Sci Rep. 2024 Dec 2;14(1):29877. doi: 10.1038/s41598-024-80691-w.

ABSTRACT

Diabetes foot ulcer (DFU) is one of the common complications of diabetes patients, which may lead to infection, necrosis and even amputation. Therefore, early diagnosis, classification of severity and related treatment are crucial for the patients. Current DFU classification methods often require experienced doctors to manually classify the severity, which is time-consuming and low accuracy. The objective of the study is to propose a few-shot DFU image classification method based on deep residual neural network and transfer learning. Considering the difficulty in obtaining clinical DFU images, it is a few-shot problem. Therefore, the methods include: (1) Data augmentation of the original DFU images by using geometric transformations and random noise; (2) Deep ResNet models selection based on different convolutional layers comparative experiments; (3) DFU classification model training with transfer learning by using the selected pre-trained ResNet model and fine tuning. To verify the proposed classification method, the experiments were performed with the original and augmented datasets by separating three classifications: zero grade, mild grade, severe grade. (1) The datasets were augmented from the original 146 to 3000 image datasets and the corresponding DFU classification's average accuracy from 0.9167 to 0.9867; (2) Comparative experiments were conducted with ResNet18, ResNet34, ResNet50, ResNet101, ResNet152 by using 3000 image datasets, and the average accuracy/loss is 0.9325/0.2927, 0.9276/0.3234, 0.9901/0.1356, 0.9865/0.1427, 0.9790/0.1583 respectively; (3) Based on the augmented 3000 image datasets, it was achieved 0.9867 average accuracy with the DFU classification model, which was trained by the pre-trained ResNet50 and hyper-parameters. The experimental results indicated that the proposed few-shot DFU image classification method based on deep ResNet and transfer learning got very high accuracy, and it is expected to be suitable for low-cost and low-computational terminal equipment, aiming at helping clinical DFU classification efficiently and auxiliary diagnosis.

PMID:39622873 | DOI:10.1038/s41598-024-80691-w

JMIR Med Inform. 2024 Nov 21;12:e60334. doi: 10.2196/60334.

ABSTRACT

BACKGROUND: Clinical named entity recognition (CNER) is a fundamental task in natural language processing used to extract named entities from electronic medical record texts. In recent years, with the continuous development of machine learning, deep learning models have replaced traditional machine learning and template-based methods, becoming widely applied in the CNER field. However, due to the complexity of clinical texts, the diversity and large quantity of named entity types, and the unclear boundaries between different entities, existing advanced methods rely to some extent on annotated databases and the scale of embedded dictionaries.

OBJECTIVE: This study aims to address the issues of data scarcity and labeling difficulties in CNER tasks by proposing a dataset augmentation algorithm based on proximity word calculation.

METHODS: We propose a Segmentation Synonym Sentence Synthesis (SSSS) algorithm based on neighboring vocabulary, which leverages existing public knowledge without the need for manual expansion of specialized domain dictionaries. Through lexical segmentation, the algorithm replaces new synonymous vocabulary by recombining from vast natural language data, achieving nearby expansion expressions of the dataset. We applied the SSSS algorithm to the Robustly Optimized Bidirectional Encoder Representations from Transformers Pretraining Approach (RoBERTa) + conditional random field (CRF) and RoBERTa + Bidirectional Long Short-Term Memory (BiLSTM) + CRF models and evaluated our models (SSSS + RoBERTa + CRF; SSSS + RoBERTa + BiLSTM + CRF) on the China Conference on Knowledge Graph and Semantic Computing (CCKS) 2017 and 2019 datasets.

RESULTS: Our experiments demonstrated that the models SSSS + RoBERTa + CRF and SSSS + RoBERTa + BiLSTM + CRF achieved F1-scores of 91.30% and 91.35% on the CCKS-2017 dataset, respectively. They also achieved F1-scores of 83.21% and 83.01% on the CCKS-2019 dataset, respectively.

