J Voice. 2024 Feb 12:S0892-1997(24)00018-3. doi: 10.1016/j.jvoice.2024.01.015. Online ahead of print.

ABSTRACT

OBJECTIVES: This study aimed to evaluate the performance of artificial intelligence (AI) models using connected speech and vowel sounds in detecting benign laryngeal diseases.

STUDY DESIGN: Retrospective.

METHODS: Voice samples from 772 patients, including 502 with normal voices and 270 with vocal cord polyps, cysts, or nodules, were analyzed. We employed deep learning architectures, including convolutional neural networks (CNNs) and time series models, to process the speech data. The primary endpoint was the area under the receiver's operating characteristic curve for binary classification.

RESULTS: CNN models analyzing speech segments significantly outperformed those using vowel sounds in distinguishing patients with and without benign laryngeal diseases. The best-performing CNN model achieved areas under the receiver operating characteristic curve of 0.895 and 0.845 for speech and vowel sounds, respectively. Correlations between AI-generated disease probabilities and perceptual assessments were more pronounced in the connected-speech analyses. However, the time series models performed worse than the CNNs.

CONCLUSION: Connected speech analysis is more effective than traditional vowel sound analysis for the diagnosis of laryngeal voice disorders. This study highlights the potential of AI technologies in enhancing the diagnostic capabilities of speech, advocating further exploration, and validation in this field.

PMID:38350806 | DOI:10.1016/j.jvoice.2024.01.015

Sci Total Environ. 2024 Feb 11:170804. doi: 10.1016/j.scitotenv.2024.170804. Online ahead of print.

ABSTRACT

As global greenhouse gas emissions increase and fossil energy sources decline dramatically, the energy transition is at the heart of many countries' development initiatives. As a biomass resource, straw plays a positive role in energy transformation and environmental improvement. However, there is still a challenge to explore the best options and models for straw production and utilization of green and efficient biomass energy in agricultural systems. This study establishes an economic-environmental-resource synergistic Straw Green recycling optimization model based on straw-electricity-biochar-biogas core (Straw Green recycling optimization model, SGROM). Firstly, we explore the effects of biochar return to the field on crop yield and greenhouse gas emission by Meta-analysis method, and on this basis, we construct SGROM to weigh the three objectives of economic-greenhouse gas emission-resource utilization, and explore the best allocation ratio between four utilization methods of straw: power generation, biochar preparation, biogas and derivatives preparation and sale, so as to obtain a straw recycling and efficient low-carbon utilization model. Exploring the response of straw green utilization patterns to crop market prices with the help of deep learning methods, SGROM has been applied to the main grain producing areas in the Sanjiang Plain of China, and the results of comparison with the traditional straw utilization (TSU) model show that the greenhouse gas emissions per unit of production value of SGROM are 19.66 % lower than that of TSU model, the electricity consumption is saved by 2.00 %, and the optimal ratios of straw for power generation, biogas and biochar production, and sale are 1.00 %, 10.75 %, 62.11 % and 26.14 %. The economic benefits and total greenhouse gas emissions of the integrated straw utilization mode are better than those of the single straw utilization mode, proving the superiority of SGROM in optimizing the straw utilization mode.

PMID:38350576 | DOI:10.1016/j.scitotenv.2024.170804

Comput Biol Med. 2024 Feb 3;171:108095. doi: 10.1016/j.compbiomed.2024.108095. Online ahead of print.

