Abstract Human Activity Recognition (HAR) systems using deep learning have shown significant promise; however, deploying such models on edge computing systems remains challenging due to constraints in inference latency, memory footprints and computational capacity. This study proposes a lightweight, end-to-end patch-based Transformer architecture designed for efficient HAR for resource constrained edge environments, evaluated through an architectural ablation study and multiple quantization strategies using TensorFlow Lite. We conducted extensive experiments by varying patch lengths and Transformer encoder depths in order to evaluate the impact on classification accuracy, latency and model complexity. The results demonstrate that moderate temporal patching achieves an effective balance between temporal representation learning and computational efficiency. Experiments have been conducted on two widely open-source benchmark datasets, WISDM and PAMAP2, where raw sensor signals are pre-processed through segmentation, normalization, and overlapping windowing before being fed into a compact Transformer model with patch embedding and multi-head self-attention for feature extraction. The trained models are converted into FP32, FP16, INT8 dynamic range, and INT8 full formats to evaluate trade-offs between different metrics such as accuracy, model size, and inference latency. The baseline model achieves 96.77% accuracy on WISDM and 98.48% on PAMAP2, with consistently high macro F1-scores. Among all quantization variants, TFLite-INT8-Dynamic reduced model size by 95.55% and 89.80% for WISDM and PAMAP2, respectively and drastically improved inference latency by around 99% for both datasets with minimal accuracy degradation. These findings demonstrate that the proposed patch-based Transformer model for HAR has achieved an effective balance between recognition performance and computational efficiency, which indicates its strong deployment feasibility for resource constrained edge environments through TensorFlow Lite benchmarking and can provide a scalable solution for edge intelligence applications.
Kisspeptin is an upstream regulator of the hypothalamic-pituitary-ovarian axis and plays a critical role in the estrous cycle and pregnancy in domestic animals. The KISS1 gene, also known as the "metastatin" gene, has emerged as a promising diagnostic biomarker in reproductive physiology across livestock species. This study aims to explore the dynamics of urinary/plasma kisspeptin and their creatinine-normalized ratio during the bovine and caprine estrous cycle and pregnancy. Experimental animals comprised two groups (cyclic and pregnant). The cyclic group consisted of 12 cyclic crossbred cows and 12 cyclic black Bengal does, and the pregnant group consisted of 12 pregnant crossbred cows and 12 pregnant black Bengal does. Early-morning fresh urine and blood samples were collected from each cyclic animal group on Day 0, 6, 12, and 21 of the estrous cycle, and at fortnightly intervals from each pregnant animal group for estimation of urinary and plasma kisspeptin. Urinary and plasma kisspeptin were estimated using enzyme-linked immunosorbent assay, and urinary and plasma creatinine values were estimated using the alkaline picrate method. Two-way repeated-measures analysis ofvariance was used to analyze the samples, followed by regression analysis and student’s t-test to assess significant differences between non-pregnant and pregnant animals. Further receiver operating curve analysis was used to assess the diagnostic performance of urinary and plasma kisspeptin and their creatinine-normalized ratios. The results obtained for the said objective revealed that urinary kisspeptin levels showed a qualitatively similar trend to plasma kisspeptin (r=0.99) during the estrous cycle in crossbred cattle and black Bengal does. The highest concentration of UKP (2.45 ± 0.21ng/ml and 2.44 ± 0.08 ng/ml) was recorded during the proestrus stage of the estrous cycle of both crossbred cows and black Bengal does. The highest plasma kisspeptin-to-creatinine ratios were also observed on Day 21 (14.49 ± 0.91 and 16.22 ± 1.53), whereas the corresponding urinary kisspeptin-to-creatinine ratios reached 11.84 ± 0.70 and 7.47 ± 0.34, respectively, in crossbred cattle and black Bengal does. Plasma and urinary kisspeptin concentrations were found to increase significantly (P<0.05) from early through late pregnancy (4.28 ± 0.19 to 33.40 ± 0.78; 2.01± 0.10 ng/ml to 23.01 ± 0.54 ng/ml) in crossbred cows and black Bengal goats (3.49 ± 0.53 to 30.50 ± 1.26 ng/mL; 1.68 ± 0.01 to 19.84 ± 0.82 ng/mL), respectively. The PKP/PC ratio (134.9 ± 10.96; 127.30 ± 6.62) was found to be significantly higher (P < 0.01) during late pregnancy than the UKP/UC ratio (71.94 ± 10.63; 68.4 ± 6.70 ng/ml) in crossbred cattle and black Bengal goats. The results of our study revealed that kisspeptin can be detected in the urine of cyclic and pregnant domestic animals, such as crossbred cattle and black Bengal does, and that its ratio to creatinine can serve as a reliable diagnostic marker for predicting ovulation and pregnancy in crossbred cows and black Bengal does.