CONCLUSIONS: The experimental results indicated that our proposed method successfully expanded the dataset and remarkably improved the performance of the model, effectively addressing the challenges of data acquisition, annotation difficulties, and insufficient model generalization performance.

PMID:39622697 | DOI:10.2196/60334

Biomed Phys Eng Express. 2024 Dec 2. doi: 10.1088/2057-1976/ad992c. Online ahead of print.

ABSTRACT

The prediction of epileptic seizures is a classical research problem, representing one of the most challenging tasks in the analysis of brain disorders. There is active research into digital twins (DT) for various healthcare applications, as they can transform research into customized and personalized healthcare. The widespread adoption of DT technology relies on ample patient data to ensure precise monitoring and decision-making, leveraging Machine Learning (ML) and Deep Learning (DL) algorithms. Given the non-stationarity of EEG recordings, characterized by substantial frequency variations over time, there is a notable preference for advanced time-frequency methods in seizure prediction. This research proposes a DT-based seizure prediction system by applying an advanced time-frequency analysis approach known as Time-Reassigned MultiSynchroSqueezing Transform (TMSST) to EEG data to extract patient-specific impulse features and subsequently, a Deep Learning strategy, CNN-BiLSTM-Attention mechanism model is utilized in learning and classifying features for seizure prediction. The proposed architecture is named as "Digital Twin-Net". By estimating the group delay in the time direction, TMSST produces the frequency components that are responsible for the EEG signal's temporal behavior and those time-frequency signatures are learned by the developed CNN-BiLSTM-Attention mechanism model. Thus the combination acts as a digital twin of a patient for the prediction of epileptic seizures. The experimental results showed that the suggested approach achieved an accuracy of 99.70% when tested on 23 patients from the publicly accessible CHB-MIT dataset. The proposed method surpasses previous solutions in terms of overall performance. Consequently, the suggested method can be regarded as an efficient approach to EEG seizure prediction.&#xD.

PMID:39622083 | DOI:10.1088/2057-1976/ad992c

Biomed Phys Eng Express. 2024 Dec 2. doi: 10.1088/2057-1976/ad992d. Online ahead of print.

ABSTRACT

Smartphone-assisted urine analyzers estimate the urinary albumin by quantifying color changes at sensor pad of test strips. These strips yield color variations due to the total protein present in the sample, making it difficult to relate to color changes due to specific analyte. We have addressed it using a Lateral Flow Assay (LFA) device for automatic detection and quantification of urinary albumin. LFAs are specific to individual analytes, allowing color changes to be linked to the specific analyte, minimizing the interference. The proposed reader performs automatic segmentation of the region of interest (ROI) using YOLOv5, a deep learning-based model. Concentrations of urinary albumin in clinical samples were classified using customized machine learning algorithms. An accuracy of 96% was achieved on the test data using the k-Nearest Neighbour (k-NN) algorithm. Performance of the model was also evaluated under different illumination conditions and with different smartphone cameras, and validated using standard nephelometer.

PMID:39622082 | DOI:10.1088/2057-1976/ad992d

Chaos. 2024 Dec 1;34(12):120401. doi: 10.1063/5.0247498.

ABSTRACT

Regime switching, the process where complex systems undergo transitions between qualitatively different dynamical states due to changes in their conditions, is a widespread phenomenon, from climate and ocean circulation, to ecosystems, power grids, and the brain. Capturing the mechanisms that give rise to isolated or sequential switching dynamics, as well as developing generic and robust methods for forecasting, detecting, and controlling them is essential for maintaining optimal performance and preventing dysfunctions or even collapses in complex systems. This Focus Issue provides new insights into regime switching, covering the recent advances in theoretical analysis harnessing the reduction approaches, as well as data-driven detection methods and non-feedback control strategies. Some of the key challenges addressed include the development of reduction techniques for coupled stochastic and adaptive systems, the influence of multiple timescale dynamics on chaotic structures and cyclic patterns in forced systems, and the role of chaotic saddles and heteroclinic cycles in pattern switching in coupled oscillators. The contributions further highlight deep learning applications for predicting power grid failures, the use of blinking networks to enhance synchronization, creating adaptive strategies to control epidemic spreading, and non-feedback control strategies to suppress epileptic seizures. These developments are intended to catalyze further dialog between the different branches of complexity.

