J Magn Reson Imaging. 2024 Feb 16. doi: 10.1002/jmri.29310. Online ahead of print.

NO ABSTRACT

PMID:38363190 | DOI:10.1002/jmri.29310

Front Vet Sci. 2024 Feb 1;11:1357109. doi: 10.3389/fvets.2024.1357109. eCollection 2024.

ABSTRACT

There is a critical need to develop and validate non-invasive animal-based indicators of affective states in livestock species, in order to integrate them into on-farm assessment protocols, potentially via the use of precision livestock farming (PLF) tools. One such promising approach is the use of vocal indicators. The acoustic structure of vocalizations and their functions were extensively studied in important livestock species, such as pigs, horses, poultry, and goats, yet cattle remain understudied in this context to date. Cows were shown to produce two types of vocalizations: low-frequency calls (LF), produced with the mouth closed, or partially closed, for close distance contacts, and open mouth emitted high-frequency calls (HF), produced for long-distance communication, with the latter considered to be largely associated with negative affective states. Moreover, cattle vocalizations were shown to contain information on individuality across a wide range of contexts, both negative and positive. Nowadays, dairy cows are facing a series of negative challenges and stressors in a typical production cycle, making vocalizations during negative affective states of special interest for research. One contribution of this study is providing the largest to date pre-processed (clean from noises) dataset of lactating adult multiparous dairy cows during negative affective states induced by visual isolation challenges. Here, we present two computational frameworks-deep learning based and explainable machine learning based, to classify high and low-frequency cattle calls and individual cow voice recognition. Our models in these two frameworks reached 87.2 and 89.4% accuracy for LF and HF classification, with 68.9 and 72.5% accuracy rates for the cow individual identification, respectively.

PMID:38362300 | PMC:PMC10867142 | DOI:10.3389/fvets.2024.1357109

iScience. 2024 Jan 26;27(2):109002. doi: 10.1016/j.isci.2024.109002. eCollection 2024 Feb 16.

ABSTRACT

This study focuses on enhancing the prediction of regulatory functional sites in DNA and RNA sequences, a crucial aspect of gene regulation. Current methods, such as motif overrepresentation and machine learning, often lack specificity. To address this issue, the study leverages evolutionary information and introduces Graphylo, a deep-learning approach for predicting transcription factor binding sites in the human genome. Graphylo combines Convolutional Neural Networks for DNA sequences with Graph Convolutional Networks on phylogenetic trees, using information from placental mammals' genomes and evolutionary history. The research demonstrates that Graphylo consistently outperforms both single-species deep learning techniques and methods that incorporate inter-species conservation scores on a wide range of datasets. It achieves this by utilizing a species-based attention model for evolutionary insights and an integrated gradient approach for nucleotide-level model interpretability. This innovative approach offers a promising avenue for improving the accuracy of regulatory site prediction in genomics.

PMID:38362268 | PMC:PMC10867641 | DOI:10.1016/j.isci.2024.109002

Laryngoscope Investig Otolaryngol. 2024 Feb 8;9(1):e1199. doi: 10.1002/lio2.1199. eCollection 2024 Feb.

ABSTRACT

OBJECTIVES: In this study, we propose a diagnostic model for automatic detection of otitis media based on combined input of otoscopy images and wideband tympanometry measurements.

METHODS: We present a neural network-based model for the joint prediction of otitis media and diagnostic difficulty. We use the subclassifications acute otitis media and otitis media with effusion. The proposed approach is based on deep metric learning, and we compare this with the performance of a standard multi-task network.

RESULTS: The proposed deep metric approach shows good performance on both tasks, and we show that the multi-modal input increases the performance for both classification and difficulty estimation compared to the models trained on the modalities separately. An accuracy of 86.5% is achieved for the classification task, and a Kendall rank correlation coefficient of 0.45 is achieved for difficulty estimation, corresponding to a correct ranking of 72.6% of the cases.

