Computers in biology and medicine最新文献

Background:

Extracting principal diagnosis from patient discharge summaries is an essential task for the meaningful use of medical data. The extraction process, usually by medical staff, is laborious and time-consuming. Although automatic models have been proposed to retrieve principal diagnoses from medical records, many rare diagnoses and a small amount of training data per rare diagnosis provide significant statistical and computational challenges.

Objective:

In this study, we aimed to extract principal diagnoses with limited available data.

Methods:

We proposed the OLR-Net, Object Label Retrieval Network, to extract principal diagnoses for discharge summaries. Our approach included semantic extraction, label localization, label retrieval, and recommendation. The semantic information of discharge summaries was mapped into the diagnoses set. Then, one-dimensional convolutional neural networks slid into the bottom-up region for diagnosis localization to enrich rare diagnoses. Finally, OLR-Net detected the principal diagnosis in the localized region. The evaluation metrics focus on the hit ratio, mean reciprocal rank, and the area under the receiver operating characteristic curve (AUROC).

Results:

12,788 desensitized discharge summary records were collected from the oncology department at Hainan Hospital of Chinese People’s Liberation Army General Hospital. We designed five distinct settings based on the number of training data per diagnosis: the full dataset, the top-50 dataset, the few-shot dataset, the one-shot dataset, and the zero-shot dataset. The performance of our model had the highest HR@5 of 0.8778 and macro-AUROC of 0.9851. In the limited available (few-shot and one-shot) dataset, the macro-AUROC were 0.9833 and 0.9485, respectively.

Conclusions:

OLR-Net has great potential for extracting principal diagnosis with limited available data through label localization and retrieval.

Background

Variations in symptoms and indistinguishable depression episodes of unipolar depression (UD) and bipolar disorder (BD) make the discrimination difficult and time-consuming. For adolescents with high disease prevalence, an efficient diagnostic tool is important for the discrimination and treatment of BU and UD.

Methods

This multi-center cross-sectional study involved 1587 UD and 246 BD adolescents aged 12–18. A combination of standard questionnaires and demographic information was collected for the construction of a full-item list. The unequal patient number was balanced with three data balancing algorithms, and 4 machine learning algorithms were compared for the discrimination ability of UD and BD in three age groups: all ages, 12–15 and 16–18. Random forest (RF) with the highest accuracy were used to rank the importance of features/items and construct the 25-item shortlist. A separate dataset was used for the final performance evaluation with the shortlist, and the discrimination ability for UD and BD was investigated.

Results

RF performed the best for UD and BD discrimination in all 3 age groups (AUC 0.88–0.90). The most important features that differentiate UD from BD belong to Parental Bonding Instrument (PBI) and Loneliness Scale of the University of California at Los Angeles (UCLA). With RF and the 25-item shortlist, the diagnostic accuracy can still reach around 80 %, achieving 95 % of the accuracy levels obtained with all features.

Conclusions

Through machine learning algorithms, the most influencing factors for UD and BD classification were recombined and applied for rapid diagnosis. This highly feasible method holds the potential for convenient and accurate diagnosis of young patients in research and clinical practice.

Objectives

Cracked tooth syndrome (CTS) is one of the major causes of tooth loss, presents the problem of early microcrack symptoms that are difficult to distinguish. This paper aims to investigate the practicality and feasibility of an improved object detection algorithm for automatically detecting cracks in dental optical images.

Methods

A total of 286 teeth were obtained from Sun Yat-sen University and Guangdong University of Technology, and simulated cracks were generated using thermal expansion and contraction. Over 3000 images of cracked teeth were collected, including 360 real clinical images. To make the model more lightweight and better suited for deployment on embedded devices, this paper improves the YOLOv8 model for detecting tooth cracks through model pruning and backbone replacement. Additionally, the impact of image enhancement modules and coordinate attention modules on optimizing our model was analyzed.

