Diagn Interv Radiol. 2024 Sep 2. doi: 10.4274/dir.2024.242790. Online ahead of print.

ABSTRACT

PURPOSE: This study aimed to evaluate the validity of two artificial intelligence (AI)-based bone age assessment programs, BoneXpert and VUNO Med-Bone Age (VUNO), compared with manual assessments using the Greulich-Pyle method in Turkish children.

METHODS: This study included a cohort of 292 pediatric cases, ranging in age from 1 to 15 years with an equal gender and number distribution in each age group. Two radiologists, who were unaware of the bone age determined by AI, independently evaluated the bone age. The statistical study involved using the intraclass correlation coefficient (ICC) to measure the level of agreement between the manual and AI-based assessments.

RESULTS: The ICC coefficients for the agreement between the manual measurements of two radiologists indicate almost perfect agreement. When all cases, regardless of gender and age group, were analyzed, a nearly perfect positive agreement was observed between the manual and software measurements. When bone age calculations were separated and analyzed separately for girls and boys, there was no statistically significant difference between the two AI-based methods for boys; however, ICC coefficients of 0.990 and 0.982 were calculated for VUNO and BoneXpert, respectively, and this difference of 0.008 was significant (z = 2.528, P = 0.012) for girls. Accordingly, VUNO showed higher agreement with manual measurements compared with BoneXpert. The difference between the agreements demonstrated by the two software packages with manual measurements in the prepubescent group was much more pronounced in girls compared with boys. After the age of 8 years for girls and 9 years for boys, the agreement between manual measurements and both AI software packages was equal.

CONCLUSION: Both BoneXpert and VUNO showed high validity in assessing bone age. Furthermore, VUNO has a statistically higher correlation with manual assessment in prepubertal girls. These results suggest that VUNO may be slightly more effective in determining bone age, indicating its potential as a highly reliable tool for bone age assessment in Turkish children.

CLINICAL SIGNIFICANCE: Investigating the most suitable AI program for the Turkish population could be clinically significant.

PMID:39221654 | DOI:10.4274/dir.2024.242790

Magn Reson Med. 2024 Sep 2. doi: 10.1002/mrm.30283. Online ahead of print.

ABSTRACT

PURPOSE: To develop an automated deep learning model for MRI-based segmentation and detection of intracranial arterial calcification.

METHODS: A novel deep learning model under the variational autoencoder framework was developed. A theoretically grounded dissimilarity loss was proposed to refine network features extracted from MRI and restrict their complexity, enabling the model to learn more generalizable MR features that enhance segmentation accuracy and robustness for detecting calcification on MRI.

RESULTS: The proposed method was compared with nine baseline methods on a dataset of 113 subjects and showed superior performance (for segmentation, Dice similarity coefficient: 0.620, area under precision-recall curve [PR-AUC]: 0.660, 95% Hausdorff Distance: 0.848 mm, Average Symmetric Surface Distance: 0.692 mm; for slice-wise detection, F1 score: 0.823, recall: 0.764, precision: 0.892, PR-AUC: 0.853). For clinical needs, statistical tests confirmed agreement between the true calcification volumes and predicted values using the proposed approach. Various MR sequences, namely T1, time-of-flight, and SNAP, were assessed as inputs to the model, and SNAP provided unique and essential information pertaining to calcification structures.

CONCLUSION: The proposed deep learning model with a dissimilarity loss to reduce feature complexity effectively improves MRI-based identification of intracranial arterial calcification. It could help establish a more comprehensive and powerful pipeline for vascular image analysis on MRI.

PMID:39221515 | DOI:10.1002/mrm.30283

PeerJ. 2024 Aug 29;12:e17896. doi: 10.7717/peerj.17896. eCollection 2024.