PMID:39621472 | DOI:10.1063/5.0247498

J Imaging Inform Med. 2024 Dec 2. doi: 10.1007/s10278-024-01337-x. Online ahead of print.

ABSTRACT

Small cohorts of certain disease states are common especially in medical imaging. Despite the growing culture of data sharing, information safety often precludes open sharing of these datasets for creating generalizable machine learning models. To overcome this barrier and maintain proper health information protection, foundational models are rapidly evolving to provide deep learning solutions that have been pretrained on the native feature spaces of the data. Although this has been optimized in Large Language Models (LLMs), there is still a sparsity of foundational models for computer vision tasks. It is in this space that we provide an investigation into pretraining Visual Geometry Group (VGG)-16, Residual Network (ResNet)-50, and Dense Network (DenseNet)-121 on an unrelated dataset of 8500 chest CTs which was subsequently fine-tuned to classify bone mineral density (BMD) in 199 breast cancer patients using the L1 vertebra on CT. These semi-foundational models showed significant improved ternary classification into mild, moderate, and severe demineralization in comparison to ground truth Hounsfield Unit (HU) measurements in trabecular bone with the semi-foundational ResNet50 architecture demonstrating the best relative performance. Specifically, the holdout testing AUC was 0.99 (p-value < 0.05, ANOVA versus no pretraining versus ImageNet transfer learning) and F1-score 0.99 (p-value < 0.05) for the holdout testing set. In this study, the use of a semi-foundational model trained on the native feature space of CT provided improved classification in a completely disparate disease state with different window levels. Future implementation with these models may provide better generalization despite smaller numbers of a disease state to be classified.

PMID:39621209 | DOI:10.1007/s10278-024-01337-x

Clin Oral Investig. 2024 Dec 2;28(12):677. doi: 10.1007/s00784-024-06068-5.

ABSTRACT

OBJECTIVES: The study aims to address the global burden of dental caries, a highly prevalent disease affecting billions of individuals, including both children and adults. Recognizing the significant health challenges posed by untreated dental caries, particularly in low- and middle-income countries, our goal is to improve early-stage detection. Though effective, traditional diagnostic methods, such as bitewing radiography, have limitations in detecting early lesions. By leveraging Artificial Intelligence (AI), we aim to enhance the accuracy and efficiency of caries detection, offering a transformative approach to dental diagnostics.

MATERIALS AND METHODS: This study proposes a novel deep learning-based approach for dental caries detection using the latest models, i.e., YOLOv7, YOLOv8, and YOLOv9. Trained on a dataset of over 3,200 images, the models address the shortcomings of existing detection methods and provide an automated solution to improve diagnostic accuracy.

RESULTS: The YOLOv7 model achieved a mean Average Precision (mAP) at 0.5 Intersection over Union (IoU) of 0.721, while YOLOv9 attained a mAP@50 IoU of 0.832. Notably, YOLOv8 outperformed both, with a mAP@0.5 of 0.982. This demonstrates robust detection capabilities across multiple categories, including caries," "Deep Caries," and "Exclusion."

CONCLUSIONS: This high level of accuracy and efficiency highlights the potential of integrating AI-driven systems into clinical workflows, improving diagnostic capabilities, reducing healthcare costs, and contributing to better patient outcomes, especially in resource-constrained environments.

