This paper introduces an advanced cancer-cell estimation technique to improve estimation accuracy. In cancer treatment, chemotherapy is commonly used to deliver potent drugs to regulate the growth of cancer/tumor cells (T_c) . Consequently, this may have adverse effects on normal/host cells (H_c) while regulating the growth of T_c . Therefore, this requires reliable information on tumor growth, particularly the behavior of T_c . However, T_c is strongly associated with the dynamics of H_c and immune cells (I_c) ; consequently, existing methods focus on identifying T_c , H_c , and I_c to achieve enhanced estimation accuracy. The difficulty in measuring I_c and the need for reliable information on H_c and T_c in the presence of uncertainties necessitate the use of model-based estimation. In this work, we use the chaotic dynamics of a tumor growth model to capture chaotic physiological dynamics. Furthermore, we integrate an advanced filtering technique with the tumor growth model to estimate the unmeasured states in the presence of uncertainties. In simulation studies, three scenarios with different growth rates of H_c are considered, and the proposed method is compared with existing cancer-cell estimation methods using root-mean-square error (RMSE). The simulation results demonstrate the improved accuracy of the proposed cancer-cell estimation technique in estimating H_c , T_c , and I_c cells.
Acute respiratory failure (ARF) requires rapid intervention to prevent patient deterioration and life-threatening complications. Early respiratory support through prehospital Continuous Positive Airway Pressure (CPAP) has been shown to improve ARF outcomes. However, existing CPAP devices present technical barriers for use in out-of-hospital emergency scenarios. This study presents the design and preliminary performance evaluation of a novel portable, ready-to-use CPAP device for out-of-hospital respiratory support. The device includes a blower providing steady air insufflation through a face mask, with Positive End-Expiratory Pressure (PEEP) regulated by two fixed-point PEEP valves. To identify optimized operational conditions ensuring target respiratory support, a lumped-parameter model (LPM) was developed to simulate device–patient interaction. Input data for the LPM included experimental characteristics of the device components. Simulations were conducted using breathing profiles representing varying ARF-like conditions and severity. Additionally, computational fluid dynamics (CFD) analysis was performed to assess the risk of CO2 rebreathing through the mask. LPM simulations described the dynamic interaction between the breathing profile and device performances, and the consequent pressure and flow patterns during the respiratory cycle. Across all simulated conditions and device settings, PEEP remained close to the predefined target of 5 cm H2O without reaching critical thresholds, confirming design choices efficacy. CFD analysis indicated an end-expiratory residual CO2 concentration of 2
The electrocardiogram (ECG) is the primary non-invasive tool for early screening and clinical management of cardiovascular disease, but deep learning deployment is limited by scarce labeled data, by architectures that ignore the periodic structure of ECG signals, and by the trade-off between long-sequence efficiency and diagnostic accuracy. We propose JEPA-Mamba, a self-supervised representation learning framework that integrates the Joint-Embedding Predictive Architecture (JEPA) with a bidirectional Mamba encoder and a channel–temporal attention module (CTAM) for multi-adaptive ECG classification on downstream tasks. High-ratio contiguous-block masking trains an exponential moving average (EMA) encoder and a narrow predictor to learn diagnostic features without hand-crafted augmentation, while the bidirectional Mamba (BiM) encoder, augmented by rotary position embedding (RoPE) and a register mechanism, couples convolutional waveform extraction with linear-complexity long-sequence modeling and CTAM recalibrates the 12 leads and focuses temporal attention on clinically salient segments. Under a strict patient-level split on PTB-XL, with an encoder pretrained on MIMIC-IV-ECG for cross-dataset validation on the same tasks, JEPA-Mamba attains macro-averaged AUCs of 0.9296 ± 0.002 and 0.9306 ± 0.002 on the fully annotated task, outperforming self-supervised baselines on most downstream tasks and surpassing the most direct JEPA counterpart in the ECG domain across all tasks. Improvements over external baselines on the all, Sub-diag., and Diag. tasks are statistically significant, and internal ablation differences are confirmed by DeLong tests. Per-class analyses, attention maps, t-SNE embeddings, and robustness experiments under noise and baseline drift support these design choices. These results offer preliminary evidence for the clinical potential of JEPA-Mamba.
Postoperative cough after lung surgery is clinically relevant, but routine assessment remains intermittent and subjective. This study evaluated the internal feasibility of an audio-based framework for cough-event detection, dry/productive cough classification, and structured follow-up reporting. More than 7,000 raw recordings were collected from 80 patients undergoing lung surgery at Tongji Hospital between June 26 and August 13, 2024 under an approved ethics protocol. After segmentation and expert-consensus screening, the cough-type cohort comprised 1,908 unique clips (1,500 dry and 408 productive). Patients, rather than clips, were partitioned into training, validation, and test subsets of 56, 12, and 12 patients, respectively. One-second recordings were resampled to 16 kHz and converted into 128-coefficient MFCC representations. K-DenseNet performed cough detection and RUNet performed dry/productive classification. Under deterministic frozen-checkpoint inference, K-DenseNet achieved 99.16
This study aims to enhance ultrasound image quality while preserving fine structures around pathological regions by adopting a Vision Transformer (ViT) encoder to integrate global context and suppress reconstruction artifacts under supervised paired training. We propose a hybrid generative model composed of a ViT encoder and a CNN decoder. Self-attention captures long-range dependencies between image patches and produces attention-guided weights that reweight decoder features to reduce boundary artifacts during local reconstruction. Using 803 low–/high-quality paired 512 × 512 images, the model is trained on 751 pairs and tested on 52 pairs, and evaluated with PSNR, SSIM, and LPIPS. The proposed method achieves PSNR 23.7204, SSIM 0.8837, and LPIPS 0.1634 on the test set, which are the best values among the compared methods. Paired supervision allows the result to be compared against a reference formed from the same acquisition, and the ViT–CNN design combines global contextual information with local detail reconstruction. The results indicate that attention computed inside the encoder can be reused as a spatial weighting signal without anatomical annotation.