ABSTRACT

Gait abnormalities are frequent in children and can be caused by different pathologies, such as cerebral palsy, neuromuscular disease, toe walker syndrome, etc. Analysis of the "gait pattern" (i.e., the way the person walks) using 3D analysis provides highly relevant clinical information. This information is used to guide therapeutic choices; however, it is underused in diagnostic processes, probably because of the lack of standardization of data collection methods. Therefore, 3D gait analysis is currently used as an assessment rather than a diagnostic tool. In this work, we aimed to determine if deep learning could be combined with 3D gait analysis data to diagnose gait disorders in children. We tested the diagnostic accuracy of deep learning methods combined with 3D gait analysis data from 371 children (148 with unilateral cerebral palsy, 60 with neuromuscular disease, 19 toe walkers, 60 with bilateral cerebral palsy, 25 stroke, and 59 typically developing children), with a total of 6400 gait cycles. We evaluated the accuracy, sensitivity, specificity, F1 score, Area Under the Curve (AUC) score, and confusion matrix of the predictions by ResNet, LSTM, and InceptionTime deep learning architectures for time series data. The deep learning-based models had good to excellent diagnostic accuracy (ranging from 0.77 to 0.99) for discrimination between healthy and pathological gait, discrimination between different etiologies of pathological gait (binary and multi-classification); and determining stroke onset time. LSTM performed best overall. This study revealed that the gait pattern contains specific, pathology-related information. These results open the way for an extension of 3D gait analysis from evaluation to diagnosis. Furthermore, the method we propose is a data-driven diagnostic model that can be trained and used without human intervention or expert knowledge. Furthermore, the method could be used to distinguish gait-related pathologies and their onset times beyond those studied in this research.

PMID:38350399 | DOI:10.1016/j.compbiomed.2024.108095

Comput Biol Med. 2024 Feb 6;171:108102. doi: 10.1016/j.compbiomed.2024.108102. Online ahead of print.

ABSTRACT

The morphological analysis of cells from optical images is vital for interpreting brain function in disease states. Extracting comprehensive cell morphology from intricate backgrounds, common in neural and some medical images, poses a significant challenge. Due to the huge workload of manual recognition, automated neuron cell segmentation using deep learning algorithms with labeled data is integral to neural image analysis tools. To combat the high cost of acquiring labeled data, we propose a novel semi-supervised cell segmentation algorithm for immunofluorescence-stained cell image datasets (ISC), utilizing a mean-teacher semi-supervised learning framework. We include a "cross comparison representation learning block" to enhance the teacher-student model comparison on high-dimensional channels, thereby improving feature compactness and separability, which results in the extraction of higher-dimensional features from unlabeled data. We also suggest a new network, the Multi Pooling Layer Attention Dense Network (MPAD-Net), serving as the backbone of the student model to augment segmentation accuracy. Evaluations on the immunofluorescence staining datasets and the public CRAG dataset illustrate our method surpasses other top semi-supervised learning methods, achieving average Jaccard, Dice and Normalized Surface Dice (NSD) indicators of 83.22%, 90.95% and 81.90% with only 20% labeled data. The datasets and code are available on the website at https://github.com/Brainsmatics/CCRL.

PMID:38350398 | DOI:10.1016/j.compbiomed.2024.108102

Comput Biol Med. 2024 Feb 7;171:108127. doi: 10.1016/j.compbiomed.2024.108127. Online ahead of print.

ABSTRACT

Identifying drug-protein interactions (DPIs) is crucial in drug discovery and repurposing. Computational methods for precise DPI identification can expedite development timelines and reduce expenses compared with conventional experimental methods. Lately, deep learning techniques have been employed for predicting DPIs, enhancing these processes. Nevertheless, the limitations observed in prior studies, where many extract features from complete drug and protein entities, overlooking the crucial theoretical foundation that pharmacological responses are often correlated with specific substructures, can lead to poor predictive performance. Furthermore, certain substructure-focused research confines its exploration to a solitary fragment category, such as a functional group. In this study, addressing these constraints, we present an end-to-end framework termed BCMMDA for predicting DPIs. The framework considers various substructure types, including branch chains, common substructures, and specific fragments. We designed a specific feature learning module by combining our proposed multi-dimensional attention mechanism with convolutional neural networks (CNNs). Deep CNNs assist in capturing the synergistic effects among these fragment sets, enabling the extraction of relevant features of drugs and proteins. Meanwhile, the multi-dimensional attention mechanism refines the relationship between drug and protein features by assigning attention vectors to each drug compound and amino acid. This mechanism empowers the model to further concentrate on pivotal substructures and elements, thereby improving its ability to identify essential interactions in DPI prediction. We evaluated the performance of BCMMDA on four well-known benchmark datasets. The results indicated that BCMMDA outperformed state-of-the-art baseline models, demonstrating significant improvement in performance.