CONCLUSION: This study demonstrates the strengths of a multi-modal diagnostic tool using both otoscopy images and wideband tympanometry measurements for the diagnosis of otitis media. Furthermore, we show that deep metric learning improves the performance of the models.

PMID:38362190 | PMC:PMC10866588 | DOI:10.1002/lio2.1199

Pol J Radiol. 2023 Dec 8;88:e562-e573. doi: 10.5114/pjr.2023.133660. eCollection 2023.

ABSTRACT

PURPOSE: To evaluate the feasibility of using a deep learning (DL) model to generate fat-suppression images and detect abnormalities on knee magnetic resonance imaging (MRI) through the fat-suppression image-subtraction method.

MATERIAL AND METHODS: A total of 45 knee MRI studies in patients with knee disorders and 12 knee MRI studies in healthy volunteers were enrolled. The DL model was developed using 2-dimensional convolutional neural networks for generating fat-suppression images and subtracting generated fat-suppression images without any abnormal findings from those with normal/abnormal findings and detecting/classifying abnormalities on knee MRI. The image qualities of the generated fat-suppression images and subtraction-images were assessed. The accuracy, average precision, average recall, F-measure, sensitivity, and area under the receiver operator characteristic curve (AUROC) of DL for each abnormality were calculated.

RESULTS: A total of 2472 image datasets, each consisting of one slice of original T1WI, original intermediate-weighted images, generated fat-suppression (FS)-intermediate-weighted images without any abnormal findings, generated FS-intermediate-weighted images with normal/abnormal findings, and subtraction images between the generated FS-intermediate-weighted images at the same cross-section, were created. The generated fat-suppression images were of adequate image quality. Of the 2472 subtraction-images, 2203 (89.1%) were judged to be of adequate image quality. The accuracies for overall abnormalities, anterior cruciate ligament, bone marrow, cartilage, meniscus, and others were 89.5-95.1%. The average precision, average recall, and F-measure were 73.4-90.6%, 77.5-89.4%, and 78.4-89.4%, respectively. The sensitivity was 57.4-90.5%. The AUROCs were 0.910-0.979.

CONCLUSIONS: The DL model was able to generate fat-suppression images of sufficient quality to detect abnormalities on knee MRI through the fat-suppression image-subtraction method.

PMID:38362017 | PMC:PMC10867951 | DOI:10.5114/pjr.2023.133660

EClinicalMedicine. 2024 Feb 9;69:102466. doi: 10.1016/j.eclinm.2024.102466. eCollection 2024 Mar.

ABSTRACT

BACKGROUND: Voiding cystourethrography (VCUG) is the gold standard for the diagnosis and grading of vesicoureteral reflux (VUR). However, VUR grading from voiding cystourethrograms is highly subjective with low reliability. This study aimed to develop a deep learning model to improve reliability for VUR grading on VCUG and compare its performance to that of clinicians.

METHODS: In this retrospective study in China, VCUG images were collected between January 2019 and September 2022 from our institution as an internal dataset for training and 4 external data sets as external testing set for validation. Samples were divided into training (N = 1000) and validation sets (N = 500), internal testing set (N = 168), and external testing set (N = 280). An ensemble learning-based model, Deep-VCUG, using Res-Net 101 and the voting methods was developed to predict VUR grade. The grading performance was assessed using heatmaps, area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, and F1 score in the internal and external testing set. The performances of four clinicians (2 pediatric urologists and 2 radiologists) with and without the Deep-VCUG assisted to predict VUR grade were explored in external testing sets.

FINDINGS: A total of 1948 VCUG images were collected (Internal dataset = 1668; multi-center external dataset = 280). For assessing unilateral VUR grading, the Deep-VCUG achieved AUCs of 0.962 (95% confidence interval [CI]: 0.943-0.978) and 0.944 (95% [CI]: 0.921-0.964) in the internal and external testing sets, respectively, for bilateral VUR grading, the Deep-VCUG also achieved high AUCs of 0.960 (95% [CI]: 0.922-0.983) and 0.924 (95% [CI]: 0.887-0.957). The Deep-VCUG model using voting method outperformed single model and clinician in terms of classification based on VCUG image. Moreover, Under the Dee-VCUG assisted, the classification ability of junior and senior clinicians was significantly improved.

