Stereotactic body radiation therapy (SBRT) is a well-established standard of treatment for non-small cell lung cancer patients. However, SBRT is a double-edged sword and a high degree of accuracy is required for dose calculations. In the presence of inhomogeneity, it should be below 3% and depends strongly on the accuracy of the algorithm used for dose calculation in treatment planning system (TPS). The aim of this study is to evaluate the accuracy of Collapsed Cone (CC) algorithm for dose calculation in the presence of lung inhomogeneity compared to the Monte Carlo (MC) algorithm as the reference. The study conducted on a wide range of photon fields from 0.5×0.5 cm2 to 10×10 cm2. Monaco® TPS equipped with CC and MC algorithms in 6 MV, 10 MV, and 18 MV photons was employed and the inhomogeneity correction factor (ICF) and PPDs were calculated inside an in-house inhomogeneous phantom. Finally, the results were compared between two algorithms. For 6 MV photons, gamma index pass rate (GIPR) with (2 mm, 3%) criteria in a 0.5×0.5 cm² field is only 71.3%. The CC algorithm won't be able to confidently reproduce the PDDs in this field. For 1×1 cm², 2×2 cm², and 3×3 cm² fields, the accuracy of the CC algorithm can increase to above 98%. For SBRT of small lung lesions, the results recommend increasing the minimum segment width to > 0.5 cm in IMRT plans if CC algorithm is used. Independent of field size, the accuracy of CC algorithm decreases with increasing energy. For 10 MV and 18 MV photons, average GIPR under (2 mm, 3%) criteria was 97.9% and 92.0%, respectively, which are not of clinical interest.
High-density metallic objects often produce missing or severely corrupted projection data, resulting in strong streak artifacts in X-ray CT images. In this paper, we propose a path-length-aware sinogram completion framework that combines interpolation in the mean attenuation domain with direct enforcement of low-order Helgason-Ludwig consistency conditions. The missing projections are first estimated using a Hermite-type interpolation that explicitly accounts for X-ray path length. Moment consistency is then restored through a probabilistic perturbation scheme based on compactly supported probability density functions, avoiding the need for solving a large-scale constrained optimization problem. Numerical experiments using both a simulated phantom and a dental CT model demonstrate that the proposed method achieves improved sinogram consistency and better image quality than LI-MAR and NMAR while preserving important anatomical structures. The proposed framework provides an efficient and physically motivated approach for metal artifact reduction in X-ray CT.
Medical image segmentation aims to accurately delineate organs, tissues, or lesion regions from complex medical images. However, existing hybrid models based on Transformers and convolutional neural networks (CNNs) still suffer from limitations in local detail modeling and cross-layer feature fusion, which often leads to blurred boundary information and loss of structural details. To address these issues, this paper proposes a Cross-layer Semantic Alignment and Context Enhancement Network (CSAE-Net) for medical image segmentation. Specifically, a semantic enhancement module is introduced into the skip connections to achieve effective fusion of high-level semantic information and shallow spatial details through spatial-channel collaborative modeling and multi-scale context extraction. In addition, a lightweight boundary refinement mechanism is employed in the decoder stage to improve the recovery capability for complex boundary regions. Experiments conducted on the Synapse, ACDC, and GlaS datasets demonstrate that the proposed method outperforms mainstream approaches in terms of Dice, HD95, and IoU metrics, validating the effectiveness of the cross-layer semantic alignment mechanism for complex medical image segmentation tasks.