PMID:38350397 | DOI:10.1016/j.compbiomed.2024.108127

Sci Rep. 2024 Feb 13;14(1):3637. doi: 10.1038/s41598-024-54251-1.

ABSTRACT

Parkinson's disease is the world's fastest-growing neurological disorder. Research to elucidate the mechanisms of Parkinson's disease and automate diagnostics would greatly improve the treatment of patients with Parkinson's disease. Current diagnostic methods are expensive and have limited availability. Considering the insidious and preclinical onset and progression of the disease, a desirable screening should be diagnostically accurate even before the onset of symptoms to allow medical interventions. We highlight retinal fundus imaging, often termed a window to the brain, as a diagnostic screening modality for Parkinson's disease. We conducted a systematic evaluation of conventional machine learning and deep learning techniques to classify Parkinson's disease from UK Biobank fundus imaging. Our results suggest Parkinson's disease individuals can be differentiated from age and gender-matched healthy subjects with 68% accuracy. This accuracy is maintained when predicting either prevalent or incident Parkinson's disease. Explainability and trustworthiness are enhanced by visual attribution maps of localized biomarkers and quantified metrics of model robustness to data perturbations.

PMID:38351326 | DOI:10.1038/s41598-024-54251-1

J Imaging Inform Med. 2024 Feb 13. doi: 10.1007/s10278-024-00995-1. Online ahead of print.

ABSTRACT

The early detection and accurate diagnosis of liver fibrosis, a progressive and potentially serious liver condition, are crucial for effective medical intervention. Invasive methods like biopsies for diagnosis can be risky and expensive. This research presents a novel computer-aided diagnosis model for liver fibrosis using a hybrid approach of minimum redundancy maximum relevance (MRMR) feature selection, bidirectional long short-term memory (BiLSTM), and convolutional neural networks (CNN). The proposed model involves multiple stages, including image acquisition, preprocessing, feature representation, fibrous tissue identification, and classification. Notably, histogram equalization is employed to enhance image quality by addressing variations in brightness levels. Performance evaluation encompasses a range of metrics such as accuracy, precision, sensitivity, specificity, F1 score, and error rate. Comparative analyses with established methods like DCNN, ANN-FLI, LungNet22, and SDAE-GAN underscore the efficacy of the proposed model. The innovative integration of hybrid MRMR-BiLSTM-CNN architecture and the horse herd optimization algorithm significantly enhances accuracy and F1 score, even with small datasets. The model tackles the complexities of hyperparameter optimization through the IHO algorithm and reduces training time by leveraging MRMR feature selection. In practical application, the proposed hybrid MRMR-BiLSTM-CNN method demonstrates remarkable performance with a 97.8% accuracy rate in identifying liver fibrosis images. It exhibits high precision, sensitivity, specificity, and minimal error rate, showcasing its potential for accurate and non-invasive diagnosis.

PMID:38351226 | DOI:10.1007/s10278-024-00995-1

Sci Rep. 2024 Feb 13;14(1):3630. doi: 10.1038/s41598-024-53345-0.

ABSTRACT

Urban park management assessment is critical to park operation and service quality. Traditional assessment methods cannot comprehensively assess park use and environmental conditions. Besides, although social media and big data have shown significant advantages in understanding public behavior or preference and park features or values, there has been little relevant research on park management assessment. This study proposes a deep learning-based framework for assessing urban park intelligent management from macro to micro levels with comment data from social media. By taking seven parks in Wuhan City as the objects, this study quantitatively assesses their overall state and performance in facilities, safety, environment, activities, and services, and reveals their main problems in management. The results demonstrate the impacts of various factors, including park type, season, and specific events such as remodeling and refurbishment, on visitor satisfaction and the characteristics of individual parks and their management. Compared with traditional methods, this framework enables real-time intelligent assessment of park management, which can accurately reflect park use and visitor feedback, and improve park service quality and management efficiency. Overall, this study provides important reference for intelligent park management assessment based on big data and artificial intelligence, which can facilitate the future development of smart cities.