INTERPRETATION: The Deep-VCUG model is a generalizable, objective, and accurate tool for vesicoureteral reflux grading based on VCUG imaging and had good assistance with clinicians to VUR grading applicability.

FUNDING: This study was supported by Natural Science Foundation of China, "Fuqing Scholar" Student Scientific Research Program of Shanghai Medical College, Fudan University, and the Program of Greater Bay Area Institute of Precision Medicine (Guangzhou).

PMID:38361995 | PMC:PMC10867607 | DOI:10.1016/j.eclinm.2024.102466

Data Brief. 2024 Feb 1;53:110131. doi: 10.1016/j.dib.2024.110131. eCollection 2024 Apr.

ABSTRACT

This paper introduces a video dataset for semantic segmentation of road potholes. This dataset contains 619 high-resolution videos captured in January 2023, covering locations in eight villages within the Hulu Sungai Tengah regency of South Kalimantan, Indonesia. The dataset is divided into three main folders, namely train, val, and test. The train, val, and test folders contain 372 videos for training, 124 videos for validation, and 123 videos for testing, respectively. Each of these main folders has two subfolders, ``RGB'' for the video in the RGB format and ``mask'' for the ground truth segmentation. These videos are precisely two seconds long, containing 48 frames each, and all are in MP4 format. The dataset offers remarkable flexibility, accommodating various research needs, from full-video segmentation to frame extraction. It enables researchers to create ground truth annotations and change the combination of videos in the folders according to their needs. This resource is an asset for researchers, engineers, policymakers, and anyone interested in advancing algorithms for pothole detection and analysis. This dataset allows for benchmarking semantic segmentation algorithms, conducting comparative studies on pothole detection methods, and exploring innovative approaches, offering valuable contributions to the computer vision community.

PMID:38361975 | PMC:PMC10867608 | DOI:10.1016/j.dib.2024.110131

Virus Evol. 2023 Dec 20;10(1):vead083. doi: 10.1093/ve/vead083. eCollection 2024.

ABSTRACT

The rapid emergence and spread of antimicrobial resistance across the globe have prompted the usage of bacteriophages (i.e. viruses that infect bacteria) in a variety of applications ranging from agriculture to biotechnology and medicine. In order to effectively guide the application of bacteriophages in these multifaceted areas, information about their host ranges-that is the bacterial strains or species that a bacteriophage can successfully infect and kill-is essential. Utilizing sixteen broad-spectrum (polyvalent) bacteriophages with experimentally validated host ranges, we here benchmark the performance of eleven recently developed computational host range prediction tools that provide a promising and highly scalable supplement to traditional, but laborious, experimental procedures. We show that machine- and deep-learning approaches offer the highest levels of accuracy and precision-however, their predominant predictions at the species- or genus-level render them ill-suited for applications outside of an ecosystems metagenomics framework. In contrast, only moderate sensitivity (<80 per cent) could be reached at the strain-level, albeit at low levels of precision (<40 per cent). Taken together, these limitations demonstrate that there remains room for improvement in the active scientific field of in silico host prediction to combat the challenge of guiding experimental designs to identify the most promising bacteriophage candidates for any given application.

PMID:38361822 | PMC:PMC10868548 | DOI:10.1093/ve/vead083

Front Bioeng Biotechnol. 2024 Feb 1;12:1330713. doi: 10.3389/fbioe.2024.1330713. eCollection 2024.