Scattered coincidences are typically rejected in conventional imaging in Positron Emission Tomography (PET) despite carrying potentially useful spatial information. We present a novel algorithm for PET imaging from single-scattered (inside tissue) events with known Time-of-Flight (TOF) and without energy information, particularly useful for plastic scintillator. The 3D annihilation loci corresponding to individual single-scattered (SS) events are modeled as spindle-torus probability distributions constrained by TOF. These event-wise probability volumes are merged using a MATLAB-based reconstruction framework to estimate the source location. In this study, we used GATE-based Monte Carlo simulations of PET acquisitions to generate realistic emission-event data, including true, SS, and multiple-scattered coincidences. Supervised machine learning classifiers, namely Random Forest (RF) and Extreme Gradient Boosting (XGB), were employed only for the identification of SS events from event-level geometric and timing-related features extracted from the simulated data. The identified SS events were subsequently processed using a dedicated 3D geometric TOF-based reconstruction framework to estimate the annihilation probability distribution and generate SS images. Both RF and XGB achieved SS-event precision of approximately 68-71%, with recall ranging from 56-79% across the evaluated scanner configurations. Using the identified SS events, we created a 3D emission image and validated its quality in terms of resolution, contrast, and uniformity. The proposed framework was also evaluated on the NEMA Image Quality (IQ) phantom under realistic imaging conditions. The results demonstrate that SS events retain meaningful spatial information and that their combination with geometric TOF-based reconstruction provides a feasible pathway for scatter-aware imaging in plastic scintillator PET systems.
This work sought to improve bioimpedance-based tissue identification model performance on an eight-tissue dataset by employing frequency response function similarity metrics as feature generators and Bayesian neural networks for accurate uncertainty estimations. Inspired by structural dynamics, frequency response function similarity metrics were applied in a new context to generate features from mean baseline measurements, thereby extracting information of how unclassified measurements compare to previous ones. Additionally, Bayesian neural networks were constructed to directly estimate model parameter uncertainty effects on classification uncertainty. Finally, a stacking ensemble technique combined base model outputs to train a meta-learner for improving performance. Models trained with similarity metric features achieved higher mean accuracies and better tissue specific F1-scores than those trained with measurements. Bayesian neural networks with temperature scaling reduced the expected calibration error of the standard feedforward networks by 19% to 83%, indicating significant enhancement of uncertainty quantification. Ensembles achieved higher mean accuracies than base models, with maximum accuracies over 75%, and maintained the enhanced uncertainty quantification. The implementation of similarity metric inputs and Bayesian neural networks for bioimpedance-based tissue identification offered a clear improvement in mean accuracy and uncertainty quantification over traditional models using measurements only. This marks an essential step towards enabling bioimpedance as a feasible sensing option for real-time tissue identification. Attaining dependable bioimpedance-based tissue identification will provide a foundation for new technologies due to economical and implementation advantages over competing methods while the added uncertainty-awareness makes it an excellent candidate for medical applications because it can provide additional context for outputs to users.
PURPOSE:The aim of this study was to investigate if the same planning target volume margins should be used for the breast/thoracic wall and lymph nodes when treated simultaneously with hypofractionated radiotherapy in deep inspiration breath-hold (DIBH) and free breathing (FB).
Methods: For 80 patients (40 treated in FB and 40 in DIBH), nine anatomy sites were evaluated. Three were located in relation to the lymph node region and six in relation to the breast/thoracic wall. For three treatment fractions, a cone beam CT was acquired before and after the treatment and compared to the planning CT. Margins were calculated based on the inter- and intra-fractional uncertainties, ensuring 90% of the patients receiving 95% of the prescribed dose to the CTV. Margin calculation was done for the full patient data set as well as for a subgroup of patients with set-up errors below the clinical action level, excluding patients with set-up errors larger than 7 mm.
Results: For the full patient data set, margins of up to 14 mm were calculated. However, for patients with set-up errors below the clinical action level, the largest margins calculated for the lymph node region for treatments in DIBH were 8.5 mm, 8.7 mm and 6.6 mm in the vertical, longitudinal, and lateral directions respectively compared to 8.4 mm, 7.9 mm and 5.9 mm for treatments in FB. For the breast/thoracic wall, the largest margins were 5.9 mm, 6.4 mm and 5.5 mm for treatments in DIBH and 5.8 mm, 7.2 mm and 5.8 mm for treatments in FB in the vertical, longitudinal and lateral directions respectively.