PMID:38351201 | DOI:10.1038/s41598-024-53345-0

Sci Rep. 2024 Feb 14;14(1):3666. doi: 10.1038/s41598-024-53988-z.

ABSTRACT

EDXRF spectrometry is a well-established and often-used analytical technique in examining materials from which cultural heritage objects are made. The analytical results are traditionally subjected to additional multivariate analysis for archaeometry studies to reduce the initial data's dimensionality based on informative features. Nowadays, artificial intelligence (AI) techniques are used more for this purpose. Different soft computing techniques are used to improve speed and accuracy. Choosing the most suitable AI method can increase the sustainability of the analytical process and postprocessing activities. An autoencoder neural network has been designed and used as a dimension reduction tool of initial [Formula: see text] data collected in the raw EDXRF spectra, containing information about the selected points' elemental composition on the canvas paintings' surface. The autoencoder network design enables the best possible reconstruction of the original EDXRF spectrum and the most informative feature extraction, which has been used for dimension reduction. Such configuration allows for efficient classification algorithms and their performances. The autoencoder neural network approach is more sustainable, especially in processing time consumption and experts' manual work.

PMID:38351176 | DOI:10.1038/s41598-024-53988-z

Sci Rep. 2024 Feb 13;14(1):3619. doi: 10.1038/s41598-024-53411-7.

ABSTRACT

We present a novel automatic preprocessing and ensemble learning technique for the segmentation of low-quality cell images. Capturing cells subjected to intense light is challenging due to their vulnerability to light-induced cell death. Consequently, microscopic cell images tend to be of low quality and it causes low accuracy for semantic segmentation. This problem can not be satisfactorily solved by classical image preprocessing methods. Therefore, we propose a novel approach of automatic enhancement preprocessing (AEP), which translates an input image into images that are easy to recognize by deep learning. AEP is composed of two deep neural networks, and the penultimate feature maps of the first network are employed as filters to translate an input image with low quality into images that are easily classified by deep learning. Additionally, we propose an automatic weighted ensemble learning (AWEL), which combines the multiple segmentation results. Since the second network predicts segmentation results corresponding to each translated input image, multiple segmentation results can be aggregated by automatically determining suitable weights. Experiments on two types of cell image segmentation confirmed that AEP can translate low-quality cell images into images that are easy to segment and that segmentation accuracy improves using AWEL.

PMID:38351053 | DOI:10.1038/s41598-024-53411-7

J Neuroeng Rehabil. 2024 Feb 13;21(1):24. doi: 10.1186/s12984-024-01320-1.

ABSTRACT

BACKGROUND: Freezing of gait (FOG) is an episodic and highly disabling symptom of Parkinson's Disease (PD). Traditionally, FOG assessment relies on time-consuming visual inspection of camera footage. Therefore, previous studies have proposed portable and automated solutions to annotate FOG. However, automated FOG assessment is challenging due to gait variability caused by medication effects and varying FOG-provoking tasks. Moreover, whether automated approaches can differentiate FOG from typical everyday movements, such as volitional stops, remains to be determined. To address these questions, we evaluated an automated FOG assessment model with deep learning (DL) based on inertial measurement units (IMUs). We assessed its performance trained on all standardized FOG-provoking tasks and medication states, as well as on specific tasks and medication states. Furthermore, we examined the effect of adding stopping periods on FOG detection performance.