ABSTRACT

Ground reaction force (GRF) data is often collected for the biomechanical analysis of running, due to the performance and injury risk insights that GRF analysis can provide. Traditional methods typically limit GRF collection to controlled lab environments, recent studies have looked to combine the ease of use of wearable sensors with the statistical power of machine learning to estimate continuous GRF data outside of these restrictions. Before such systems can be deployed with confidence outside of the lab they must be shown to be a valid and accurate tool for a wide range of users. The aim of this study was to evaluate how accurately a consumer-priced sensor system could estimate GRFs whilst a heterogeneous group of runners completed a treadmill protocol with three different personalised running speeds and three gradients. Fifty runners (25 female, 25 male) wearing pressure insoles made up of 16 resistive sensors and an inertial measurement unit ran at various speeds and gradients on an instrumented treadmill. A long short term memory (LSTM) neural network was trained to estimate both vertical ( G R F v ) and anteroposterior ( G R F a p ) force traces using leave one subject out validation. The average relative root mean squared error (rRMSE) was 3.2% and 3.1%, respectively. The mean ( G R F v ) rRMSE across the evaluated participants ranged from 0.8% to 8.8% and from 1.3% to 17.3% in the ( G R F a p ) estimation. The findings from this study suggest that current consumer-priced sensors could be used to accurately estimate two-dimensional GRFs for a wide range of runners at a variety of running intensities. The estimated kinetics could be used to provide runners with individualised feedback as well as form the basis of data collection for running injury risk factor studies on a much larger scale than is currently possible with lab based methods.

PMID:39221284 | PMC:PMC11366233 | DOI:10.7717/peerj.17896

Mater Today Bio. 2024 Aug 13;28:101201. doi: 10.1016/j.mtbio.2024.101201. eCollection 2024 Oct.

ABSTRACT

Label-free three-dimensional imaging plays a crucial role in unraveling the complexities of cellular functions and interactions in biomedical research. Conventional single-cell optical tomography techniques offer affordability and the convenience of bypassing laborious cell labelling protocols. However, these methods are encumbered by restricted illumination scanning ranges on abaxial plane, resulting in the loss of intricate cellular imaging details. The ability to fully control cellular rotation across all angles has emerged as an optimal solution for capturing comprehensive structural details of cells. Here, we introduce a label-free, cost-effective, and readily fabricated contactless acoustic-induced vibration system, specifically designed to enable multi-degree-of-freedom rotation of cells, ultimately attaining stable in-situ rotation. Furthermore, by integrating this system with advanced deep learning technologies, we perform 3D reconstruction and morphological analysis on diverse cell types, thus validating groups of high-precision cell identification. Notably, long-term observation of cells reveals distinct features associated with drug-induced apoptosis in both cancerous and normal cells populations. This methodology, based on deep learning-enabled cell 3D reconstruction, charts a novel trajectory for groups of real-time cellular visualization, offering promising advancements in the realms of drug screening and post-single-cell analysis, thereby addressing potential clinical requisites.

PMID:39221213 | PMC:PMC11364901 | DOI:10.1016/j.mtbio.2024.101201

Biomed Eng Comput Biol. 2024 Aug 30;15:11795972241277081. doi: 10.1177/11795972241277081. eCollection 2024.

ABSTRACT

AIM: The Insilco study uses deep learning algorithms to predict the protein-coding pg m RNA sequences.

MATERIAL AND METHODS: The NCBI GEO DATA SET GSE218606's GEO R tool discovered P.G's outer membrane vesicles' most differentially expressed mRNA. Genemania analyzed differentially expressed gene networks. Transcriptomics data were collected and labeled on P. gingivalis protein-coding mRNA sequence and pseudogene, lincRNA, and bidirectional promoter lincRNA. Orange, a machine learning tool, analyzed and predicted data after preprocessing. Naïve Bayes, neural networks, and gradient descent partition data into training and testing sets, yielding accurate results. Cross-validation, model accuracy, and ROC curve were evaluated after model validation.

RESULTS: Three models, Neural Networks, Naive Bayes, and Gradient Boosting, were evaluated using metrics like Area Under the Curve (AUC), Classification Accuracy (CA), F1 Score, Precision, Recall, and Specificity. Gradient Boosting achieved a balanced performance (AUC: 0.72, CA: 0.41, F1: 0.32) compared to Neural Networks (AUC: 0.721, CA: 0.391, F1: 0.314) and Naive Bayes (AUC: 0.701, CA: 0.172, F1: 0.114). While statistical tests revealed no significant differences between the models, Gradient Boosting exhibited a more balanced precision-recall relationship.