Conclusion: Treatments in DIBH and FB can use the same margins. However, the lymph node region requires a larger margin than the breast/thoracic wall region, due to greater anatomical mobility.
This study focuses on the restricted accessibility for people who are unable to undertake lower limb resistance exercise independently. Many typical exercises rely on external equipment or assistance, which limits the options for many rehabilitation patients, such as those who have had a stroke, neuromuscular conditions, or surgery recovery. This study aimed to develop a low-cost, innovative 3D-printed exoskeleton knee brace (KB) for independent resistance training, targeting the knee extensors and flexors. The KB prototype was designed to provide comfort, safety, and a mix of lightweight and strength. Kinematic analysis was used to evaluate the forces generated by resistance bands, and the design underwent finite element analysis simulation to ensure its structural integrity. Results indicate high user satisfaction, specifically regarding the KB's lightweight design and perceived safety. Participants perceived that the brace provided appropriate resistance for targeting the quadriceps and hamstring muscle groups during knee extension and flexion exercises. Participants recommended design improvements, including adjustments to the strap and length. The study suggests that KBs are a suitable method for independent resistance training, and future research should focus on evaluating their effectiveness in rehabilitation patients who are unable to perform current lower limb activities.
Beam angle selection plays an important role in shaping dose distribution in fixed-field radiotherapy, yet additional beam placement in multi-beam settings is often empirically determined. This study proposes AutoBeams, a transparent dose-driven rule-based framework for generating six-beam configurations in left-sided breast fixed-field IMRT, and evaluates whether these configurations can produce clinically acceptable plans under a standardized commercial treatment planning system (TPS) optimization protocol. A retrospective cohort of 41 patients with left-sided breast cancer treated with six-beam coplanar fixed-field IMRT was included. For each patient, a clinical-angle reference (CR) was defined from the original clinical beam angles, and an AutoBeams configuration (AP) was generated using the same patient geometry and isocenter. CR and AP plans were generated and evaluated in Pinnacle under identical planning conditions, with beam-angle configuration being the only intended difference. AutoBeams sequentially selected beam angles from a predefined coplanar candidate set using dose-based scoring rules with angular separation constraints. AutoBeams generated beam configurations that differed from the clinical beam arrangements through limited angular adjustments. Under commercial TPS optimization, target dose metrics remained broadly comparable between AP and CR. AP showed lower heart dose metrics and spinal cord maximum dose, while contralateral lung low-dose exposure and contralateral breast mean dose increased, indicating OAR-specific trade-offs. In addition, AP plans demonstrated significantly lower total monitor units compared with CR (p< 0.001). Criterion-level evaluation showed that AutoBeams-generated configurations could generally be optimized into clinically acceptable plans. These findings support AutoBeams as a transparent and reproducible approach to beam-angle generation in left-sided breast fixed-field IMRT.
Cardiovascular diseases remain the leading cause of mortality worldwide, driving the rapid advancement of cardiac tissue engineering (CTE) as a strategy to develop effective therapeutic interventions. Among its notable achievements is the cardiac patch, a supportive scaffold designed to repair damaged tissue following myocardial infarction. This study selected alginate (Alg) due to its superior biocompatibility and printability for the fabrication of cardiac patches through extrusion-based 3D bioprinting, a novel technology for scaffold construction. To further enhance scaffold performance, graphene oxide (GO) nanosheets were incorporated into the Alg matrix to improve electrical conductivity, mechanical strength, cell adhesion, and printability. Bioinks comprising 8% Alg and different concentrations of GO (0, 0.05, 0.1, and 0.2 mg ml-1) were formulated and thoroughly assessed for their physicochemical characteristics and cytocompatibility. All formulations exhibited suitable rheological characteristics for 3D printing. Moreover, the addition of GO reduced strand thickness, increased pore size, and improved overall printability. Mechanical testing revealed an increase in tensile strength from 0.18 MPa to 0.79 MPa with the addition of 0.2 mg ml-1of GO, while electrical conductivity reached (0.3 ± 0.00) × 10-6S m-1, approaching the range required for CTE. MTT assays demonstrated a concentration-dependent cytotoxic effect, showing decreased viability at 0.1 and 0.2 mg ml-1GO, while scaffolds with 0 and 0.05 mg ml-1GO preserved high viability. Scanning electron microscopy further confirmed enhanced cell adhesion and filopodia extension in Alg-GO-1 (0.05 mg ml-1) compared to pure Alg scaffolds. The findings generally suggested that Alg/GO composites, particularly at optimized concentrations, hold significant promise as bioinks for CTE due to their favorable mechanical, electrical, and biological properties.