ABSTRACT

Over the past 35 years, studies conducted worldwide have revealed a threefold increase in the incidence of thyroid cancer. Strain elastography is a new imaging technique to identify benign and malignant thyroid nodules due to its sensitivity to tissue stiffness. However, there are certain limitations of this technique, particularly in terms of standardization of the compression process, evaluation of results and several assumptions used in commercial strain elastography modes for the purpose of simplifying imaging analysis. In this work, we propose a novel conditional generative adversarial network (TSE-GAN) for automatically generating thyroid strain elastograms, which adopts a global-to-local architecture to improve the ability of extracting multi-scale features and develops an adaptive deformable U-net structure in the sub-generator to apply effective deformation. Furthermore, we introduce a Lab-based loss function to induce the networks to generate realistic thyroid elastograms that conform to the probability distribution of the target domain. Qualitative and quantitative assessments are conducted on a clinical dataset provided by Shanghai Sixth People's Hospital. Experimental results demonstrate that thyroid elastograms generated by the proposed TSE-GAN outperform state-of-the-art image translation methods in meeting the needs of clinical diagnostic applications and providing practical value.

PMID:38361791 | PMC:PMC10867782 | DOI:10.3389/fbioe.2024.1330713

Front Oncol. 2024 Feb 1;14:1281572. doi: 10.3389/fonc.2024.1281572. eCollection 2024.

ABSTRACT

OBJECTIVE: This study aimed to evaluate the value of 18F-FDG PET/CT radiomics in predicting EGFR gene mutations in non-small cell lung cancer by meta-analysis.

METHODS: The PubMed, Embase, Cochrane Library, Web of Science, and CNKI databases were searched from the earliest available date to June 30, 2023. The meta-analysis was performed using the Stata 15.0 software. The methodological quality and risk of bias of included studies were assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 and Radiomics Quality Score criteria. The possible causes of heterogeneity were analyzed by meta-regression.

RESULTS: A total of 17 studies involving 3763 non-small cell lung cancer patients were finally included. We analyzed 17 training cohorts and 10 validation cohorts independently. Within the training cohort, the application of 18F-FDG PET/CT radiomics in predicting EGFR mutations in NSCLC demonstrated a sensitivity of 0.76 (95% CI: 0.70-0.81) and a specificity of 0.78 (95% CI: 0.74-0.82), accompanied by a positive likelihood ratio of 3.5 (95% CI:3.0-4.2), a negative likelihood ratio of 0.31 (95% CI: 0.24-0.39), a diagnostic odds ratio of 11.0 (95% CI: 8.0-16.0), and an area under the curve (AUC) of 0.84 (95% CI: 0.80-0.87). In the validation cohort, the values included a sensitivity of 0.76 (95% CI: 0.67-0.83), a specificity of 0.75 (95% CI: 0.68-0.80), a positive likelihood ratio of 3.0 (95% CI:2.4-3.8), a negative likelihood ratio of 0.32 (95% CI: 0.24-0.44), a diagnostic odds ratio of 9 (95% CI: 6-15), and an AUC of 0.82 (95% CI: 0.78-0.85). The average Radiomics Quality Score (RQS) across studies was 10.47 ± 4.72. Meta-regression analysis identifies the application of deep learning and regions as sources of heterogeneity.

CONCLUSION: 18F-FDG PET/CT radiomics may be useful in predicting mutation status of the EGFR gene in non-small cell lung cancer.

SYSTEMATIC REVIEW REGISTRATION: https://www.crd.york.ac.uk/PROSPERO, identifier CRD42022385364.

PMID:38361781 | PMC:PMC10867100 | DOI:10.3389/fonc.2024.1281572

iScience. 2024 Jan 30;27(3):109027. doi: 10.1016/j.isci.2024.109027. eCollection 2024 Mar 15.