CONCLUSION: In silico analysis using machine learning techniques successfully predicted protein-coding mRNA sequences within Porphyromonas gingivalis OMVs. Gradient Boosting outperformed other models (Neural Networks, Naive Bayes) by achieving a balanced performance across metrics like AUC, classification accuracy, and precision-recall, suggests its potential as a reliable tool for protein-coding mRNA prediction in P. gingivalis OMVs.

PMID:39221175 | PMC:PMC11365027 | DOI:10.1177/11795972241277081

Front Neurosci. 2024 Aug 16;18:1352129. doi: 10.3389/fnins.2024.1352129. eCollection 2024.

ABSTRACT

BACKGROUND: Mild cognitive impairment is a heterogeneous syndrome. The heterogeneity of the syndrome and the absence of consensus limited the advancement of MCI. The purpose of our research is to create a visual framework of the last decade, highlight the hotspots of current research, and forecast the most fruitful avenues for future MCI research.

METHODS: We collected all the MCI-related literature published between 1 January 2013, and 24 April 2023, on the "Web of Science." The visual graph was created by the CiteSpace and VOSviewer. The current research hotspots and future research directions are summarized through the analysis of keywords and co-cited literature.

RESULTS: There are 6,075 articles were included in the final analysis. The number of publications shows an upward trend, especially after 2018. The United States and the University of California System are the most prolific countries and institutions, respectively. Petersen is the author who ranks first in terms of publication volume and influence. Journal of Alzheimer's Disease was the most productive journal. "neuroimaging," "fluid markers," and "predictors" are the focus of current research, and "machine learning," "electroencephalogram," "deep learning," and "blood biomarkers" are potential research directions in the future.

CONCLUSION: The cognition of MCI has been continuously evolved and renewed by multiple countries' joint efforts in the past decade. Hotspots for current research are on diagnostic biomarkers, such as fluid markers, neuroimaging, and so on. Future hotspots might be focused on the best prognostic and diagnostic models generated by machine learning and large-scale screening tools such as EEG and blood biomarkers.

PMID:39221008 | PMC:PMC11361971 | DOI:10.3389/fnins.2024.1352129

Heliyon. 2024 Aug 6;10(16):e35865. doi: 10.1016/j.heliyon.2024.e35865. eCollection 2024 Aug 30.

ABSTRACT

The digital era has expanded social exposure with easy internet access for mobile users, allowing for global communication. Now, people can get to know what is going on around the globe with just a click; however, this has also resulted in the issue of fake news. Fake news is content that pretends to be true but is actually false and is disseminated to defraud. Fake news poses a threat to harmony, politics, the economy, and public opinion. As a result, bogus news detection has become an emerging research domain to identify a given piece of text as genuine or fraudulent. In this paper, a new framework called Generative Bidirectional Encoder Representations from Transformers (GBERT) is proposed that leverages a combination of Generative pre-trained transformer (GPT) and Bidirectional Encoder Representations from Transformers (BERT) and addresses the fake news classification problem. This framework combines the best features of both cutting-edge techniques-BERT's deep contextual understanding and the generative capabilities of GPT-to create a comprehensive representation of a given text. Both GPT and BERT are fine-tuned on two real-world benchmark corpora and have attained 95.30 % accuracy, 95.13 % precision, 97.35 % sensitivity, and a 96.23 % F1 score. The statistical test results indicate the effectiveness of the fine-tuned framework for fake news detection and suggest that it can be a promising approach for eradicating this global issue of fake news in the digital landscape.

PMID:39220956 | PMC:PMC11365402 | DOI:10.1016/j.heliyon.2024.e35865

Heliyon. 2024 Aug 8;10(16):e35769. doi: 10.1016/j.heliyon.2024.e35769. eCollection 2024 Aug 30.