Background.Major depressive disorder (MDD) is a prevalent mental health condition with profound consequences on daily functioning and quality of life. Existing studies have focused on functional connectivity (FC), with limited investigation of effective connectivity (EC) and its frequency-specific manifestations within machine learning frameworks.New Method.This study proposes an automated MDD detection framework based on graph-theoretical features derived from two complementary connectivity measures: the phase slope index (PSI), representing directed EC, and the weighted phase lag index (WPLI), representing FC. Statistically selected features were extracted across five traditional frequency bands and evaluated using multiple machine learning classifiers.Results.PSI-based features showed a marginal advantage in accuracy and area under the curve (63.4% and 0.63) compared with WPLI-based features (62.8% and 0.60). These small differences cannot be interpreted as a meaningful performance difference between the two connectivity measures. The delta and beta bands, in both connectivity methods, consistently yielded the most discriminative results across both connectivity methods, with the theta band additionally discriminative for FC.Comparison with existing methods:Most existing studies rely on FC measures, whereas the proposed framework incorporates directed EC and systematically evaluates the discriminative contribution of individual frequency bands through graph-theoretical analysis.Conclusions.Directed EC and FC achieved comparable classification performance, but frequency-specific graph-theoretical analysis of EC could provide additional interpretability by capturing the directionality of disrupted brain communication in MDD, supporting the development of more neurophysiologically meaningful diagnostic tools.
The distinct qualities and broad spectrum of applications of carbon dots (CDs) make them an attractive class of carbon-based nanomaterials these are nanoscale materials, often fewer than 10 nanometres. Their interesting biological, electronics and chemical features make them useful for a variety of purposes like energy conversion, sensing, bioimaging and catalysis, plus several more. CDs are synthesised via either top-down or bottom-up techniques. The surface of CDs comprises modified oxygen, polymer-based or amino groups, which allow for an abundance of chemical modification. Rare earth elements (REE) are an ideal choice for doping with CDs, yielding a combination termed RE-CDs that can enhance luminescence characteristics, utility and quantum yields. By combing these two materials each of their properties will help in many aspects like technological and biomedical applications such as increased photo luminescence, targeted drug delivery, bio imaging of tumour cells, structure modification, etc in cancer studies the hybrid materials shows increased bio compatibility and low adverse effects which shows efficient cellular uptake and ROS for destroys cancer cells along that which acts a nanocarriers to deliver anti-cancer medications. In this review, we provide an in-depth analysis of the structure classification, synthesis methodologies, photoluminescent properties, and anticancer applications of CDs doped with REEs. Moreover, this review describes the current limitations and future outlooks to enhance the utilisation of RE-CDs in biomedical applications.