Results

Through experimental validation, we conclude that that model pruning reduction maintains performance better than replacing a lightweight backbone network on a tooth crack detection task. This approach achieved a reduction in parameters and GFLOPs by 16.8 % and 24.3 %, respectively, with minimal impact on performance. These results affirm the effectiveness of the proposed method in identifying and labeling tooth fractures. In addition, this paper demonstrated that the impact of image enhancement modules and coordinate attention mechanisms on YOLOv8's performance in the task of tooth crack detection was minimal.

Conclusions

An improved object detection algorithm has been proposed to reduce model parameters. This lightweight model is easier to deploy and holds potential for assisting dentists in identifying cracks on tooth surfaces.

Objectives

CT perfusion (CTP) imaging is vital in treating acute ischemic stroke by identifying salvageable tissue and the infarcted core. CTP images allow quantitative estimation of CT perfusion parameters, which can provide information on the degree of tissue hypoperfusion and its salvage potential. Traditional methods for estimating perfusion parameters, such as singular value decomposition (SVD) and its variations, are known to be sensitive to noise and inaccuracies in the arterial input function. To our knowledge, there has been no implementation of deep learning methods for CT perfusion parameter estimation.

Materials & methods

In this work, we propose a deep learning method based on the Transformer model, named CTPerformer-Net, for CT perfusion parameter estimation. In addition, our method incorporates some physical priors. We integrate physical consistency prior, smoothness prior and the physical model prior through the design of the loss function. We also generate a simulation dataset based on physical model prior for training the network model.

Results

In the simulation dataset, CTPerformer-Net exhibits a 23.4 % increase in correlation coefficients, a 95.2 % decrease in system error, and a 90.7 % reduction in random error when contrasted with block-circulant SVD. CTPerformer-Net successfully identifies hypoperfused and infarcted lesions in 103 real CTP images from the ISLES 2018 challenge dataset. It achieves a mean dice score of 0.36 for the infarct core segmentation, which is slightly higher than the commercially available software (dice coefficient: 0.34) used as a reference level by the challenge.

Conclusion

Experimental results on the simulation dataset demonstrate that CTPerformer-Net achieves better performance compared to block-circulant SVD. The real-world patient dataset confirms the validity of CTPerformer-Net.

Background

Submaximal muscle strength grading is clinically significant to monitor the progress of rehabilitation. Especially muscle strength grading of core back muscles is challenging using the conventional manual muscle testing (MMT) methods. The muscles are crucial to recovery from back pain, spinal cord injury, stroke and other related diseases. The subjective nature of MMT, adds more ambiguity to grade fine progressions in submaximal strength levels involving 4-, 4 and 4+ grades. Electromyogram (EMG) has been widely used as a quantitative measure to provide insight into the progress of muscle strength. However, several EMG features have been reported in previous studies, and the selection of suitable features pertaining to the problem has remained a challenge.

Method

Principal Component Analysis (PCA) biplot visualization is employed in this study to select EMG features that highlight fine changes in muscle strength spanning the submaximal range. Features that offer maximum loading in the principal component subspace, as observed in the PCA biplot, are selected for grading submaximal strength.

The performance of the proposed feature set is compared with conventional Principal Component (PC) scores. Submaximal muscle strength grades of 4-, 4, 4+ or 5 are assigned using K-means and Gaussian mixture model clustering methods. Clustering performance of the two feature selection methods is compared using the silhouette score metric.

Results

The proposed feature set from biplot visualization involving Root Mean Square (RMS) EMG and Waveform Length in combination with Gaussian Mixture Model (GMM) clustering method was observed to offer maximum accuracy. Muscle-wise mean Silhouette Index (SI) scores (p < 0.05) of .81, .74 (Longissimus thoracis left, right) and .73, .77 (Iliocostalis lumborum left, right) were observed. Similarly grade wise mean SI scores (p < 0.05) of .80, .76, .73, and .981 for grades 4-, 4, 4+, and 5 respectively, were observed.