ABSTRACT

Angiogenesis is an essential process in tumorigenesis, tumor invasion, and metastasis, and is an intriguing pathway for drug discovery. Targeting vascular endothelial growth factor receptor 2 (VEGFR2) to inhibit tumor angiogenic pathways has been widely explored and adopted in clinical practice. However, most drugs, such as the Food and Drug Administration -approved drug axitinib (ATC code: L01EK01), have considerable side effects and limited tolerability. Therefore, there is an urgent need for the development of novel VEGFR2 inhibitors. In this study, we propose a novel strategy to design potential candidates targeting VEGFR2 using three-dimensional (3D) deep learning and structural modeling methods. A geometric-enhanced molecular representation learning method (GEM) model employing a graph neural network (GNN) as its underlying predictive algorithm was used to predict the activity of the candidates. In the structural modeling method, flexible docking was performed to screen data with high affinity and explore the mechanism of the inhibitors. Small -molecule compounds with consistently improved properties were identified based on the intersection of the scores obtained from both methods. Candidates identified using the GEM-GNN model were selected for in silico modeling using molecular dynamics simulations to further validate their efficacy. The GEM-GNN model enabled the identification of candidate compounds with potentially more favorable properties than the existing drug, axitinib, while achieving higher efficacy.

PMID:39220924 | PMC:PMC11365333 | DOI:10.1016/j.heliyon.2024.e35769

JDS Commun. 2023 Nov 17;5(4):310-316. doi: 10.3168/jdsc.2023-0442. eCollection 2024 Jul.

ABSTRACT

Respiratory rate (RR) is commonly employed for identifying animals experiencing heat-stress conditions and respiratory diseases. Recent advancements in computer vision algorithms have enabled the estimation of the RR in dairy cows through image-based approaches, with a primary focus on standing positions, thermal imaging, and deep learning techniques. In this study, our objective was to develop a system capable of accurately predicting the RR of lying Holstein cows under unrestrained conditions using red, green, and blue (RGB) and infrared (IR) night vision images. Thirty lactating cows were continuously recorded for 12 h per day over a 3-d period, capturing at least one 30-s video segment of each cow during lying time. A total of 95 videos were manually annotated with rectangular bounding boxes encompassing the flank area (region of interest; ROI) of the lying cows. For future applications, we trained a model for ROI identification using YOLOv8 to avoid manual annotations. The observed RR was determined by visual counting of breaths in each video. To predict the RR, we devised an image processing pipeline involving (1) capturing the ROI for the entire video, (2) reshaping the pixel intensity of each image channel into a 2-dimensional object and calculating its per-frame mean, (3) applying fast Fourier transform (FFT) to the average pixel intensity vector, (4) filtering frequencies specifically associated with respiratory movements, and (5) executing inverse FFT on the denoized data and identifying peaks on the resulting plot, with the count of peaks serving as the predicted RR per minute. The evaluation metrics, root mean squared error of prediction (RMSEP) and R2, yielded values of 8.3 breaths/min (17.1% of the mean RR) and 0.77, respectively. To further validate the method, an additional dataset comprising preweaning dairy calves was used, consisting of 42 observations from 25 calves. The RMSEP and R2 values for this dataset were 13.0 breaths/min and 0.73, respectively. The model trained to identify the ROI exhibited a precision of 100%, a recall of 71.8%, and an F 1 score of 83.6% for bounding box detection. These are promising results for the implementation of this pipeline in future studies. The application of FFT to signals acquired from both RGB and IR images proved to be an effective and accurate method for computing the RR of cows in unrestrained conditions.

PMID:39220844 | PMC:PMC11365218 | DOI:10.3168/jdsc.2023-0442

Stat (Int Stat Inst). 2024;13(1):e649. doi: 10.1002/sta4.649. Epub 2024 Feb 7.