Predicting the acoustic response of coated elastic systems remains a challenging problem in nondestructive evaluation, particularly when interfacial mechanical contrast and layer thickness jointly govern complex dispersive behavior. In this purely theoretical study, generalized Rayleigh-wave propagation in dental restorative bilayer systems is theoretically investigated within a scanning acoustic microscopy framework. Four coating materials composite resin, amalgam, cobalt chromium alloy, and gold alloy are examined on glass-ionomer, silicate, and zinc-phosphate cement substrates. The Rayleigh critical angle and surface wave velocity are analyzed as functions of the normalized thicknessh/λTover the range 0-2. The results reveal two distinct propagation regimes separated ath/λT≈ 1. In the interaction-dominated regime, anomalous and non-monotonic dispersion emerges from strong layer-substrate mechanical coupling and is governed by a single dimensionless mismatch parameterξ= (EL/ES)/(ρL/ρS). In the asymptotic regime, dispersion vanishes as propagation becomes layer controlled. A four-type acoustic classification (Types A, B, C, D) is established from the joint analysis ofξ, the anomaly severity index, and the velocity sensitivity index (VSI), with a near unity linearξ-VSI correlation (R2= 0.98). The natural enamel-dentin bilayer (ξ= 3.31, VSI=+79.8%) is identified as the acoustic reference target for restorative design. These findings provide a unified theoretical framework for evaluating acoustic compatibility in dental restorative assemblies, offering potential theoretical guidance pending experimental validation.
Alzheimer's disease (AD) is a neurodegenerative disorder, and mild cognitive impairment (MCI) represents a transitional stage between AD and cognitively normal (CN) individuals. Early diagnosis is clinically important for delaying disease progression. To address the limitations of single-modal approaches and the insufficient modeling of complex cross-modal interactions in existing multimodal fusion methods, this paper proposes a multimodal deep learning (DL) classification framework integrating structural magnetic resonance imaging (sMRI) and clinical features. The framework employs a 3D ResNet-34 to extract imaging features and a multilayer perceptron to encode clinical data. A bidirectional cross-modal attention mechanism enhances associations between imaging and clinical modalities, followed by an adaptive gated fusion module that dynamically integrates concatenated global multimodal features with cross-modal interaction features. To evaluate the robustness of the proposed framework, all experiments were repeated using five different random seeds, and the results are reported as mean ± standard deviation. Experimental results demonstrate competitive performance across multiple classification tasks, achieving accuracies of 95.67% ± 1.70% for three-class classification (CN vs MCI vs AD) and 93.64% ± 2.36% for four-class classification (CN vs early MCI (EMCI) vs late MCI (LMCI) vs AD). For binary classification (AD vs CN, AD vs MCI, MCI vs CN, and EMCI vs LMCI), the method achieves accuracies of 96.83% ± 1.53%, 95.33% ± 1.00%, 95.83% ± 1.39%, and 93.61% ± 2.35%, respectively. The proposed framework provides an effective solution for multimodal DL-based computer-aided diagnosis of AD.
The increasing clinical use of stereotactic radiosurgery, stereotactic body radiotherapy and intensity-modulated radiotherapy has heightened the need for accurate small-field dosimetry. In sub-centimetre photon fields, dosimetric uncertainties arise from steep dose gradients, loss of lateral charged particle equilibrium, source occlusion, and detector-related volume-averaging and perturbation effects. This study evaluated the influence of detector construction and sensitive volume on percentage depth dose (PDD) measurements in a 1 × 1 cm2, 6 MV photon field delivered by an Elekta Versa HD linear accelerator. PDD measurements were acquired using three detectors with differing sensitive volumes: the PTW Semiflex 3D ionisation chamber (0.07 cm3), PTW PinPoint 3D ionisation chamber (0.016 cm3), and PTW microDiamond detector (0.004 mm3). Measurements were performed in a PTW BeamScan water phantom at a source-to-surface distance of 100 cm and repeated to assess reproducibility. Three repeated measurements were acquired for each detector. Corresponding dose calculations were generated using the Monaco treatment planning system (TPS) Monte Carlo algorithm with a 1 mm grid resolution and 0.5% statistical uncertainty. Substantial deviations between measured and calculated PDDs were observed in the surface and build-up regions, with differences reaching +106% at the phantom surface. These discrepancies reflect the combined effects of steep dose gradients and the absence of charged particle equilibrium. Within the build-up region, the ionisation chambers exhibited under-response of up to approximately -20% at shallow depths, while beyond the depth of maximum dose they demonstrated increasing over-response with depth, reaching approximately +7% at 250 mm for the Semiflex detector. This behaviour is consistent with energy dependence and small-field cavity-theory effects. In contrast, the microDiamond detector-maintained agreement with TPS calculations within approximately ±2% across the entire depth range beyond the build-up region. One-dimensional gamma analysis (global normalisation, 10% dose threshold) yielded pass rates of 98.8%, 72.1%, and 25.9% for the microDiamond, PinPoint, and Semiflex detectors, respectively, using the 1%/1 mm criterion. The findings demonstrate that detector construction and sensitive volume significantly influence PDD accuracy in very small photon fields. The microDiamond detector showed the closest agreement with Monte Carlo TPS calculations, highlighting the advantages of solid-state detectors for small-field dosimetry. Nevertheless, accurate dosimetry in the build-up region remains challenging irrespective of detector type.
