Medical & Biological Engineering & Computing最新文献

Doxorubicin (DOXO) is a primary treatment for breast cancer but can cause cardiotoxicity in over 25% of patients within the first year post-chemotherapy. Recognizing at-risk patients before DOXO initiation offers pathways for alternative treatments or early protective actions. We analyzed data from 78 Brazilian breast cancer patients, with 34.6% developing cardiotoxicity within a year of their final DOXO dose. To address the limited sample size, we utilized the DAS (Data Augmentation and Smoothing) method, creating 4892 synthetic samples that exhibited high statistics fidelity to the original data. By integrating routine blood biomarkers (C-Reactive protein, total cholesterol, LDL-c, HDL-c, hematocrit, and hemoglobin) and two clinical measures (weighted smoking status and body mass index), our model achieved an AUROC of 0.85±0.10, a sensitivity of 0.89, and a specificity of 0.69, positioning it as a potential screening instrument. Notably, DAS outperformed the established methods, Adaptive Synthetic Sampling (ADASYN), Synthetic Minority Over-Sampling Technique (SMOTE), and Synthetic Data Vault (SDV), underscoring its promise for medical synthetic data generation and pioneering a cardiotoxicity prediction model specifically for DOXO.

Source-free domain adaptation (SFDA) has become crucial in medical image analysis, enabling the adaptation of source models across diverse datasets without labeled target domain images. Self-training, a popular SFDA approach, iteratively refines self-generated pseudo-labels using unlabeled target domain data to adapt a pre-trained model from the source domain. However, it often faces model instability due to incorrect pseudo-label accumulation and foreground-background class imbalance. This paper presents a pioneering SFDA framework, named cascaded network-guided class-balanced multi-prototype auxiliary learning (C $_{}^2$ MAL), to enhance model stability. Firstly, we introduce the cascaded translation-segmentation network (CTS-Net), which employs iterative learning between translation and segmentation networks to generate accurate pseudo-labels. The CTS-Net employs a translation network to synthesize target-like images from unreliable predictions of the initial target domain images. The synthesized results refine segmentation network training, ensuring semantic alignment and minimizing visual disparities. Subsequently, reliable pseudo-labels guide the class-balanced multi-prototype auxiliary learning network (CMAL-Net) for effective model adaptation. CMAL-Net incorporates a new multi-prototype auxiliary learning strategy with a memory network to complement source domain data. We propose a class-balanced calibration loss and multi-prototype-guided symmetry cross-entropy loss to tackle class imbalance issue and enhance model adaptability to the target domain. Extensive experiments on four benchmark fundus image datasets validate the superiority of C $_{}^2$ MAL over state-of-the-art methods, especially in scenarios with significant domain shifts. Our code is available at https://github.com/yxk-art/C2MAL .

In bone tumor resection surgery, patient-specific cutting guides aid the surgeon in the resection of a precise part of the bone. Despite the use of automation methodologies in surgical guide modeling, to date, the placement of cutting planes is a manual task. This work presents an algorithm for the automatic positioning of cutting planes to reduce healthy bone resected and thus improve post-operative outcomes. The algorithm uses particle swarm optimization to search for the optimal positioning of points defining a cutting surface composed of planes parallel to a surgical approach direction. The quality of a cutting surface is evaluated by an objective function that considers two key variables: the volumes of healthy bone resected and tumor removed. The algorithm was tested on three tumor cases in long bone epiphyses (two tibial, one humeral) with varying plane numbers. Optimal optimization parameters were determined, with varying parameters through iterations providing lower mean and standard deviation of the objective function. Initializing particle swarm optimization with a plausible cutting surface configuration further improved stability and minimized healthy bone resection. Future work is required to reach 3D optimization of the planes positioning, further improving the solution.