METHODS: Twelve PD patients with self-reported FOG (mean age 69.33 ± 6.02 years) completed a FOG-provoking protocol, including timed-up-and-go and 360-degree turning-in-place tasks in On/Off dopaminergic medication states with/without volitional stopping. IMUs were attached to the pelvis and both sides of the tibia and talus. A temporal convolutional network (TCN) was used to detect FOG episodes. FOG severity was quantified by the percentage of time frozen (%TF) and the number of freezing episodes (#FOG). The agreement between the model-generated outcomes and the gold standard experts' video annotation was assessed by the intra-class correlation coefficient (ICC).

RESULTS: For FOG assessment in trials without stopping, the agreement of our model was strong (ICC (%TF) = 0.92 [0.68, 0.98]; ICC(#FOG) = 0.95 [0.72, 0.99]). Models trained on a specific FOG-provoking task could not generalize to unseen tasks, while models trained on a specific medication state could generalize to unseen states. For assessment in trials with stopping, the agreement of our model was moderately strong (ICC (%TF) = 0.95 [0.73, 0.99]; ICC (#FOG) = 0.79 [0.46, 0.94]), but only when stopping was included in the training data.

CONCLUSION: A TCN trained on IMU signals allows valid FOG assessment in trials with/without stops containing different medication states and FOG-provoking tasks. These results are encouraging and enable future work investigating automated FOG assessment during everyday life.

PMID:38350964 | DOI:10.1186/s12984-024-01320-1

Fluids Barriers CNS. 2024 Feb 13;21(1):15. doi: 10.1186/s12987-024-00516-w.

ABSTRACT

BACKGROUND: Peri-sinus structures such as arachnoid granulations (AG) and the parasagittal dural (PSD) space have gained much recent attention as sites of cerebral spinal fluid (CSF) egress and neuroimmune surveillance. Neurofluid circulation dysfunction may manifest as morphological changes in these structures, however, automated quantification of these structures is not possible and rather characterization often requires exogenous contrast agents and manual delineation.

METHODS: We propose a deep learning architecture to automatically delineate the peri-sinus space (e.g., PSD and intravenous AG structures) using two cascaded 3D fully convolutional neural networks applied to submillimeter 3D T2-weighted non-contrasted MRI images, which can be routinely acquired on all major MRI scanner vendors. The method was evaluated through comparison with gold-standard manual tracing from a neuroradiologist (n = 80; age range = 11-83 years) and subsequently applied in healthy participants (n = 1,872; age range = 5-100 years), using data from the Human Connectome Project, to provide exemplar metrics across the lifespan. Dice-Sørensen and a generalized linear model was used to assess PSD and AG changes across the human lifespan using quadratic restricted splines, incorporating age and sex as covariates.

RESULTS: Findings demonstrate that the PSD and AG volumes can be segmented using T2-weighted MRI with a Dice-Sørensen coefficient and accuracy of 80.7 and 74.6, respectively. Across the lifespan, we observed that total PSD volume increases with age with a linear interaction of gender and age equal to 0.9 cm3 per year (p < 0.001). Similar trends were observed in the frontal and parietal, but not occipital, PSD. An increase in AG volume was observed in the third to sixth decades of life, with a linear effect of age equal to 0.64 mm3 per year (p < 0.001) for total AG volume and 0.54 mm3 (p < 0.001) for maximum AG volume.

CONCLUSIONS: A tool that can be applied to quantify PSD and AG volumes from commonly acquired T2-weighted MRI scans is reported and exemplar volumetric ranges of these structures are provided, which should provide an exemplar for studies of neurofluid circulation dysfunction. Software and training data are made freely available online ( https://github.com/hettk/spesis ).

PMID:38350930 | DOI:10.1186/s12987-024-00516-w

BMC Med Imaging. 2024 Feb 13;24(1):43. doi: 10.1186/s12880-024-01217-4.

ABSTRACT

BACKGROUND: A deep learning (DL) model that automatically detects cardiac pathologies on cardiac MRI may help streamline the diagnostic workflow. To develop a DL model to detect cardiac pathologies on cardiac MRI T1-mapping and late gadolinium phase sensitive inversion recovery (PSIR) sequences were used.