CLINICAL RELEVANCE: Integrating these latest YOLO advanced AI models into dental diagnostics could transform the landscape of caries detection. Enhancing early-stage diagnosis accuracy can lead to more precise and cost-effective treatment strategies, with significant implications for improving patient outcomes, particularly in low-resource settings where traditional diagnostic capabilities are often limited.

PMID:39621193 | DOI:10.1007/s00784-024-06068-5

Insights Imaging. 2024 Dec 2;15(1):290. doi: 10.1186/s13244-024-01861-y.

ABSTRACT

OBJECTIVE: To develop a multi-scene model that can automatically segment acute vertebral compression fractures (VCFs) from spine radiographs.

METHODS: In this multicenter study, we collected radiographs from five hospitals (Hospitals A-E) between November 2016 and October 2019. The study included participants with acute VCFs, as well as healthy controls. For the development of the Positioning and Focus Network (PFNet), we used a training dataset consisting of 1071 participants from Hospitals A and B. The validation dataset included 458 participants from Hospitals A and B, whereas external test datasets 1-3 included 301 participants from Hospital C, 223 from Hospital D, and 261 from Hospital E, respectively. We evaluated the segmentation performance of the PFNet model and compared it with previously described approaches. Additionally, we used qualitative comparison and gradient-weighted class activation mapping (Grad-CAM) to explain the feature learning and segmentation results of the PFNet model.

RESULTS: The PFNet model achieved accuracies of 99.93%, 98.53%, 99.21%, and 100% for the segmentation of acute VCFs in the validation dataset and external test datasets 1-3, respectively. The receiver operating characteristic curves comparing the four models across the validation and external test datasets consistently showed that the PFNet model outperformed other approaches, achieving the highest values for all measures. The qualitative comparison and Grad-CAM provided an intuitive view of the interpretability and effectiveness of our PFNet model.

CONCLUSION: In this study, we successfully developed a multi-scene model based on spine radiographs for precise preoperative and intraoperative segmentation of acute VCFs.

CRITICAL RELEVANCE STATEMENT: Our PFNet model demonstrated high accuracy in multi-scene segmentation in clinical settings, making it a significant advancement in this field.

KEY POINTS: This study developed the first multi-scene deep learning model capable of segmenting acute VCFs from spine radiographs. The model's architecture consists of two crucial modules: an attention-guided module and a supervised decoding module. The exceptional generalization and consistently superior performance of our model were validated using multicenter external test datasets.

PMID:39621135 | DOI:10.1186/s13244-024-01861-y

Bioprocess Biosyst Eng. 2024 Dec 2. doi: 10.1007/s00449-024-03110-4. Online ahead of print.

ABSTRACT

Fast and accurate detection of infectious bacteria in wounds is crucial for effective clinical treatment. However, traditional methods take over 24 h to yield results, which is inadequate for urgent clinical needs. Here, we introduce a deep learning-driven framework that detects and classifies four bacteria commonly found in wounds: Acinetobacter baumannii (AB), Escherichia coli (EC), Pseudomonas aeruginosa (PA), and Staphylococcus aureus (SA). This framework leverages the pretrained ResNet50 deep learning architecture, trained on manually collected periodic bacterial colony-growth images from high-content imaging. In in vitro samples, our method achieves a detection rate of over 95% for early colonies cultured for 8 h, reducing detection time by more than 12 h compared to traditional Environmental Protection Agency (EPA)-approved methods. For colony classification, it identifies AB, EC, PA, and SA colonies with accuracies of 96%, 97%, 96%, and 98%, respectively. For mixed bacterial samples, it identifies colonies with 95% accuracy and classifies them with 93% precision. In mouse wound samples, the method identifies over 90% of developing bacterial colonies and classifies colony types with an average accuracy of over 94%. These results highlight the framework's potential for improving the clinical treatment of wound infections. Besides, the framework provides the detection results with key feature visualization, which enhance the prediction credibility for users. To summarize, the proposed framework enables high-throughput identification, significantly reducing detection time and providing a cost-effective tool for early bacterial detection.

