Computers in biology and medicine最新文献

Breast cancer poses a significant health threat worldwide. Contrastive learning has emerged as an effective method to extract critical lesion features from mammograms, thereby offering a potent tool for breast cancer screening and analysis. A crucial aspect of contrastive learning is negative sampling, where the selection of hard negative samples is essential for driving representations to retain detailed lesion information. In large-scale contrastive learning applied to natural images, it is often assumed that extracted features can sufficiently capture semantic content, and that each mini-batch inherently includes ideal hard negative samples. However, the unique characteristics of breast lumps challenge these assumptions when dealing with mammographic data. In response, we introduce ManiNeg, a novel approach that leverages manifestations as proxies to select hard negative samples. As a condensed representation of a physician's domain knowledge, manifestations represent observable symptoms or signs of a disease and can provide a robust basis for choosing hard negative samples. This approach benefits from its invariance to model optimization, facilitating efficient sampling. We tested ManiNeg on the task of distinguishing between benign and malignant breast lumps. Our results demonstrate that ManiNeg not only improves representation in both unimodal and multimodal contexts but also offers benefits that extend to datasets beyond the initial pretraining phase. To support ManiNeg and future research endeavors, we have developed the MVKL mammographic dataset. This dataset includes multi-view mammograms, corresponding reports, meticulously annotated manifestations, and pathologically confirmed benign-malignant outcomes for each case. The MVKL dataset and our codes are publicly available at https://github.com/wxwxwwxxx/ManiNeg to foster further research within the community.

An efficient novel approach is introduced to predict the effectiveness of chemotherapy treatment in lung cancer by monitoring the serum N-glycome of patients combined with artificial intelligence-based data analysis. The study involved thirty-three lung cancer patients undergoing chemotherapy treatments. Serum samples were taken before and after the treatment. The N-linked oligosaccharides were enzymatically released, fluorophore-labeled, and analyzed by capillary electrophoresis with laser-induced fluorescence detection. The resulting electropherograms were thoroughly processed and evaluated by artificial intelligence-based classifiers, i.e., utilizing a machine learning algorithm to categorize the data into two (binary) classes. The classifier analysis method revealed a strong association between the structural changes in the N-glycans and the outcomes of the chemotherapy treatments (ROC >0.9). This novel combination of bioanalytical and AI methods provided a precise and rapid tool for predicting the effectiveness of chemotherapy.

Background: Malaria is a critical and potentially fatal disease caused by the Plasmodium parasite and is responsible for more than 600,000 deaths globally. Early and accurate detection of malaria parasites is crucial for effective treatment, yet conventional microscopy faces limitations in variability and efficiency.

Methods: We propose a novel computer-aided detection framework based on deep learning and attention mechanisms, extending the YOLO-SPAM and YOLO-PAM models. Our approach facilitates the detection and classification of malaria parasites across all infection stages and supports multi-species identification.

Results: The framework was evaluated on three publicly available datasets, demonstrating high accuracy in detecting four distinct malaria species and their life stages. Comparative analysis against state-of-the-art methodologies indicates significant improvements in both detection rates and diagnostic utility.

Conclusion: This study presents a robust solution for automated malaria detection, offering valuable support for pathologists and enhancing diagnostic practices in real-world scenarios.

Musculoskeletal modeling based on inverse dynamics provides a cost-effective non-invasive means for calculating intersegmental joint reaction forces and moments, solely relying on kinematic data, easily obtained from smart wearables. On the other hand, the accuracy and precision of such models strongly hinge upon the selected scaling methodology tailored to subject-specific data. This study investigates the impact of upper body mass distribution on internal and external kinetics computed using a comprehensive musculoskeletal model during level walking in both normal weight and obese individuals. Human motion data was collected using seventeen body worn inertial measuring units for nineteen (19) healthy subjects. The results indicate that variations in segmental masses and centers of mass, resulting from diverse mass scaling techniques, significantly affect ground reaction force estimations in obese subjects, particularly in the vertical component, with a root mean square error (RMSE) of 54.7 ± 23.8 %BW; followed by 12.3 ± 8.0 %BW (medio-lateral); and 6.2 ± 3.2 %BW (antero-posterior). The vertical component of hip, knee, and ankle joint reaction forces also exhibit sensitivity to personalized mass distribution variations. Importantly, the degree of deviation in model predictions increases with body mass index. Statistical analysis using single sample Wilcoxon-Signed Rank test for non-normal data and t-test for normal data, revealed significant differences (p < 0.05) in the computed errors in kinetic parameters between the two scaling approaches. The body shape-based scaling approach significantly impacts musculoskeletal modeling in clinical applications where the upper body mass distribution is crucial, such as in spinal deformities, obesity, and low back pain. This approach accounts for the body shape inherent variability within the same BMI category and enhances the predicted joint kinetics.

Fetal echocardiography (ultrasound of the fetal heart) plays a vital role in identifying heart defects, allowing clinicians to establish prenatal and postnatal management plans. Machine learning-based methods are emerging to support the automation of fetal echocardiographic analysis; this review presents the findings from a literature review in this area. Searches were queried at leading indexing platforms ACM, IEEE Xplore, PubMed, Scopus, and Web of Science, including papers published until July 2023. In total, 343 papers were found, where 48 papers were selected to compose the detailed review. The reviewed literature presents research on neural network-based methods to identify fetal heart anatomy in classification and segmentation modelling. The reviewed literature uses five categorical technical analysis terms: attention and saliency, coarse to fine, dilated convolution, generative adversarial networks, and spatio-temporal. This review offers a technical overview for those already working in the field and an introduction to those new to the topic.

With advances in sequencing technologies, the use of high-throughput sequencing to characterize microbial communities is becoming increasingly feasible. However, metagenomic assembly poses computational challenges in reconstructing genes and organisms from complex samples. To address this issue, we introduce a new concept called Adaptive Sequence Alignment (ASA) for analyzing metagenomic DNA sequence data. By iteratively adapting a set of partial alignments of reference sequences to match the sample data, the approach can be applied in multiple scenarios, from taxonomic identification to assembly of target regions of interest. To demonstrate the benefits of ASA, we present two application scenarios and compare the results with state-of-the-art methods conventionally used for the same tasks. In the first, ASA accurately detected microorganisms from a sequenced metagenomic sample with a known composition. The second illustrated the utility of ASA in assembling target genetic regions of the microorganisms. An example implementation of the ASA concept is available at https://github.com/elolab/ASA.

Surface electromyography (sEMG), a non-invasive technique, offers the ability to identify insights into the activities of muscles in the form of electrical pulses. During the process of recording, the sEMG signals frequently become contaminated by a multitude of different artifacts, the origin of which may be attributed to numerous sources. These artifacts affect the reliability and accuracy of the pure sEMG activity, and subsequently reduce the quality of analysis and interpretation. This can lead to a misinterpretation of sEMG signals, incorrect diagnostic, or a false decision in the case of human-machine interfaces (HMI), etc. Currently, several approaches have been developed to remove or reduce the effect of artifacts on the sEMG activity. In this paper, a comprehensive review of the current studies dealing with identification, detection, and removal of artifacts from sEMG signals is proposed. In addition, this study presents different features used to characterize the artifacts from that of the clean sEMG recordings. Finally, in order to improve the quality of denoising methods, the associated challenges of detection and artifact removal approaches are discussed to be addressed carefully in the future works.