.

The detection of schizophrenia using neuroimaging modalities remains challenging due to the complex brain structures and subtle pathological changes in the brain. In this work, we propose a novel hybrid feature extraction approach using a fusion of wavelet-based handcrafted features and dense deep features extracted from structural MRI (sMRI) scans to achieve a robust computer-aided diagnosis system for the accurate detection of schizophrenia. The sMRI scans are first pre-processed using median filtering technique to mitigate noise and ameliorate image quality. Following pre-processing, we extract handcrafted features using discrete wavelet transform technique which results in wavelet approximation and detail coefficients capturing the spatial and frequency components from sMRI data. Further, we employ principal component analysis to reduce data dimensionality of wavelet coefficients and improve the computational efficiency. Additionally, we extract deep features from the linear orthogonal transformations of sMRI data using modified CNN architecture with dense connectivity. The hybrid feature set is then used as input to three machine learning classifiers- k-nearest neighbor, support vector machine, and random forest. The proposed approach achieves exceptional performance, with k-nearest neighbor classifier, achieving 99.2% accuracy on stratified 10-fold cross validation. The results indicate a highly reliable and accurate computer-aided diagnosis system integrating wavelet-based and dense deep orthogonal features extracted from sMRI scans for the detection of schizophrenia.

Seismocardiography (SCG) utilises an accelerometer to monitor the heart's vibrations. The accelerometer's compact size and light weight make it suitable for wearable heart monitoring. However, motion artefacts from walking or subject movement contaminate the signal in both the time and frequency domains, making monitoring challenging. Monitoring during activity provides deeper insight into cardiovascular function as blood flow demands increase. Existing denoising methods often suppress heart sounds along with high-amplitude motion noise. In contrast, this paper introduces a novel approach to classify and eliminate noise components without affecting the heart sounds, thereby enhancing segmentation accuracy. A triaxial accelerometer is strategically positioned on the chest to capture SCG (z-axis) and motion noise from walking (x-axis). An adaptive Yen's threshold method is developed to detect peaks in the SCG signal. Machine learning models are then trained using time- and frequency-domain features to classify these components, enabling the removal of noise peaks from the SCG signal. Clustering-based post-processing further improves the segmentation. The decision tree model achieves 99% accuracy in identifying Fundamental Heart Sounds (FHSs) in healthy subjects. These findings suggest that, with further clinical validation across diverse patient populations, the proposed approach could revolutionize home monitoring, offering a more accurate and insightful analysis of cardiac activity during daily activities.