ABSTRACT

Glaucoma is a major cause of blindness and vision impairment worldwide, and visual field (VF) tests are essential for monitoring the conversion of glaucoma. While previous studies have primarily focused on using VF data at a single time point for glaucoma prediction, there has been limited exploration of longitudinal trajectories. Additionally, many deep learning techniques treat the time-to-glaucoma prediction as a binary classification problem (glaucoma Yes/No), resulting in the misclassification of some censored subjects into the nonglaucoma category and decreased power. To tackle these challenges, we propose and implement several deep-learning approaches that naturally incorporate temporal and spatial information from longitudinal VF data to predict time-to-glaucoma. When evaluated on the Ocular Hypertension Treatment Study (OHTS) dataset, our proposed convolutional neural network (CNN)-long short-term memory (LSTM) emerged as the top-performing model among all those examined. The implementation code can be found online (https://github.com/rivenzhou/VF_prediction).

PMID:39220673 | PMC:PMC11364364 | DOI:10.1002/sta4.649

Proc SPIE Int Soc Opt Eng. 2024 Feb;12930:129300K. doi: 10.1117/12.3009084. Epub 2024 Apr 2.

ABSTRACT

Whole brain segmentation with magnetic resonance imaging (MRI) enables the non-invasive measurement of brain regions, including total intracranial volume (TICV) and posterior fossa volume (PFV). Enhancing the existing whole brain segmentation methodology to incorporate intracranial measurements offers a heightened level of comprehensiveness in the analysis of brain structures. Despite its potential, the task of generalizing deep learning techniques for intracranial measurements faces data availability constraints due to limited manually annotated atlases encompassing whole brain and TICV/PFV labels. In this paper, we enhancing the hierarchical transformer UNesT for whole brain segmentation to achieve segmenting whole brain with 133 classes and TICV/PFV simultaneously. To address the problem of data scarcity, the model is first pretrained on 4859 T1-weighted (T1w) 3D volumes sourced from 8 different sites. These volumes are processed through a multi-atlas segmentation pipeline for label generation, while TICV/PFV labels are unavailable. Subsequently, the model is finetuned with 45 T1w 3D volumes from Open Access Series Imaging Studies (OASIS) where both 133 whole brain classes and TICV/PFV labels are available. We evaluate our method with Dice similarity coefficients(DSC). We show that our model is able to conduct precise TICV/PFV estimation while maintaining the 132 brain regions performance at a comparable level. Code and trained model are available at: https://github.com/MASILab/UNesT/wholebrainSeg.

PMID:39220623 | PMC:PMC11364374 | DOI:10.1117/12.3009084

Psychiatry Res Neuroimaging. 2024 Aug 28;344:111886. doi: 10.1016/j.pscychresns.2024.111886. Online ahead of print.

ABSTRACT

Verifying schizophrenia (SZ) can be assisted by deep learning techniques and patterns in brain activity observed in alpha-EEG recordings. The suggested research provides evidence of the reliability of alpha-EEG rhythm in a Gated-Recurrent-Unit-based deep-learning model for investigating SZ. This study suggests Rudiment Densely-Coupled Convolutional Gated Recurrent Unit (RDCGRU) for the various EEG-rhythm-based (gamma, beta, alpha, theta, and delta) diagnoses of SZ. The model includes multiple 1-D-Convolution (Con-1-D) folds with steps greater than 1, which enables the model to programmatically and effectively learn how to reduce the incoming signal. The Con-1-D layers and numerous Gated Recurrent Unit (GRU) layers comprise the Exponential-Linear-Unit activation function. This powerful activation function facilitates in-deep-network training and improves classification performance. The Densely-Coupled Convolutional Gated Recurrent Unit (DCGRU) layers enable RDCGRU to address the training accuracy loss brought on by vanishing or exploding gradients, and this might make it possible to develop intense, deep versions of RDCGRU for more complex problems. The sigmoid activation function is implemented in the digital (binary) classifier's output nodes. The RDCGRU deep learning model attained the most excellent accuracy, 88.88 %, with alpha-EEG rhythm. The research achievements: The RDCGRU deep learning model's GRU cells responded superiorly to the alpha-EEG rhythm in EEG-based verification of SZ.

PMID:39217668 | DOI:10.1016/j.pscychresns.2024.111886

Water Res. 2024 Aug 23;266:122315. doi: 10.1016/j.watres.2024.122315. Online ahead of print.