ABSTRACT

By providing high-resolution of cell-to-cell variation in gene expression, single-cell RNA sequencing (scRNA-seq) offers insights into cell heterogeneity, differentiating dynamics, and disease mechanisms. However, challenges such as low capture rates and dropout events can introduce noise in data analysis. Here, we propose a deep neural generative framework, the dynamic batching adversarial autoencoder (DB-AAE), which excels at denoising scRNA-seq datasets. DB-AAE directly captures optimal features from input data and enhances feature preservation, including cell type-specific gene expression patterns. Comprehensive evaluation on simulated and real datasets demonstrates that DB-AAE outperforms other methods in denoising accuracy and biological signal preservation. It also improves the accuracy of other algorithms in establishing pseudo-time inference. This study highlights DB-AAE's effectiveness and potential as a valuable tool for enhancing the quality and reliability of downstream analyses in scRNA-seq research.

PMID:38361616 | PMC:PMC10867661 | DOI:10.1016/j.isci.2024.109027

Sci Rep. 2024 Feb 15;14(1):3826. doi: 10.1038/s41598-024-54067-z.

ABSTRACT

In this work, we present GAROM, a new approach for reduced order modeling (ROM) based on generative adversarial networks (GANs). GANs attempt to learn to generate data with the same statistics of the underlying distribution of a dataset, using two neural networks, namely discriminator and generator. While widely applied in many areas of deep learning, little research is done on their application for ROM, i.e. approximating a high-fidelity model with a simpler one. In this work, we combine the GAN and ROM framework, introducing a data-driven generative adversarial model able to learn solutions to parametric differential equations. In the presented methodology, the discriminator is modeled as an autoencoder, extracting relevant features of the input, and a conditioning mechanism is applied to the generator and discriminator networks specifying the differential equation parameters. We show how to apply our methodology for inference, provide experimental evidence of the model generalization, and perform a convergence study of the method.

PMID:38361023 | DOI:10.1038/s41598-024-54067-z

J Imaging Inform Med. 2024 Feb 15. doi: 10.1007/s10278-024-01022-z. Online ahead of print.

ABSTRACT

We aimed to develop and validate a deep learning-based system using pre-therapy computed tomography (CT) images to detect epidermal growth factor receptor (EGFR)-mutant status in patients with non-small cell lung cancer (NSCLC) and predict the prognosis of advanced-stage patients with EGFR mutations treated with EGFR tyrosine kinase inhibitors (TKI). This retrospective, multicenter study included 485 patients with NSCLC from four hospitals. Of them, 339 patients from three centers were included in the training dataset to develop an EfficientNetV2-L-based model (EME) for predicting EGFR-mutant status, and the remaining patients were assigned to an independent test dataset. EME semantic features were extracted to construct an EME-prognostic model to stratify the prognosis of EGFR-mutant NSCLC patients receiving EGFR-TKI. A comparison of EME and radiomics was conducted. Additionally, we included patients from The Cancer Genome Atlas lung adenocarcinoma dataset with both CT images and RNA sequencing data to explore the biological associations between EME score and EGFR-related biological processes. EME obtained an area under the curve (AUC) of 0.907 (95% CI 0.840-0.926) on the test dataset, superior to the radiomics model (P = 0.007). The EME and radiomics fusion model showed better (AUC, 0.941) but not significantly increased performance (P = 0.895) compared with EME. In prognostic stratification, the EME-prognostic model achieved the best performance (C-index, 0.711). Moreover, the EME-prognostic score showed strong associations with biological pathways related to EGFR expression and EGFR-TKI efficacy. EME demonstrated a non-invasive and biologically interpretable approach to predict EGFR status, stratify survival prognosis, and correlate biological pathways in patients with NSCLC.

PMID:38361006 | DOI:10.1007/s10278-024-01022-z

Methods. 2024 Feb 13:S1046-2023(24)00042-2. doi: 10.1016/j.ymeth.2024.01.018. Online ahead of print.