PMID:39621107 | DOI:10.1007/s00449-024-03110-4

Abdom Radiol (NY). 2024 Dec 2. doi: 10.1007/s00261-024-04708-8. Online ahead of print.

ABSTRACT

Large language models (LLMs) and multi-modal large language models (MLLMs) represent the cutting-edge in artificial intelligence. This review provides a comprehensive overview of their capabilities and potential impact on radiology. Unlike most existing literature reviews focusing solely on LLMs, this work examines both LLMs and MLLMs, highlighting their potential to support radiology workflows such as report generation, image interpretation, EHR summarization, differential diagnosis generation, and patient education. By streamlining these tasks, LLMs and MLLMs could reduce radiologist workload, improve diagnostic accuracy, support interdisciplinary collaboration, and ultimately enhance patient care. We also discuss key limitations, such as the limited capacity of current MLLMs to interpret 3D medical images and to integrate information from both image and text data, as well as the lack of effective evaluation methods. Ongoing efforts to address these challenges are introduced.

PMID:39621074 | DOI:10.1007/s00261-024-04708-8

J Med Chem. 2024 Dec 2. doi: 10.1021/acs.jmedchem.4c02193. Online ahead of print.

ABSTRACT

Molecular property prediction with deep learning often employs self-supervised learning techniques to learn common knowledge through masked atom prediction. However, the common knowledge gained by masked atom prediction dramatically differs from the graph-level optimization objective of downstream tasks, which results in suboptimal problems. Particularly for properties with limited data, the failure to consider domain knowledge results in a direct search in an immense common space, rendering it infeasible to identify the global optimum. To address this, we propose MPCD, which enhances pretraining transferability by aligning the optimization objectives between pretraining and fine-tuning with domain knowledge. MPCD also leverages multitask learning to improve data utilization and model robustness. Technically, MPCD employs a relation-aware self-attention mechanism to capture molecules' local and global structures comprehensively. Extensive validation demonstrates that MPCD outperforms state-of-the-art methods for absorption, distribution, metabolism, excretion, and toxicity (ADMET) and physicochemical prediction across various data sizes.

PMID:39620982 | DOI:10.1021/acs.jmedchem.4c02193

Med Phys. 2024 Dec 2. doi: 10.1002/mp.17503. Online ahead of print.

ABSTRACT

BACKGROUND: Accurate musculoseletal ultrasound (MSKUS) image segmentation is crucial for diagnosis and treatment planning. Compared with traditional segmentation methods, deploying deep learning segmentation methods that balance segmentation efficiency, accuracy, and model size on edge devices has greater advantages.

PURPOSE: This paper aims to design a MSKUS image segmentation method that has fewer parameters, lower computation complexity and higher segmentation accuracy.

METHODS: In this study, an attention mechanism-based lightweight UNet (AML-UNet) is designed to segment target muscle regions in MSKUS images. To suppress the transmission of redundant feature, Channel Reconstruction and Spatial Attention Module is designed in the encoding path. In addition, considering the inherent characteristic of MSKUS image, Multiscale Aggregation Module is developed to replace the skip connection architecture of U-Net. Deep supervision is also introduced to the decoding path to refine predicted masks gradually. Our method is evaluated on two MSKUS 2D-image segmentation datasets, including 3917 MSKUS and 1534 images respectively. In the experiments, a five-fold cross-validation method is adopted in ablation experiments and comparison experiments. In addition, Wilcoxon Signed-Rank Test and Bonferroni correction are employed to validate the significance level. 0.01 was used as the statistical significance level in our paper.

RESULTS: AML-UNet yielded a mIoU of 84.17% and 90.14% on two datasets, representing a 3.38% ( p < 0.001 $p&lt;0.001$ ) and 3.48% ( p < 0.001 $p&lt;0.001$ ) over the Unext model. The number of parameters and FLOPs are only 0.21M and 0.96G, which are 1/34 and 1/29 of those in comparison with UNet.