ABSTRACT

Accurately predicting the concentration of organochlorine pesticides (OCPs) presents a challenge due to their complex sources and environmental behaviors. In this study, we introduced a novel and advanced model that combined the power of three distinct techniques: Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Variational Mode Decomposition (VMD), and a deep learning network of Long Short-Term Memory (LSTM). The objective is to characterize the variation in OCPs concentrations with high precision. Results show that the hybrid two-stage decomposition coupled models achieved an average symmetric mean absolute percentage error (SMAPE) of 23.24 % in the empirical analysis of typical surface water. It exhibited higher predictive power than the given individual benchmark models, which yielded an average SMAPE of 40.88 %, and single decomposition coupled models with an average SMAPE of 29.80 %. The proposed CEEMDAN-VMD-LSTM model, with an average SMAPE of 13.55 %, consistently outperformed the other models, yielding an average SMAPE of 33.53 %. A comparative analysis with shallow neural network methods demonstrated the advantages of the LSTM algorithm when coupled with secondary decomposition techniques for processing time series datasets. Furthermore, the interpretable analysis derived by the SHAP approach revealed that precipitation followed by the total phosphorus had strong effects on the predicted concentration of OCPs in the given water. The data presented herein shows the effectiveness of decomposition technique-based deep learning algorithms in capturing the dynamic characteristics of pollutants in surface water.

PMID:39217646 | DOI:10.1016/j.watres.2024.122315

Eur Heart J. 2024 Sep 1:ehae595. doi: 10.1093/eurheartj/ehae595. Online ahead of print.

ABSTRACT

BACKGROUND AND AIMS: Deep learning applied to electrocardiograms (ECG-AI) is an emerging approach for predicting atrial fibrillation or flutter (AF). This study introduces an ECG-AI model developed and tested at a tertiary cardiac centre, comparing its performance with clinical and AF polygenic scores (PGS).

METHODS: ECG in sinus rhythm from the Montreal Heart Institute were analysed, excluding those from patients with preexisting AF. The primary outcome was incident AF at 5 years. An ECG-AI model was developed by splitting patients into non-overlapping datasets: 70% for training, 10% for validation, and 20% for testing. Performance of ECG-AI, clinical models and PGS was assessed in the test dataset. The ECG-AI model was externally validated in the Medical Information Mart for Intensive Care-IV (MIMIC-IV) hospital dataset.

RESULTS: A total of 669,782 ECGs from 145,323 patients were included. Mean age was 61±15 years, and 58% were male. The primary outcome was observed in 15% of patients and the ECG-AI model showed an area under the receiver operating characteristic curve (AUC) of 0.78. In time-to-event analysis including the first ECG, ECG-AI inference of high risk identified 26% of the population with a 4.3-fold increased risk of incident AF (95% confidence interval 4.02-4.57). In a subgroup analysis of 2,301 patients, ECG-AI outperformed CHARGE-AF (AUC=0.62) and PGS (AUC=0.59). Adding PGS and CHARGE-AF to ECG-AI improved goodness-of-fit (likelihood ratio test p<0.001), with minimal changes to the AUC (0.76-0.77). In the external validation cohort (mean age 59±18 years, 47% male, median follow-up 1.1 year) ECG-AI model performance= remained consistent (AUC=0.77).

CONCLUSIONS: ECG-AI provides an accurate tool to predict new-onset AF in a tertiary cardiac centre, surpassing clinical and polygenic scores.

PMID:39217446 | DOI:10.1093/eurheartj/ehae595

Comput Med Imaging Graph. 2024 Aug 28;117:102425. doi: 10.1016/j.compmedimag.2024.102425. Online ahead of print.