ABSTRACT

MOTIVATION: Drug-target interaction prediction is an important area of research to predict whether there is an interaction between a drug molecule and its target protein. It plays a critical role in drug discovery and development by facilitating the identification of potential drug candidates and expediting the overall process. Given the time-consuming, expensive, and high-risk nature of traditional drug discovery methods, the prediction of drug-target interactions has become an indispensable tool. Using machine learning and deep learning to tackle this class of problems has become a mainstream approach, and graph-based models have recently received much attention in this field. However, many current graph-based Drug-Target Interaction (DTI) prediction methods rely on manually defined rules to construct the Drug-Protein Pair (DPP) network during the DPP representation learning process. However, these methods fail to capture the true underlying relationships between drug molecules and target proteins.

RESULTS: We propose GSL-DTI, an automatic graph structure learning model used for predicting drug-target interactions (DTIs). Initially, we integrate large-scale heterogeneous networks using a graph convolution network based on meta-paths, effectively learning the representations of drugs and target proteins. Subsequently, we construct drug-protein pairs based on these representations. In contrast to previous studies that construct DPP networks based on manual rules, our method introduces an automatic graph structure learning approach. This approach utilizes a filter gate on the affinity scores of DPPs and relies on the classification loss of downstream tasks to guide the learning of the underlying DPP network structure. Based on the learned DPP network, we transform the prediction of drug-target interactions into a node classification problem. The comprehensive experiments conducted on three public datasets have shown the superiority of GSL-DTI in the tasks of DTI prediction. Additionally, GSL-DTI provides a fresh perspective for advancing research in graph structure learning for DTI prediction.

PMID:38360082 | DOI:10.1016/j.ymeth.2024.01.018

J Nucl Med. 2024 Feb 15:jnumed.123.266018. doi: 10.2967/jnumed.123.266018. Online ahead of print.

ABSTRACT

Reliable performance of PET segmentation algorithms on clinically relevant tasks is required for their clinical translation. However, these algorithms are typically evaluated using figures of merit (FoMs) that are not explicitly designed to correlate with clinical task performance. Such FoMs include the Dice similarity coefficient (DSC), the Jaccard similarity coefficient (JSC), and the Hausdorff distance (HD). The objective of this study was to investigate whether evaluating PET segmentation algorithms using these task-agnostic FoMs yields interpretations consistent with evaluation on clinically relevant quantitative tasks. Methods: We conducted a retrospective study to assess the concordance in the evaluation of segmentation algorithms using the DSC, JSC, and HD and on the tasks of estimating the metabolic tumor volume (MTV) and total lesion glycolysis (TLG) of primary tumors from PET images of patients with non-small cell lung cancer. The PET images were collected from the American College of Radiology Imaging Network 6668/Radiation Therapy Oncology Group 0235 multicenter clinical trial data. The study was conducted in 2 contexts: (1) evaluating conventional segmentation algorithms, namely those based on thresholding (SUVmax40% and SUVmax50%), boundary detection (Snakes), and stochastic modeling (Markov random field-Gaussian mixture model); (2) evaluating the impact of network depth and loss function on the performance of a state-of-the-art U-net-based segmentation algorithm. Results: Evaluation of conventional segmentation algorithms based on the DSC, JSC, and HD showed that SUVmax40% significantly outperformed SUVmax50%. However, SUVmax40% yielded lower accuracy on the tasks of estimating MTV and TLG, with a 51% and 54% increase, respectively, in the ensemble normalized bias. Similarly, the Markov random field-Gaussian mixture model significantly outperformed Snakes on the basis of the task-agnostic FoMs but yielded a 24% increased bias in estimated MTV. For the U-net-based algorithm, our evaluation showed that although the network depth did not significantly alter the DSC, JSC, and HD values, a deeper network yielded substantially higher accuracy in the estimated MTV and TLG, with a decreased bias of 91% and 87%, respectively. Additionally, whereas there was no significant difference in the DSC, JSC, and HD values for different loss functions, up to a 73% and 58% difference in the bias of the estimated MTV and TLG, respectively, existed. Conclusion: Evaluation of PET segmentation algorithms using task-agnostic FoMs could yield findings discordant with evaluation on clinically relevant quantitative tasks. This study emphasizes the need for objective task-based evaluation of image segmentation algorithms for quantitative PET.