CONCLUSIONS: Our proposed model achieved superior results with fewer parameters while maintaining segmentation efficiency and accuracy compared to other methods.

PMID:39620487 | DOI:10.1002/mp.17503

Microsc Res Tech. 2024 Dec 2. doi: 10.1002/jemt.24739. Online ahead of print.

ABSTRACT

Diabetes is a chronic disease that occurs when the body cannot regulate blood sugar levels. Nowadays, the screening tests for diabetes are developed using multivariate regression methods. An increasing amount of data is automatically collected to provide an opportunity for creating challenging and accurate prediction modes that are updated constantly with the help of machine learning techniques. In this manuscript, a Dual Multi Scale Attention Network optimized with Archerfish Hunting Optimization Algorithm is proposed for Diabetes Prediction (DMSAN-AHO-DP). Here, the data is gathered through PIMA Indian Diabetes Dataset (PIDD). The collected data is fed towards the preprocessing to remove the noise of input data and improves the data quality by using Contrast Limited Adaptive Histogram Equalization Filtering (CLAHEF) method. Then the preprocessed data are fed to Multi-Level Haar Wavelet Features Fusion Network (MHWFFN) based feature extraction. Then the extracted data is supplied to the Dual Multi Scale Attention Network (DMSAN) for diabetic or non-diabetic classification. The hyper parameter of Dual Multi Scale Attention Network is tuned with Archerfish Hunting Optimization (AHO) algorithm, which classifies diabetic or non-diabetic accurately. The proposed DMSAN-AHO-DP technique is implemented in Python. The efficacy of the DMSAN-AHO-DP approach is examined with some metrics, like Accuracy, F-scores, Sensitivity, Specificity, Precision, Recall, Computational time. The DMSAN-AHO-DP technique achieves 23.52%, 36.12%, 31.12% higher accuracy and 16.05%, 21.14%, 31.02% lesser error rate compared with existing models: Enhanced Deep Neural Network based Model for Diabetes Prediction (EDNN-DP), Indian PIMA Dataset using Deep Learning for Diabetes Prediction (ANN-DP), and Enhanced Support Vector Machine with Deep Neural Network Learning strategies for Diabetes Prediction (SVM-DNN-DP).

PMID:39620437 | DOI:10.1002/jemt.24739

Curr Diabetes Rev. 2024 Nov 29. doi: 10.2174/0115733998307556240819093038. Online ahead of print.

ABSTRACT

BACKGROUND: Recent research demonstrates that diabetes can lead to heart problems, neurological damage, and other illnesses.

METHOD: In this paper, we design a low-complexity Deep Learning (DL)-based model for the diagnosis of type 2 diabetes. In our experiments, we use the publicly available PIMA Indian Diabetes Dataset (PIDD). To obtain a low-complexity and accurate DL architecture, we perform an accuracy-versus-complexity study on several DL models.

RESULT: The results show that the proposed DL structure, including Convolutional Neural Networks and Multi-Layer Perceptron models (i.e., CNN+MLP model) outperforms other models with an accuracy of 93.89%.

CONCLUSION: With these features, the proposed hybrid model can be used in wearable devices and IoT-based health monitoring applications.

PMID:39620332 | DOI:10.2174/0115733998307556240819093038

J Orthop Res. 2024 Dec 2. doi: 10.1002/jor.26024. Online ahead of print.