ABSTRACT

Photoacoustic tomography (PAT) is a powerful imaging modality for visualizing tissue physiology and exogenous contrast agents. However, PAT faces challenges in visualizing deep-seated vascular structures due to light scattering, absorption, and reduced signal intensity with depth. Optical coherence tomography angiography (OCTA) offers high-contrast visualization of vasculature networks, yet its imaging depth is limited to a millimeter scale. Herein, we propose OCPA-Net, a novel unsupervised deep learning method that utilizes the rich vascular feature of OCTA to enhance PAT images. Trained on unpaired OCTA and PAT images, OCPA-Net incorporates a vessel-aware attention module to enhance deep-seated vessel details captured from OCTA. It leverages a domain-adversarial loss function to enforce structural consistency and a novel identity invariant loss to mitigate excessive image content generation. We validate the structural fidelity of OCPA-Net on simulation experiments, and then demonstrate its vascular enhancement performance on in vivo imaging experiments of tumor-bearing mice and contrast-enhanced pregnant mice. The results show the promise of our method for comprehensive vessel-related image analysis in preclinical research applications.

PMID:39216343 | DOI:10.1016/j.compmedimag.2024.102425

Med Biol Eng Comput. 2024 Aug 31. doi: 10.1007/s11517-024-03188-8. Online ahead of print.

ABSTRACT

Deep learning has been widely used in ultrasound image analysis, and it also benefits kidney ultrasound interpretation and diagnosis. However, the importance of ultrasound image resolution often goes overlooked within deep learning methodologies. In this study, we integrate the ultrasound image resolution into a convolutional neural network and explore the effect of the resolution on diagnosis of kidney tumors. In the process of integrating the image resolution information, we propose two different approaches to narrow the semantic gap between the features extracted by the neural network and the resolution features. In the first approach, the resolution is directly concatenated with the features extracted by the neural network. In the second approach, the features extracted by the neural network are first dimensionally reduced and then combined with the resolution features to form new composite features. We compare these two approaches incorporating the resolution with the method without incorporating the resolution on a kidney tumor dataset of 926 images consisting of 211 images of benign kidney tumors and 715 images of malignant kidney tumors. The area under the receiver operating characteristic curve (AUC) of the method without incorporating the resolution is 0.8665, and the AUCs of the two approaches incorporating the resolution are 0.8926 (P < 0.0001) and 0.9135 (P < 0.0001) respectively. This study has established end-to-end kidney tumor classification systems and has demonstrated the benefits of integrating image resolution, showing that incorporating image resolution into neural networks can more accurately distinguish between malignant and benign kidney tumors in ultrasound images.

PMID:39215783 | DOI:10.1007/s11517-024-03188-8

Biomed Eng Online. 2024 Aug 31;23(1):90. doi: 10.1186/s12938-024-01280-6.

ABSTRACT

Medical imaging datasets for research are frequently collected from multiple imaging centers using different scanners, protocols, and settings. These variations affect data consistency and compatibility across different sources. Image harmonization is a critical step to mitigate the effects of factors like inherent differences between various vendors, hardware upgrades, protocol changes, and scanner calibration drift, as well as to ensure consistent data for medical image processing techniques. Given the critical importance and widespread relevance of this issue, a vast array of image harmonization methodologies have emerged, with deep learning-based approaches driving substantial advancements in recent times. The goal of this review paper is to examine the latest deep learning techniques employed for image harmonization by analyzing cutting-edge architectural approaches in the field of medical image harmonization, evaluating both their strengths and limitations. This paper begins by providing a comprehensive fundamental overview of image harmonization strategies, covering three critical aspects: established imaging datasets, commonly used evaluation metrics, and characteristics of different scanners. Subsequently, this paper analyzes recent structural MRI (Magnetic Resonance Imaging) harmonization techniques based on network architecture, network learning algorithm, network supervision strategy, and network output. The underlying architectures include U-Net, Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), flow-based generative models, transformer-based approaches, as well as custom-designed network architectures. This paper investigates the effectiveness of Disentangled Representation Learning (DRL) as a pivotal learning algorithm in harmonization. Lastly, the review highlights the primary limitations in harmonization techniques, specifically the lack of comprehensive quantitative comparisons across different methods. The overall aim of this review is to serve as a guide for researchers and practitioners to select appropriate architectures based on their specific conditions and requirements. It also aims to foster discussions around ongoing challenges in the field and shed light on promising future research directions with the potential for significant advancements.