Conclusion

The study addresses the problem of selecting minimum features that offer maximum variability for EMG assisted submaximal muscle strength grading. The proposed method emphasizes using biplot visualization to overcome the difficulty in choosing appropriate EMG features of the core back muscles that significantly distinguishes between grades 4-, 4, 4+ and 5.

Accurate gestational age (GA) prediction is crucial for monitoring fetal development and ensuring optimal prenatal care. Traditional methods often face challenges in terms of precision and prediction efficiency. In this context, leveraging modern deep learning (DL) techniques is a promising solution. This paper introduces a novel DL approach for GA prediction using fetal brain images obtained via magnetic resonance imaging (MRI), which combines the strength of the Xception pretrained model with a multihead attention (MHA) mechanism. The proposed model was trained on a diverse dataset comprising 52,900 fetal brain images from 741 patients. The images encompass a GA ranging from 19 to 39 weeks. These pretrained models served as feature extraction components during the training process. The extracted features were subsequently used as the inputs of different configurable MHAs, which produced GA predictions in days. The proposed model achieved promising results with 8 attention heads, 32 dimensionality of the key space and 32 dimensionality of the value space, with an R-squared (R²) value of 96.5 %, a mean absolute error (MAE) of 3.80 days, and a Pearson correlation coefficient (PCC) of 98.50 % for the test set. Additionally, the 5-fold cross-validation results reinforce the model's reliability, with an average R² of 95.94 %, an MAE of 3.61 days, and a PCC of 98.02 %. The proposed model excels in different anatomical views, notably the axial and sagittal views. A comparative analysis of multiple planes and a single plane highlights the effectiveness of the proposed model against other state-of-the-art (SOTA) models reported in the literature. The proposed model could help clinicians accurately predict GA.

Background and objectives

This study introduced an animal cognitive function assessment system using a novel circular eight-room maze (CERM). The CERM, designed for tracking path trajectories in animal models of cognitive impairment pathologies such as traumatic brain injury (TBI), comprised a 120-cm diameter disk with eight rooms (30 cm × 25 cm × 30 cm).

Methods

These rooms have magnetic interfaces for modular assembly and disassembly. Notably, one room remained dark and contained food, while the remaining seven rooms automatically lit up when a rat entry, facilitating the assessment of the rat's learning and spatial memory. An infrared night vision camera captured the animal's search trajectory, and binary image processing techniques were employed to eliminate noise and extract the rat's position coordinates to record the rat's path trajectories. The system automatically calculated various cognitive assessment parameters, such as latency, distance traveled, time spent in each quadrant, inner and outer area exploration times, short-term and long-term memory errors, and the number of entries to all the rooms by chance/by memory.

Results

The analysis of overall path trajectories revealed increasingly erratic movement and a growing reliance on chance to enter rooms in rats with TBI over time, likely due to declining memory and the consequent inability to locate the food room. Moreover, increased trajectories in the first quadrant and inner area characterized the behavior of rats with TBI, with statistically significant differences from the sham group observed on day 7. By day 28, all cognitive parameters except short-term memory error significantly differed between the two groups.

Conclusion

Experimental data indicated a substantial increase in irregular search behavior in the TBI group over time, suggesting deterioration in cognitive function and an inability to accurately recall the food room. Conversely, the sham group exhibited consistent search trajectories, typically following the walls and rapidly locating the food room. Moreover, their room entries were guided by memory rather than by chance. Compared with traditional maze tests, this system's strengths lie in its ability to provide more quantitative data and vividly portray behavioral patterns. Therefore, the proposed CERM system can be used as an effective tool for cognitive assessment.

Background

This study seeks to quantify the effects of upper body veno-arterial extracorporeal membrane oxygenation (VA ECMO) on the anatomical distribution of oxygen delivery in the setting of hypoxic respiratory failure and provide new insights that will guide clinical use of this support strategy to bridge patients to lung transplant.