Continuous electrocardiogram (ECG) monitoring requires practical engineering solutions that balance predictive performance, response time, and resource consumption in distributed wearable-oriented systems. We present a wearable-fog-cloud framework for continuous ECG decision orchestration in which wearable devices perform first-pass triage and the fog gateway evaluates patient-specific temporal criticality, uncertainty and resource admissibility, reserving cloud consultation for ambiguous or critical cases. The method explicitly distinguishes logical cloud demand from actual cloud execution and incorporates exact record-level decision memoization for repeated beat-level tuples derived from the same ECG record. The framework was evaluated in iFogSim as a controlled system-level simulation using synchronized wearable-side and cloud-side prediction streams derived from PTB-XL. We compared the proposed method with three baselines: wearable-only, cloud-only, and no-reuse. We also assessed scalability, parameter sensitivity, cache realism, component-level ablations, and bootstrap uncertainty at the record and patient levels. At the selected operating point, the method achieved a Matthews correlation coefficient of 0.6073, an F1-score of 0.8255, and an area under the receiver operating characteristic curve of 0.8803, with an average latency of 27.9 ms and an actual cloud offload ratio of 6.13%. Among logically cloud-resolved tuples, 84.66% were served through exact record-level cache reuse. The proposed framework improved predictive performance compared with wearable-only inference while maintaining low latency; compared with the no-reuse configuration, it achieved higher predictive performance with slightly lower real cloud execution. These results indicate that selective cloud escalation and exact record-level decision reuse can improve the trade-off between predictive quality and resource usage in a controlled ECG decision-orchestration framework, while prospective validation on native wearable or ambulatory ECG data remains necessary.
This study investigates how architectural design influences the behavior of lightweight convolutional neural networks in multilabel classification of retinal diseases using the ocular disease intelligent recognition-5 K dataset. Rather than focusing solely on predictive performance, we analyze multiple complementary dimensions, including accuracy, stability across training runs, inter-model agreement, error complementarity, and image-level consensus. Results show that no single architecture consistently achieves both optimal performance and stability. Moreover, different architectures exhibit distinct prediction patterns and fail on partially disjoint subsets of the data, leading to significant complementarity. Agreement and consensus analyses further reveal that disagreement across models correlates with classification difficulty, suggesting its potential as a proxy for uncertainty. These findings demonstrate that architectural inductive bias plays a central role in shaping model behavior beyond aggregate metrics. Importantly, the observed diversity enables improved performance and reliability through model combination, supporting the development of ensemble-based systems for clinically relevant decision support.