PMID:38360049 | DOI:10.2967/jnumed.123.266018

Healthc Inform Res. 2024 Jan;30(1):42-48. doi: 10.4258/hir.2024.30.1.42. Epub 2024 Jan 31.

ABSTRACT

OBJECTIVES: Telemedicine is firmly established in the healthcare landscape of many countries. Acute respiratory infections are the most common reason for telemedicine consultations. A throat examination is important for diagnosing bacterial pharyngitis, but this is challenging for doctors during a telemedicine consultation. A solution could be for patients to upload images of their throat to a web application. This study aimed to develop a deep learning model for the automated diagnosis of exudative pharyngitis. Thereafter, the model will be deployed online.

METHODS: We used 343 throat images (139 with exudative pharyngitis and 204 without pharyngitis) in the study. ImageDataGenerator was used to augment the training data. The convolutional neural network models of MobileNetV3, ResNet50, and EfficientNetB0 were implemented to train the dataset, with hyperparameter tuning.

RESULTS: All three models were trained successfully; with successive epochs, the loss and training loss decreased, and accuracy and training accuracy increased. The EfficientNetB0 model achieved the highest accuracy (95.5%), compared to MobileNetV3 (82.1%) and ResNet50 (88.1%). The EfficientNetB0 model also achieved high precision (1.00), recall (0.89) and F1-score (0.94).

CONCLUSIONS: We trained a deep learning model based on EfficientNetB0 that can diagnose exudative pharyngitis. Our model was able to achieve the highest accuracy, at 95.5%, out of all previous studies that used machine learning for the diagnosis of exudative pharyngitis. We have deployed the model on a web application that can be used to augment the doctor's diagnosis of exudative pharyngitis.

PMID:38359848 | DOI:10.4258/hir.2024.30.1.42

Healthc Inform Res. 2024 Jan;30(1):3-15. doi: 10.4258/hir.2024.30.1.3. Epub 2024 Jan 31.

ABSTRACT

OBJECTIVES: Medical artificial intelligence (AI) has recently attracted considerable attention. However, training medical AI models is challenging due to privacy-protection regulations. Among the proposed solutions, federated learning (FL) stands out. FL involves transmitting only model parameters without sharing the original data, making it particularly suitable for the medical field, where data privacy is paramount. This study reviews the application of FL in the medical domain.

METHODS: We conducted a literature search using the keywords "federated learning" in combination with "medical," "healthcare," or "clinical" on Google Scholar and PubMed. After reviewing titles and abstracts, 58 papers were selected for analysis. These FL studies were categorized based on the types of data used, the target disease, the use of open datasets, the local model of FL, and the neural network model. We also examined issues related to heterogeneity and security.

RESULTS: In the investigated FL studies, the most commonly used data type was image data, and the most studied target diseases were cancer and COVID-19. The majority of studies utilized open datasets. Furthermore, 72% of the FL articles addressed heterogeneity issues, while 50% discussed security concerns.

CONCLUSIONS: FL in the medical domain appears to be in its early stages, with most research using open data and focusing on specific data types and diseases for performance verification purposes. Nonetheless, medical FL research is anticipated to be increasingly applied and to become a vital component of multi-institutional research.

PMID:38359845 | DOI:10.4258/hir.2024.30.1.3

Sci Rep. 2024 Feb 15;14(1):3845. doi: 10.1038/s41598-024-54502-1.