METHODS: Subjects in this study were either diagnosed with cardiac pathology (n = 137) including acute and chronic myocardial infarction, myocarditis, dilated cardiomyopathy, and hypertrophic cardiomyopathy or classified as normal (n = 63). Cardiac MR imaging included T1-mapping and PSIR sequences. Subjects were split 65/15/20% for training, validation, and hold-out testing. The DL models were based on an ImageNet pretrained DenseNet-161 and implemented using PyTorch and fastai. Data augmentation with random rotation and mixup was applied. Categorical cross entropy was used as the loss function with a cyclic learning rate (1e-3). DL models for both sequences were developed separately using similar training parameters. The final model was chosen based on its performance on the validation set. Gradient-weighted class activation maps (Grad-CAMs) visualized the decision-making process of the DL model.

RESULTS: The DL model achieved a sensitivity, specificity, and accuracy of 100%, 38%, and 88% on PSIR images and 78%, 54%, and 70% on T1-mapping images. Grad-CAMs demonstrated that the DL model focused its attention on myocardium and cardiac pathology when evaluating MR images.

CONCLUSIONS: The developed DL models were able to reliably detect cardiac pathologies on cardiac MR images. The diagnostic performance of T1 mapping alone is particularly of note since it does not require a contrast agent and can be acquired quickly.

PMID:38350900 | DOI:10.1186/s12880-024-01217-4

BMC Bioinformatics. 2024 Feb 14;25(1):69. doi: 10.1186/s12859-024-05679-9.

ABSTRACT

BACKGROUND: Technological advances have enabled the generation of unique and complementary types of data or views (e.g. genomics, proteomics, metabolomics) and opened up a new era in multiview learning research with the potential to lead to new biomedical discoveries.

RESULTS: We propose iDeepViewLearn (Interpretable Deep Learning Method for Multiview Learning) to learn nonlinear relationships in data from multiple views while achieving feature selection. iDeepViewLearn combines deep learning flexibility with the statistical benefits of data and knowledge-driven feature selection, giving interpretable results. Deep neural networks are used to learn view-independent low-dimensional embedding through an optimization problem that minimizes the difference between observed and reconstructed data, while imposing a regularization penalty on the reconstructed data. The normalized Laplacian of a graph is used to model bilateral relationships between variables in each view, therefore, encouraging selection of related variables. iDeepViewLearn is tested on simulated and three real-world data for classification, clustering, and reconstruction tasks. For the classification tasks, iDeepViewLearn had competitive classification results with state-of-the-art methods in various settings. For the clustering task, we detected molecular clusters that differed in their 10-year survival rates for breast cancer. For the reconstruction task, we were able to reconstruct handwritten images using a few pixels while achieving competitive classification accuracy. The results of our real data application and simulations with small to moderate sample sizes suggest that iDeepViewLearn may be a useful method for small-sample-size problems compared to other deep learning methods for multiview learning.

CONCLUSION: iDeepViewLearn is an innovative deep learning model capable of capturing nonlinear relationships between data from multiple views while achieving feature selection. It is fully open source and is freely available at https://github.com/lasandrall/iDeepViewLearn .

PMID:38350879 | DOI:10.1186/s12859-024-05679-9

Acad Radiol. 2024 Feb 12:S1076-6332(24)00004-7. doi: 10.1016/j.acra.2024.01.004. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: To develop and validate a deep learning (DL)-based method for pancreas segmentation on CT and automatic measurement of pancreatic volume in pancreatic cancer.

MATERIALS AND METHODS: This retrospective study used 3D nnU-net architecture for fully automated pancreatic segmentation in patients with pancreatic cancer. The study used 851 portal venous phase CT images (499 pancreatic cancer and 352 normal pancreas). This dataset was divided into training (n = 506), internal validation (n = 126), and external test set (n = 219). For the external test set, the pancreas was manually segmented by two abdominal radiologists (R1 and R2) to obtain the ground truth. In addition, the consensus segmentation was obtained using Simultaneous Truth and Performance Level Estimation (STAPLE) algorithm. Segmentation performance was assessed using the Dice similarity coefficient (DSC). Next, the pancreatic volumes determined by automatic segmentation were compared to those determined by manual segmentation by two radiologists.