PMID:39217355 | DOI:10.1186/s12938-024-01280-6

Commun Biol. 2024 Aug 31;7(1):1071. doi: 10.1038/s42003-024-06758-6.

ABSTRACT

The assessment of mutagenicity is essential in drug discovery, as it may lead to cancer and germ cells damage. Although in silico methods have been proposed for mutagenicity prediction, their performance is hindered by the scarcity of labeled molecules. However, experimental mutagenicity testing can be time-consuming and costly. One solution to reduce the annotation cost is active learning, where the algorithm actively selects the most valuable molecules from a vast chemical space and presents them to the oracle (e.g., a human expert) for annotation, thereby rapidly improving the model's predictive performance with a smaller annotation cost. In this paper, we propose muTOX-AL, a deep active learning framework, which can actively explore the chemical space and identify the most valuable molecules, resulting in competitive performance with a small number of labeled samples. The experimental results show that, compared to the random sampling strategy, muTOX-AL can reduce the number of training molecules by about 57%. Additionally, muTOX-AL exhibits outstanding molecular structural discriminability, allowing it to pick molecules with high structural similarity but opposite properties.

PMID:39217273 | DOI:10.1038/s42003-024-06758-6

Sci Rep. 2024 Aug 31;14(1):20270. doi: 10.1038/s41598-024-70774-z.

ABSTRACT

Dysphagia, a disorder affecting the ability to swallow, has a high prevalence among the older adults and can lead to serious health complications. Therefore, early detection of dysphagia is important. This study evaluated the effectiveness of a newly developed deep learning model that analyzes syllable-segmented data for diagnosing dysphagia, an aspect not addressed in prior studies. The audio data of daily conversations were collected from 16 patients with dysphagia and 24 controls. The presence of dysphagia was determined by videofluoroscopic swallowing study. The data were segmented into syllables using a speech-to-text model and analyzed with a convolutional neural network to perform binary classification between the dysphagia patients and control group. The proposed model in this study was assessed in two different aspects. Firstly, with syllable-segmented analysis, it demonstrated a diagnostic accuracy of 0.794 for dysphagia, a sensitivity of 0.901, a specificity of 0.687, a positive predictive value of 0.742, and a negative predictive value of 0.874. Secondly, at the individual level, it achieved an overall accuracy of 0.900 and area under the curve of 0.953. This research highlights the potential of deep learning modal as an early, non-invasive, and simple method for detecting dysphagia in everyday environments.

PMID:39217249 | DOI:10.1038/s41598-024-70774-z

Pediatr Cardiol. 2024 Aug 31. doi: 10.1007/s00246-024-03630-6. Online ahead of print.

ABSTRACT

The volumetric data obtained from the cardiac CT scan of congenital heart disease patients is important for defining patient's status and making decision for proper management. The objective of this study is to evaluate the intra-observer, inter-observer, and interstudy reproducibility of left ventricular (LV) and right ventricular (RV) or functional single-ventricle (FSV) volume. And compared those between manual and using semi-automated segmentation tool. Total of 127 patients (56 female, 71 male; mean age 82.1 months) underwent pediatric protocol cardiac CT from January 2020 to December 2022. The volumetric data including both end-systolic and -diastolic volume and calculated EF were derived from both conventional semiautomatic region growing algorithms (CM, TeraRecon, TeraRecon, Inc., San Mateo, CA, USA) and deep learning-based annotation program (DLS, Medilabel, Ingradient, Inc., Seoul, Republic of Korea) by three readers, who have different background knowledge or experience of radiology or image extraction before. The reproducibility was compared using intra- and inter-observer agreements. And the usability was measured using time for reconstruction and number of tests that were reconfigured before the reconfiguration time was reduced to less than 5 min. Inter- and intra-observer agreements showed better agreements degrees in DLS than CM in all analyzers. The time used for reconstruction showed significantly shorter in DLS compared with CM. And significantly small numbers of tests before the reconfiguration is needed in DLS than CM. Deep learning-based annotation program can be more accurate way for measurement of volumetric data for congenital heart disease patients with better reproducibility than conventional method.

PMID:39217235 | DOI:10.1007/s00246-024-03630-6