Methods

Employing a patient-specific vascular geometry and a quantitative model of oxygen transport, computational simulations were performed to determine hemodynamics and oxygen delivery in the ascending and descending aorta, left and right coronary arteries, and great vessels during upper body VA ECMO support. Oxygen content in ECMO circuit blood flow was varied while considering different degrees of lung failure severity. Using lumped parameter models to dynamically apply perfusion boundary conditions, hemodynamic parameters and oxygen content were analyzed to assess the effect of ECMO supply titration.

Results

The results emphasize the importance of anatomical distribution for tissue oxygen delivery in severe lung failure, with ECMO-derived flow primarily augmenting oxygen content in specific vascular beds. They also demonstrate that although cannulating the subclavian artery can enhance cerebral oxygen delivery, its ability to ensure sufficient oxygen delivery to the coronary circulation seems to be comparatively restricted.

Conclusions

The oxygen delivery to a specific vascular area is primarily determined by the oxygen content in the source of perfusion. Caution is advised with upper body VA ECMO for patients with hypoxic respiratory failure and right ventricle dysfunction, due to potential coronary ischemia. Management of these patients is challenging due to disease progression and organ availability uncertainties.

Esophageal squamous cell carcinoma (ESCC) is a prevalent malignant tumor of the digestive tract. Clinical findings reveal that the five-year survival rate for mid-to late-stage ESCC patients is merely around 20 %, whereas those diagnosed at an early stage can achieve up to a 95 % survival rate. Consequently, early detection is paramount to improving ESCC patient survival. Protein markers are essential for diagnosing diseases, and the identification of new candidate proteins associated with ESCC through the protein-protein interaction (PPI) network is aimed for in this paper. The PPI network related to ESCC was constructed using protein data, comprising 2094 nodes and 19,660 edges. To assess the nodes' importance in the network, three metrics—degree centrality, betweenness centrality, and closeness centrality—were employed, leading to the identification of 81 key proteins. Subsequently, the biological significance of these proteins in the network was explored, combining biomedical knowledge from three perspectives: network, node, and cluster. The results demonstrated that 52 out of 81 key proteins were confirmed to be linked to ESCC. Among the remaining 29 unreported proteins, 18 displayed significant biological significance, indicating their potential as protein markers related to ESCC.

Magnetic Resonance Imaging (MRI) plays a pivotal role in modern clinical practice, providing detailed anatomical visualization with exceptional spatial resolution and soft tissue contrast. Dynamic MRI, aiming to capture both spatial and temporal characteristics, faces challenges related to prolonged acquisition times and susceptibility to motion artifacts. Balancing spatial and temporal resolutions becomes crucial in real-world clinical scenarios. In the realm of dynamic MRI reconstruction, while Convolutional Recurrent Neural Networks (CRNNs) struggle with long-range dependencies, CRNNs require extensive iterations, impacting efficiency. Transformers, known for their effectiveness in high-dimensional imaging, are underexplored in dynamic MRI reconstruction. Additionally, prevailing algorithms fall short of achieving superior results in demanding generative reconstructions at high acceleration rates. This research proposes a novel approach for dynamic MRI reconstruction, named CRNN-Refined Spatiotemporal Transformer Network (CST-Net). The spatiotemporal Transformer initiates reconstruction, modeling temporal and spatial correlations, followed by refinement using the CRNN. This integration mitigates inaccuracies caused by damaged frames and reduces CRNN iterations, enhancing computational efficiency without compromising reconstruction quality. Our study compares the performance of the proposed CST-Net at 6 × and 12 × undersampling rates, showcasing its superiority over existing algorithms. Particularly, in challenging 25× generative reconstructions, the CST-Net outperforms current methods. The comparison includes experiments under both radial and Cartesian undersampling patterns. In conclusion, CST-Net successfully addresses the limitations inherent in existing generative reconstruction algorithms, thereby paving the way for further exploration and optimization of Transformer-based approaches in dynamic MRI reconstruction. Code and Datasets can be available: https://github.com/XWangBin/CST-Net.