RESULTS: The DL-based model for pancreatic segmentation showed a mean DSC of 0.764 in the internal validation dataset and DSC of 0.807, 0.805, and 0.803 using R1, R2, and STAPLE as references in the external test dataset. The pancreas parenchymal volume measured by automatic and manual segmentations were similar (DL-based model: 65.5 ± 19.3 cm3 and STAPLE: 65.1 ± 21.4 cm3; p = 0.486). The pancreatic parenchymal volume difference between the DL-based model predictions and the manual segmentation by STAPLE was 0.5 cm3, with correlation coefficients of 0.88.

CONCLUSION: The DL-based model efficiently generates automatic segmentation of the pancreas and measures the pancreatic volume in patients with pancreatic cancer.

PMID:38350812 | DOI:10.1016/j.acra.2024.01.004

Med Image Anal. 2024 Feb 8;93:103104. doi: 10.1016/j.media.2024.103104. Online ahead of print.

ABSTRACT

Automated lesion detection in retinal optical coherence tomography (OCT) scans has shown promise for several clinical applications, including diagnosis, monitoring and guidance of treatment decisions. However, segmentation models still struggle to achieve the desired results for some complex lesions or datasets that commonly occur in real-world, e.g. due to variability of lesion phenotypes, image quality or disease appearance. While several techniques have been proposed to improve them, one line of research that has not yet been investigated is the incorporation of additional semantic context through the application of anomaly detection models. In this study we experimentally show that incorporating weak anomaly labels to standard segmentation models consistently improves lesion segmentation results. This can be done relatively easy by detecting anomalies with a separate model and then adding these output masks as an extra class for training the segmentation model. This provides additional semantic context without requiring extra manual labels. We empirically validated this strategy using two in-house and two publicly available retinal OCT datasets for multiple lesion targets, demonstrating the potential of this generic anomaly guided segmentation approach to be used as an extra tool for improving lesion detection models.

PMID:38350222 | DOI:10.1016/j.media.2024.103104

Comput Methods Programs Biomed. 2024 Feb 1;246:108054. doi: 10.1016/j.cmpb.2024.108054. Online ahead of print.

ABSTRACT

Deep-learning-based medical image segmentation techniques can assist doctors in disease diagnosis and rapid treatment. However, existing medical image segmentation models do not fully consider the dependence between feature segments in the feature extraction process, and the correlated features can be further extracted. Therefore, a recurrent positional encoding circular attention mechanism network (RPECAMNet) is proposed based on relative positional encoding for medical image segmentation. Multiple residual modules are used to extract the primary features of the medical images, which are thereafter converted into one-dimensional data for relative positional encoding. The recursive former is used to further extract features from medical images, and decoding is performed using deconvolution. An adaptive loss function is designed to train the model and achieve accurate medical-image segmentation. Finally, the proposed model is used to conduct comparative experiments on the synapse and self-constructed kidney datasets to verify the accuracy of the proposed model for medical image segmentation.

PMID:38350190 | DOI:10.1016/j.cmpb.2024.108054

Comput Methods Programs Biomed. 2024 Feb 8;246:108060. doi: 10.1016/j.cmpb.2024.108060. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVE: Vital sign monitoring in the Intensive Care Unit (ICU) is crucial for enabling prompt interventions for patients. This underscores the need for an accurate predictive system. Therefore, this study proposes a novel deep learning approach for forecasting Heart Rate (HR), Systolic Blood Pressure (SBP), and Diastolic Blood Pressure (DBP) in the ICU.

METHODS: We extracted 24,886 ICU stays from the MIMIC-III database which contains data from over 46 thousand patients, to train and test the model. The model proposed in this study, Transformer-based Diffusion Probabilistic Model for Sparse Time Series Forecasting (TDSTF), merges Transformer and diffusion models to forecast vital signs. The TDSTF model showed state-of-the-art performance in predicting vital signs in the ICU, outperforming other models' ability to predict distributions of vital signs and being more computationally efficient. The code is available at https://github.com/PingChang818/TDSTF.