Lower limb biomechanics of chronic ankle instability (CAI) individuals has been widely investigated, but few have evaluated the internal foot mechanics in CAI. This study evaluated bone and soft tissue stress in CAI contrasted with copers and non-injured participants during a cutting task. Integrating scanned 3D foot shapes and free-form deformation, sixty-six personalized finite element foot models were developed. Computed Achilles tendon forces and measured regional plantar pressure were applied as boundary loading conditions for simulation. It was observed that the primary group differences in foot stress occurred during midstance and heel-off phases of the cutting task. Specifically, healthy individuals had significantly higher stress in the talus and soft tissue around the talus compared to CAI participants. In contrast, CAI participants had significantly higher stress in the cuneiforms and lateral forefoot bones during mid-stance and push-off phases. CAI participants appeared to adopt a protective strategy by transferring greater force to the lateral forefoot at the heel-off phase while lowering stress around the talus, which may be associated with pain relief near the ankle. These findings suggest further attention should be placed on internal stress in CAI at the push-off phase with implications for long-term foot adaptation.

Proximal femoral fractures in children are challenging in clinical treatment due to their unique anatomical and biomechanical characteristics. The distribution and characteristics of fracture lines directly affect the selection of treatment options and prognosis. Pediatric proximal femur fractures exhibit distinctive features, with the distribution and characteristics of the fracture line playing a crucial role in deciding optimal treatment. The study aims to investigate the morphological characteristics of pediatric femoral neck fracture (FNF) from clinical cases by fracture mapping technology and to analyze the relationship between fracture classifications and age. The CT data were collected from 46 consecutive pediatric inpatients' diagnoses of FNF from March 2009 to December 2022. The fracture imaging was reconstructed in three dimensions and performed the simulated anatomical reduction by Mimics and 3-matic. Both Delbet classification and Pauwels angle classification were documented according to the fracture line in each patient. Furthermore, all of the fracture lines in these patients were superimposed to form a fracture map and a heat map. This study included 24 boys and 22 girls (average age, 9.61 ± 3.17 years (4 to 16 years)). The fracture lines of the anterior and superior femoral neck were found to be mainly located in the middle and lower regions of the femoral neck, while fracture lines of the posterior and inferior neck were mainly concentrated in the middle region. Most children younger than 10 years had Delbet type III of fracture (69%), whereas those older than 10 years had Delbet type II of fracture (73%). Furthermore, most children had Pauwels angle type III of fracture (63%), especially in those over 10 years old (80%) (p = 0.0001). FNF in children is predominantly located in the middle and lower regions of the neck. Older children may be prone to be affected with higher fracture location of FNF or unstable type of fracture.

Chronic obstructive pulmonary disease (COPD) is a leading cause of death worldwide and greatly reduces the quality of life. Utilizing remote monitoring has been shown to improve quality of life and reduce exacerbations, but remains an ongoing area of research. We introduce a novel method for estimating changes in ease of breathing for COPD patients, using obstructed breathing data collected via wearables. Physiological signals were recorded, including respiratory airflow, acceleration, audio, and bio-impedance. By comparing patient-specific measurements, this approach enables non-intrusive remote monitoring. We analyze the influence of signal selection, window parameters, feature engineering, and classification models on predictive performance, finding that acceleration signals are most effective, complemented by audio signals. The best model achieves an F1-score of 0.83. To facilitate clinical adoption, we incorporate interpretability by designing novel saliency map methods, highlighting important aspects of the signals. We adapt local explainability techniques to time series and introduce a novel imputation method for periodic signals, improving faithfulness to the data and interpretability.

Detection of early mild cognitive impairment (EMCI) is clinically challenging as it involves subtle alterations in multiple brain sub-anatomic regions. Among different brain regions, the corpus callosum and lateral ventricles are primarily affected due to EMCI. In this study, an improved deep canonical correlation analysis (CCA) based framework is proposed to fuse magnetic resonance (MR) image features from lateral ventricular and corpus callosal structures for the detection of EMCI condition. For this, obtained structural MR images of healthy controls and EMCI subjects are preprocessed. Lateral ventricles and corpus callosum structures are segmented from these images and features are extracted. Extracted features from different brain structures are fused using non-linear orthogonal iteration-based deep CCA. Fused features are employed to differentiate healthy controls and EMCI condition using extreme learning machine classifier. Results indicate that fused callosal and ventricular features are able to detect EMCI. Improved deep CCA algorithm with tuned hyperparameters achieves the highest classifier performance with an F-score of 82.15%. The proposed framework is compared with state-of-the-art CCA approaches, and the results demonstrate its improved performance in EMCI detection. This highlights the potential of the proposed framework in the automated diagnosis of preclinical MCI conditions.