ABSTRACT

Magnetic resonance imaging (MRI) of the knee is the recommended diagnostic method before invasive arthroscopy surgery. Nevertheless, interpreting knee MRI scans is a time-consuming process that is vulnerable to inaccuracies and inconsistencies. We proposed a multitask learning network MCSNetatt which efficiently introduces segmentation prior features and enhances classification results through multiscale feature fusion and spatial attention modules. The MRI studies and subsequent arthroscopic diagnosis of 259 knees were collected retrospectively. Models were trained based on multitask loss with coronal and sagittal sequences and fused using logistic regression (LR). We visualized the network's interpretability by the gradient-weighted class activation mapping method. The LR model achieved higher area under the curve and mean average precision of medial and lateral menisci than models trained on a single sagittal or coronal sequence. Our multitask model MCSNetat outperformed the single-task model CNet and two clinicians in classification, with accuracy, precision, recall, F1-score of 0.980, 1.000, 0.952, 0.976 for medial and 0.920, 0.905, 0.905, 0.905 for the lateral, respectively. With the assistance of model results and visualized saliency maps, both clinicians showed improvement in their diagnostic performance. Compared to the baseline segmentation model, our model improved dice similarity coefficient and the 95% Hausdorff distance (HD95) of the lateral meniscus for 2.3% and 0.860 mm in coronal images and 4.4% and 2.253 mm in sagittal images. Our multitask learning network quickly generated accurate clinicopathological classification and segmentation of knee MRI, demonstrating its potential to assist doctors in a clinical setting.

PMID:39620311 | DOI:10.1002/jor.26024

Front Radiol. 2024 Nov 15;4:1412404. doi: 10.3389/fradi.2024.1412404. eCollection 2024.

NO ABSTRACT

PMID:39620146 | PMC:PMC11608602 | DOI:10.3389/fradi.2024.1412404

Mar Life Sci Technol. 2024 Sep 27;6(4):631-650. doi: 10.1007/s42995-024-00252-y. eCollection 2024 Nov.

ABSTRACT

Genomic selection (GS) applied to the breeding of aquatic animals has been of great interest in recent years due to its higher accuracy and faster genetic progress than pedigree-based methods. The genetic analysis of complex traits in GS does not escape the current excitement around artificial intelligence, including a renewed interest in deep learning (DL), such as deep neural networks (DNNs), convolutional neural networks (CNNs), and autoencoders. This article reviews the current status and potential of DL applications in phenotyping, genotyping and genomic estimated breeding value (GEBV) prediction of GS. It can be seen from this article that CNNs obtain phenotype data of aquatic animals efficiently, and without injury; DNNs as single nucleotide polymorphism (SNP) variant callers are critical to have shown higher accuracy in assessments of genotyping for the next-generation sequencing (NGS); autoencoder-based genotype imputation approaches are capable of highly accurate genotype imputation by encoding complex genotype relationships in easily portable inference models; sparse DNNs capture nonlinear relationships among genes to improve the accuracy of GEBV prediction for aquatic animals. Furthermore, future directions of DL in aquaculture are also discussed, which should expand the application to more aquaculture species. We believe that DL will be applied increasingly to molecular breeding of aquatic animals in the future.

SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s42995-024-00252-y.

PMID:39620094 | PMC:PMC11602929 | DOI:10.1007/s42995-024-00252-y

Mar Life Sci Technol. 2024 Apr 29;6(4):736-749. doi: 10.1007/s42995-024-00222-4. eCollection 2024 Nov.

ABSTRACT

Improving the accuracy of fishing ground prediction for oceanic economic species has always been one of the most concerning issues in fisheries research. Recent studies have confirmed that deep learning has achieved superior results over traditional methods in the era of big data. However, the deep learning-based fishing ground prediction model with a single environment suffers from the problem that the area of the fishing ground is too large and not concentrated. In this study, we developed a deep learning-based fishing ground prediction model with multiple environmental factors using neon flying squid (Ommastrephes bartramii) in Northwest Pacific Ocean as an example. Based on the modified U-Net model, the approach involves the sea surface temperature, sea surface height, sea surface salinity, and chlorophyll a as inputs, and the center fishing ground as the output. The model is trained with data from July to November in 2002-2019, and tested with data of 2020. We considered and compared five temporal scales (3, 6, 10, 15, and 30 days) and seven multiple environmental factor combinations. By comparing different cases, we found that the optimal temporal scale is 30 days, and the optimal multiple environmental factor combination contained SST and Chl a. The inclusion of multiple factors in the model greatly improved the concentration of the center fishing ground. The selection of a suitable combination of multiple environmental factors is beneficial to the precise spatial distribution of fishing grounds. This study deepens the understanding of the mechanism of environmental field influence on fishing grounds from the perspective of artificial intelligence and fishery science.