RESULTS: The results of the study showed that TDSTF achieved a Standardized Average Continuous Ranked Probability Score (SACRPS) of 0.4438 and a Mean Squared Error (MSE) of 0.4168, an improvement of 18.9% and 34.3% over the best baseline model, respectively. The inference speed of TDSTF is more than 17 times faster than the best baseline model.

CONCLUSION: TDSTF is an effective and efficient solution for forecasting vital signs in the ICU, and it shows a significant improvement compared to other models in the field.

PMID:38350189 | DOI:10.1016/j.cmpb.2024.108060

Comput Methods Programs Biomed. 2024 Feb 1;246:108052. doi: 10.1016/j.cmpb.2024.108052. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVE: Atrial Fibrillation (AF) is a supraventricular tachyarrhythmia that can lead to thromboembolism, hearlt failure, ischemic stroke, and a decreased quality of life. Characterizing the locations where the mechanisms of AF are initialized and maintained is key to accomplishing an effective ablation of the targets, hence restoring sinus rhythm. Many methods have been investigated to locate such targets in a non-invasive way, such as Electrocardiographic Imaging, which enables an on-invasive and panoramic characterization of cardiac electrical activity using recording Body Surface Potentials (BSP) and a torso model of the patient. Nonetheless, this technique entails some major issues stemming from solving the inverse problem, which is known to be severely ill-posed. In this context, many machine learning and deep learning approaches aim to tackle the characterization and classification of AF targets to improve AF diagnosis and treatment.

METHODS: In this work, we propose a method to locate AF drivers as a supervised classification problem. We employed a hybrid form of the convolutional-recurrent network which enables feature extraction and sequential data modeling utilizing labeled realistic computerized AF models. Thus, we used 16 AF electrograms, 1 atrium, and 10 torso geometries to compute the forward problem. Previously, the AF models were labeled by assigning each sample of the signals a region from the atria from 0 (no driver) to 7, according to the spatial location of the AF driver. The resulting 160 BSP signals, which resemble a 64-lead vest recording, are preprocessed and then introduced into the network following a 4-fold cross-validation in batches of 50 samples.

RESULTS: The results show a mean accuracy of 74.75% among the 4 folds, with a better performance in detecting sinus rhythm, and drivers near the left superior pulmonary vein (R1), and right superior pulmonary vein (R3) whose mean sensitivity bounds around 84%-87%. Significantly good results are obtained in mean sensitivity (87%) and specificity (83%) in R1.

CONCLUSIONS: Good results in R1 are highly convenient since AF drivers are commonly found in this area: the left atrial appendage, as suggested in some previous studies. These promising results indicate that using CNN-LSTM networks could lead to new strategies exploiting temporal correlations to address this challenge effectively.

PMID:38350188 | DOI:10.1016/j.cmpb.2024.108052

J Nat Prod. 2024 Feb 13. doi: 10.1021/acs.jnatprod.3c00879. Online ahead of print.

ABSTRACT

Many machine learning techniques are used as drug discovery tools with the intent to speed characterization by determining relationships between compound structure and biological function. However, particularly in anticancer drug discovery, these models often make only binary decisions about the biological activity for a narrow scope of drug targets. We present a feed-forward neural network, PECAN (Prediction Engine for the Cytostatic Activity of Natural product-like compounds), that simultaneously classifies the potential antiproliferative activity of compounds against 59 cancer cell lines. It predicts the activity to be one of six categories, indicating not only if activity is present but the degree of activity. Using an independent subset of NCI data as a test set, we show that PECAN can reach 60.1% accuracy in a six-way classification and present further evidence that it classifies based on useful structural features of compounds using a "within-one" measure that reaches 93.0% accuracy.

PMID:38349959 | DOI:10.1021/acs.jnatprod.3c00879