The use of AR technology in image-guided neurosurgery enables visualization of lesions that are concealed deep within the brain. Accurate AR registration is required to precisely match virtual lesions with anatomical structures displayed under a microscope. The purpose of this work was to develop a real-time augmented surgical navigation system using contactless line-structured light registration, microscope calibration, and visible optical tracking. Contactless discrete sparse line-structured light point cloud is utilized to construct patient-image registration. Microscope calibration optimization with dimensional invariant calibrator is employed to enable real-time tracking of the microscope. The visible optical tracking integrates a 3D medical model with surgical microscope video in real time, generating an augmented microscope stream. The proposed patient-image registration algorithm yielded an average root mean square error (RMSE) of 0.78 ± 0.14 mm. The pixel match ratio error (PMRE) of the microscope calibration was found to be 0.646%. The RMSE and PMRE of the system experiments are 0.79 ± 0.10 mm and 3.30 ± 1.08%, respectively. Experimental evaluations confirmed the feasibility and efficiency of microscope AR surgical navigation (MASN) registration. By means of registration technology, MASN overlays virtual lesions onto the microscopic view of the real lesions in real time, which can help surgeons to localize lesions hidden deep in tissue.

In the context of chronic liver diseases, where variability in progression necessitates early and precise diagnosis, this study addresses the limitations of traditional histological analysis and the shortcomings of existing deep learning approaches. A novel patch-level classification model employing multi-scale feature extraction and fusion was developed to enhance the grading accuracy and interpretability of liver biopsies, analyzing 1322 cases across various staining methods. The study also introduces a slide-level aggregation framework, comparing different diagnostic models, to efficiently integrate local histological information. Results from extensive validation show that the slide-level model consistently achieved high F1 scores, notably 0.9 for inflammatory activity and steatosis, and demonstrated rapid diagnostic capabilities with less than one minute per slide on average. The patch-level model also performed well, with an F1 score of 0.64 for ballooning and 0.99 for other indicators, and proved transferable to public datasets. The conclusion drawn is that the proposed analytical framework offers a reliable basis for the diagnosis and treatment of chronic liver diseases, with the added benefit of robust interpretability, suggesting its practical utility in clinical settings.

Performing automatic and standardized 4D TEE segmentation and mitral valve analysis is challenging due to the limitations of echocardiography and the scarcity of manually annotated 4D images. This work proposes a semi-supervised training strategy using pseudo labelling for MV segmentation in 4D TEE; it employs a Teacher-Student framework to ensure reliable pseudo-label generation. 120 4D TEE recordings from 60 candidates for MV repair are used. The Teacher model, an ensemble of three convolutional neural networks, is trained on end-systole and end-diastole frames and is used to generate MV pseudo-segmentations on intermediate frames of the cardiac cycle. The pseudo-annotated frames augment the Student model's training set, improving segmentation accuracy and temporal consistency. The Student outperforms individual Teachers, achieving a Dice score of 0.82, an average surface distance of 0.37 mm, and a 95% Hausdorff distance of 1.72 mm for MV leaflets. The Student model demonstrates reliable frame-by-frame MV segmentation, accurately capturing leaflet morphology and dynamics throughout the cardiac cycle, with a significant reduction in inference time compared to the ensemble. This approach greatly reduces manual annotation workload and ensures reliable, repeatable, and time-efficient MV analysis. Our method holds strong potential to enhance the precision and efficiency of MV diagnostics and treatment planning in clinical settings.