PMID:39620085 | PMC:PMC11602920 | DOI:10.1007/s42995-024-00222-4

Eur J Radiol Open. 2024 Nov 15;13:100610. doi: 10.1016/j.ejro.2024.100610. eCollection 2024 Dec.

ABSTRACT

PURPOSE: To evaluate the effectiveness of a constructed deep learning model in predicting early recurrence after surgery in hepatocellular carcinoma (HCC) patients with solitary tumors ≤5 cm.

MATERIALS AND METHODS: Our study included a total of 331 HCC patients who underwent curative resection, with all patients having preoperative dynamic contrast-enhanced MRI (DCE-MRI). Patients who recurred within two years after surgery were defined as early recurrence. The enrolled patients were randomly divided into the training group and the testing group. A ResNet-based deep learning model with eight conventional neural network branches was built to predict the early recurrence status of these patients. Patient characteristics and laboratory tests were further filtered by regression models and then integrated with deep learning models to improve the prediction performance.

RESULTS: Among 331 HCC patients, 70 (21.1 %) experienced early recurrence. In multivariate Cox regression analysis, only tumor size (Hazard ratio (HR=1.394, 95 %CI:1.011-1.920, p value=0.043) and deep learning extracted image features (HR: 38440, 95 %CI:2321-636600, p value<0.001) were significant risk factors for early recurrence. In the training and testing cohort, the AUCs of the image-based deep learning prediction model were 0.839 and 0.833. By integrating tumor size with image-based deep learning model to construct a combined model, we found that the AUCs of the combined model to assess early recurrence in the training and validation cohort were 0.846 and 0.842. We further developed a nomogram to visualize the preoperative combined model, and the prediction performance of nomogram showed a good fitness in the testing cohort.

CONCLUSIONS: The proposed deep learning-based prediction model using DCE-MRI is useful for assessing early recurrence in HCC patients with single tumors ≤5 cm.

PMID:39619794 | PMC:PMC11607649 | DOI:10.1016/j.ejro.2024.100610

Proc AAAI Conf Artif Intell. 2024;38(14):15634-15643. doi: 10.1609/aaai.v38i14.29491. Epub 2024 Mar 24.

ABSTRACT

Utilizing electronic health records (EHR) for machine learning-driven clinical research has great potential to enhance outcome predictions and treatment personalization. Nonetheless, due to privacy and security concerns, the secondary use of EHR data is regulated, constraining researchers' access to EHR data. Generating synthetic EHR data with deep learning methods is a viable and promising approach to mitigate privacy concerns, offering not only a supplementary resource for downstream applications but also sidestepping the privacy risks associated with real patient data. While prior efforts have concentrated on EHR data synthesis, significant challenges persist: addressing the heterogeneity of features including temporal and non-temporal features, structurally missing values, and irregularity of the temporal measures, and ensuring rigorous privacy of the real data used for model training. Existing works in this domain only focused on solving one or two aforementioned challenges. In this work, we propose IGAMT, an innovative framework to generate privacy-preserved synthetic EHR data that not only maintains high quality with heterogeneous features, missing values, and irregular measures but also achieves differential privacy with enhanced privacy-utility trade-off. Extensive experiments prove that IGAMT significantly outperforms baseline and state-of-the-art models in terms of resemblance to real data and performance of downstream applications. Ablation studies also prove the effectiveness of the techniques applied in IGAMT.

PMID:39619768 | PMC:PMC11606572 | DOI:10.1609/aaai.v38i14.29491