Background.Seismocardiography (SCG) is a non-invasive, wearable technology that captures mechanical vibrations of the heart, offering a potential low-cost tool for assessing cardiac function. While conventional imaging modalities such as transthoracic echocardiography (TTE) and cardiac magnetic resonance (CMR) provide detailed structural and volumetric measures, they are operator-dependent and require specialized training. This study aimed to evaluate the correlation between SCG, TTE and CMR markers of systolic and diastolic function in healthy individuals and those with cardiovascular disease (CVD).Methods.Twenty-six subjects (age 45 ± 16 years and 77% male) were included, 10 healthy subjects and 16 subjects with CVD. All subjects underwent SCG, TTE, and CMR measurements before and after intravenous infusion of 2.0 l isotonic saline. SCG signals were segmented into individual heartbeats, and fiducial points were automatically detected and manually validated.Results.SCG amplitudes Cd-to-Dd and Dd demonstrated moderate correlations with the diastolic parametere' from TTE (r= 0.42-0.66), while SCG time intervals Gs-Bd and Ls-Bd correlated moderately with LVET (r= 0.51-0.66). These associations were consistent across both healthy participants and patients with CVD.Conclusion.SCG-derived measures correlate moderately with TTE markers of diastolic (e') and systolic (LVET) function, and these associations are preserved in both healthy and CVD subgroups, indicating robustness across clinical phenotypes. The findings in this study support SCG's potential as a non-invasive tool for assessing mechanical cardiac function. However, further validation in real-world wearable settings is required.
To address the significant challenges in brain tumor magnetic resonance imaging detection-including high morphological heterogeneity of lesions, blurred boundaries, and severe background noise interference-this study proposes a multi-domain cooperative perception model, BTA-DETR (Brain Tumor Aware-DEtection TRansformer). First, we design a content-aware fusion unit, which leverages parallel spatial and channel branches alongside an adaptive gating mechanism to dynamically re-weight local textural details and global semantic features, thereby adaptively capturing the diverse morphologies of lesions. Second, we propose the learnable temperature attention module, which generates a pixel-level spatially heterogeneous temperature field via a lightweight convolutional branch to differentially modulate the sharpness of the attention distribution, thereby improving the model's localization performance in lesion regions with blurred boundaries. Finally, we construct a frequency-spatial cooperative module, which parses pathological textures through a tri-band gating mechanism in the frequency-domain branch, complemented by asymmetric depth-wise directional convolutions in the spatial-domain branch to supplement geometric boundary information, enhancing the representational capacity for fine-grained lesion features. Experimental results on the public Roboflow Brain Tumor Detection dataset demonstrate that, compared to the baseline RT-DETR, BTA-DETR improves mean average precision50 by 4.36 percentage points while reducing parameters by 0.82 M and computational cost by approximately 9.1%. Experimental results on the two external datasets, Kaggle BrainTumor and BraTS 2021 T1ce, further demonstrate that, under a dataset-specific retraining setting, BTA-DETR achieves higher detection metrics than the vanilla RT-DETR baseline.
Electrooculography (EOG) for eye movement detection and condition monitoring are affected by variable baseline drifts and are susceptible to motion artifact-induced noise. To address these limitations, this work introduces an eye movement classification approach, covering blink, saccade, smooth pursuit, vergence, and vestibulo-ocular movements, using impedance measurement from an eye wearable system. Impedance oculography (IOG) data were collected from subjects performing the specified eye movements using a spectacle-mounted two-electrode system connected to a IOG measurement set up, which, with appropriate modifications can be realized as a compact wearable device. The collected data were processed through baseline drift correction, wavelet filtering, and windowing. Notably, the proposed approach eliminates the need for explicit feature extraction for identifying the inherent spatial characteristics of IOG signal for activity classification. Here, a convolution neural network with stratified 5-fold cross validation was implemented to classify the eye movements, with 80% of the data used for training and 20% for testing. The IOG data collected from subjects over extended durations indicted uniform baseline drift, demonstrating the superiority of the IOG over EOG for eye motion signal acquisitions. High class-specific accuracies of 95%, 97%, 97%, 100% and 93% for blink, saccade, smooth pursuit, vergence and vestibulo-ocular movements, respectively, confirm the efficacy of the proposed method for accurate eye-movement classification in wearable eye-tracking systems.