Patients with end-stage renal disease receiving maintenance hemodialysis are at increased risk of developing active tuberculosis (TB) and are therefore, considered a priority population for latent tuberculosis infection (LTBI) screening and management. This study aimed to estimate the prevalence of LTBI and identify factors independently associated with QuantiFERON-TB Gold Plus (QFT-Plus) positivity among patients receiving maintenance hemodialysis in 13 selected governorates in Egypt. A retrospective registry-based cross-sectional study was conducted using records from the Egyptian Tuberculosis Health Initiative collected between November 2022 and December 2023 from designated screening facilities. Descriptive statistics were used to summarize the data, and between-group comparisons were performed using the Chi-square test and analysis of variance (ANOVA), as appropriate. Multivariable logistic regression and generalized linear mixed models (GLMMs) were applied to identify factors independently associated with QFT-Plus positivity. Geographic distribution was visualized using a choropleth map generated with ArcGIS Desktop version 10.8. Among the 18,060 screened individuals, 3668 (20.31%) had positive QFT-Plus results when calculated using all tested participants as the denominator. Among participants with determinate results (n = 16,477), the QFT-Plus positivity rate was 22.26%, while 1583 (8.77%) had indeterminate results. In the GLMM analysis, older age (adjusted odds ratio [aOR] = 1.03, 95% CI 1.02–1.03; p < 0.001), male sex (aOR = 1.59, 95% CI 1.47–1.72; p < 0.001), and smoking (aOR = 1.21, 95% CI 1.05–1.41; p = 0.011) were significant factors associated with QFT-Plus positivity after adjustment for diabetes, hypertension, infection with human immunodeficiency virus, hepatitis C virus, and hepatitis B virus, region, and patient load, while accounting for clustering at the governorate level. Descriptive geographic mapping showed substantial geographic variation in LTBI prevalence across governorates. Regional QFT-Plus positivity among participants with determinate QFT-Plus results was similar in Nile Delta (22.5%) and Upper Egypt (21.7%), although marked heterogeneity was observed. The highest positivity rates were reported in Luxor (27.1%) and Sharqiya (25.7%). Geographic variation was also observed in indeterminate QFT-Plus results. These findings support further evaluation of structured LTBI screening and follow-up pathways within renal care services, particularly for older male patients and smokers. Targeted surveillance strategies may facilitate identification of individuals at increased risk and help inform future TB prevention policies in this population. Additional prospective studies should evaluate treatment uptake, progression to active TB, clinical benefit, cost-effectiveness, and optimal screening strategies.
Electroconvulsive therapy (ECT) is an effective psychiatric treatment and relies on an iatrogenically induced, generalized seizure. Seizure quality scores have been developed to predict treatment outcomes of ECT. Its intensity can be governed by various parameters like electrode placement, anesthesia, and the systemic levels of CO₂. The aim of this retrospective study was to investigate whether reducing end-tidal CO₂ (etCO₂) through hyperventilation prior to ECT seizure would improve seizure quality score considering electrode placement and anesthetic agent. Case notes from 2023 were reviewed and seizure quality scores, pre-seizure etCO₂, electrode placement, and anesthetic agent were obtained. The resulting 732 ECTs from 68 patients were analyzed. Lower etCO₂ and its interaction with main anesthetic agent and different bilateral electrode placement were not significantly associated with improved seizure quality overall. Higher age was associated with a decrease in seizure quality (β = -0.03, SE = 0.01, z = − 5.58, p < .001). Post-hoc power analysis indicated sample sizes of at least 400 patients to detect isolated etCO2 effects. Seizure quality should be understood as a complex interplay of technical, neurophysiological, and patient-related factors. Large and standardized multicenter studies are needed to explore the influence of treatment protocols on seizure quality.
Mycoplasma pneumoniae pneumonia (MPP) is associated with a high risk of liver injury, which adversely affects clinical outcomes and healthcare costs. Early identification of at-risk children remains challenging, and no validated predictive tool is currently available. This multicenter retrospective study included 1321 children with MPP from two centers in Nanjing, China. The development cohort was randomly split into training (67%) and test (33%) sets, with an independent external validation cohort (n = 640) from another hospital. Liver injury was defined as alanine aminotransferase (ALT) > 80 U/L. Feature selection was performed using the boruta algorithm, and ten machine learning algorithms were developed and evaluated. Model performance was assessed using area under the receiver operating characteristic curve (AUC), accuracy (ACC), sensitivity, specificity, calibration metrics, and decision curve analysis (DCA). The SHAP method was used for model interpretation. Among 1321 patients, 55 (4.2%) developed liver injury. The boruta algorithm identified six predictors: LDH, D-dimer, CK, age, IL-6, and CK-MB. The neural network (nnet) model demonstrated optimal performance, with AUC values of 0.914 in the training set, 0.829 in the test set, and 0.755 in the external validation set. The model showed acceptable calibration and modest clinical utility on decision curve analysis. SHAP analysis revealed LDH as the most important predictor. An interactive web-based application was developed to facilitate clinical implementation. We developed and validated a machine learning-based model using six readily available predictors that identifies children with MPP at risk for liver injury. The accompanying web-based tool may assist clinicians in early risk stratification and timely intervention.