ABSTRACT

To assess the image quality parameters of dual-energy computed tomography angiography (DECTA) 40-, and 60 keV virtual monoenergetic images (VMIs) combined with deep learning-based image reconstruction model (DLM) and iterative reconstructions (IR). CT scans of 28 post EVAR patients were enrolled. The 60 s delayed phase of DECTA was evaluated. Objective [noise, contrast-to-noise ratio (CNR), signal-to-noise ratio (SNR)] and subjective (overall image quality and endoleak conspicuity - 3 blinded readers assessment) image quality analyses were performed. The following reconstructions were evaluated: VMI 40, 60 keV VMI; IR VMI 40, 60 keV; DLM VMI 40, 60 keV. The noise level of the DLM VMI images was approximately 50% lower than that of VMI reconstruction. The highest CNR and SNR values were measured in VMI DLM images. The mean CNR in endoleak in 40 keV was accounted for as 1.83 ± 1.2; 2.07 ± 2.02; 3.6 ± 3.26 in VMI, VMI IR, and VMI DLM, respectively. The DLM algorithm significantly reduced noise and increased lesion conspicuity, resulting in higher objective and subjective image quality compared to other reconstruction techniques. The application of DLM algorithms to low-energy VMIs significantly enhances the diagnostic value of DECTA in evaluating endoleaks. DLM reconstructions surpass traditional VMIs and IR in terms of image quality.

PMID:38360941 | DOI:10.1038/s41598-024-54502-1

Insights Imaging. 2024 Feb 16;15(1):50. doi: 10.1186/s13244-024-01636-5.

ABSTRACT

Kidney diseases result from various causes, which can generally be divided into neoplastic and non-neoplastic diseases. Deep learning based on medical imaging is an established methodology for further data mining and an evolving field of expertise, which provides the possibility for precise management of kidney diseases. Recently, imaging-based deep learning has been widely applied to many clinical scenarios of kidney diseases including organ segmentation, lesion detection, differential diagnosis, surgical planning, and prognosis prediction, which can provide support for disease diagnosis and management. In this review, we will introduce the basic methodology of imaging-based deep learning and its recent clinical applications in neoplastic and non-neoplastic kidney diseases. Additionally, we further discuss its current challenges and future prospects and conclude that achieving data balance, addressing heterogeneity, and managing data size remain challenges for imaging-based deep learning. Meanwhile, the interpretability of algorithms, ethical risks, and barriers of bias assessment are also issues that require consideration in future development. We hope to provide urologists, nephrologists, and radiologists with clear ideas about imaging-based deep learning and reveal its great potential in clinical practice.Critical relevance statement The wide clinical applications of imaging-based deep learning in kidney diseases can help doctors to diagnose, treat, and manage patients with neoplastic or non-neoplastic renal diseases.Key points• Imaging-based deep learning is widely applied to neoplastic and non-neoplastic renal diseases.• Imaging-based deep learning improves the accuracy of the delineation, diagnosis, and evaluation of kidney diseases.• The small dataset, various lesion sizes, and so on are still challenges for deep learning.

PMID:38360904 | DOI:10.1186/s13244-024-01636-5

Sci Rep. 2024 Feb 15;14(1):3839. doi: 10.1038/s41598-023-50797-8.

ABSTRACT

Breast cancer has the highest incidence rate among women in Ethiopia compared to other types of cancer. Unfortunately, many cases are detected at a stage where a cure is delayed or not possible. To address this issue, mammography-based screening is widely accepted as an effective technique for early detection. However, the interpretation of mammography images requires experienced radiologists in breast imaging, a resource that is limited in Ethiopia. In this research, we have developed a model to assist radiologists in mass screening for breast abnormalities and prioritizing patients. Our approach combines an ensemble of EfficientNet-based classifiers with YOLOv5, a suspicious mass detection method, to identify abnormalities. The inclusion of YOLOv5 detection is crucial in providing explanations for classifier predictions and improving sensitivity, particularly when the classifier fails to detect abnormalities. To further enhance the screening process, we have also incorporated an abnormality detection model. The classifier model achieves an F1-score of 0.87 and a sensitivity of 0.82. With the addition of suspicious mass detection, sensitivity increases to 0.89, albeit at the expense of a slightly lower F1-score of 0.79.

PMID:38360869 | DOI:10.1038/s41598-023-50797-8