Heliyon. 2024 Jul 4;10(14):e33973. doi: 10.1016/j.heliyon.2024.e33973. eCollection 2024 Jul 30.

ABSTRACT

Major depression is a complex psychiatric disorder that includes genetic, neurological, and cognitive factors. Early detection and intervention can prevent progression, and help select the best treatment. Traditional clinical diagnosis tends to be subjective and misdiagnosed. Based on this, this study leverages clinical scale assessments and sequencing data to construct disease prediction models. Firstly, data undergoes preprocessing involving normalization and other requisite procedures. Feature engineering is then applied to curate subsets of features, culminating in the construction of a model through the implementation of machine learning and deep learning algorithms. In this study, 18 features with significant differences between patients and healthy controls were selected. The depression recognition model was constructed by deep learning with an accuracy of 87.26 % and an AUC of 91.56 %, which can effectively distinguish patients with depression from healthy controls. In addition, 33 features selected by recursive feature elimination method were used to construct a prognostic effect model of patients after 2 weeks of treatment, with an accuracy of 75.94 % and an AUC of 83.33 %. The results show that the deep learning algorithm based on clinical and sequencing data has good accuracy and provides an objective and accurate method for the diagnosis and pharmacodynamic prediction of depression. Furthermore, the selected differential features can serve as candidate biomarkers to provide valuable clues for diagnosis and efficacy prediction.

PMID:39130405 | PMC:PMC11315137 | DOI:10.1016/j.heliyon.2024.e33973

Heliyon. 2024 Jul 15;10(14):e34685. doi: 10.1016/j.heliyon.2024.e34685. eCollection 2024 Jul 30.

ABSTRACT

Today, the number of published scientific articles is increasing day by day, and this has made the process of searching for articles more difficult. The need to provide specific recommender systems (RSs) for suggesting scientific articles is strongly felt in this situation. Because searching for articles based only on matching the titles or content of other articles is not an efficient process. In this research, the combination of two content analysis and citation network is used to design an RS for scientific articles (RECSA). In RECSA, natural language processing and deep learning techniques are used to process the titles and extract the content attributes of the articles. For this purpose, first, the titles of the articles are pre-processed, and by using the Term Frequency Inverse Document Frequency (TF-IDF) criterion, the importance of each word in the title is estimated. Then the dimensions of the obtained attributes are reduced by using a convolutional neural network (CNN). Then, by using the cosine similarity criterion, the content similarity matrix of the articles is calculated based on the attribute vectors. Also, the link prediction approach is used to analyze the connections of scientific articles' citation network. Finally, in the third step of RECSA, the two similarity matrices calculated in the previous steps are combined using an influence coefficient parameter to obtain the final similarity matrix, and the recommendation operation is based on the highest similarity value. The efficiency of RECSA has been evaluated from different aspects and the results have been compared with previous works. According to the results, utilizing the combination of TF-IDF and CNN for analyzing content-based features, leads to at least 0.32 % improvement in terms of precision compared to previous works. Also, by integrating citation and content-based data, the precision of first suggestion in RECSA would be 99.01 % which indicates the minimum improvement of 0.9 % compared to compared methods. The results show that by using RECSA, the recommendation can be done with higher accuracy and efficiency.

PMID:39130403 | PMC:PMC11315118 | DOI:10.1016/j.heliyon.2024.e34685

Gastro Hep Adv. 2022 Feb 3;1(1):29-37. doi: 10.1016/j.gastha.2021.09.003. eCollection 2022.

ABSTRACT

BACKGROUND AND AIMS: Radiofrequency ablation (RFA) is a widely accepted, minimally invasive treatment for hepatocellular carcinoma (HCC). This study aimed to develop a machine learning (ML) model to predict the risk of HCC recurrence after RFA treatment for individual patients.

METHODS: We included a total of 1778 patients with treatment-naïve HCC who underwent RFA. The cumulative probability of overall recurrence after the initial RFA treatment was 78.9% and 88.0% at 5 and 10 years, respectively. We developed a conventional Cox proportional hazard model and 6 ML models-including the deep learning-based DeepSurv model. Model performance was evaluated using Harrel's c-index and was validated externally using the split-sample method.

RESULTS: The gradient boosting decision tree (GBDT) model achieved the best performance with a c-index of 0.67 from external validation, and it showed a high discriminative ability in stratifying the external validation sample into 2, 3, and 4 different risk groups (P < .001 among all risk groups). The c-index of DeepSurv was 0.64. In order of significance, the tumor number, serum albumin level, and des-gamma-carboxyprothrombin level were the most important variables for the prediction of HCC recurrence in the GBDT model. Also, the current GBDT model enabled the output of a personalized cumulative recurrence prediction curve for each patient.

CONCLUSION: We developed a novel ML model for the personalized risk prediction of HCC recurrence after RFA treatment. The current model may lead to the personalization of effective follow-up strategies after RFA treatment according to the risk stratification of HCC recurrence.

PMID:39129938 | PMC:PMC11308827 | DOI:10.1016/j.gastha.2021.09.003

Gastro Hep Adv. 2022 Feb 3;1(1):70-78. doi: 10.1016/j.gastha.2021.11.001. eCollection 2022.

ABSTRACT

The access to increasing volumes of scientific and clinical data, particularly with the implementation of electronic health records, has reignited an enthusiasm for artificial intelligence and its application to the health sciences. This interest has reached a crescendo in the past few years with the development of several machine learning- and deep learning-based medical technologies. The impact on research and clinical practice within gastroenterology and hepatology has already been significant, but the near future promises only further integration of artificial intelligence and machine learning into this field. The concepts underlying artificial intelligence and machine learning initially seem intimidating, but with increasing familiarity, they will become essential skills in every clinician's toolkit. In this review, we provide a guide to the fundamentals of machine learning, a concentrated area of study within artificial intelligence that has been built on a foundation of classical statistics. The most common machine learning methodologies, including those involving deep learning, are also described.

PMID:39129929 | PMC:PMC11307451 | DOI:10.1016/j.gastha.2021.11.001

Front Plant Sci. 2024 Jul 26;15:1395558. doi: 10.3389/fpls.2024.1395558. eCollection 2024.

ABSTRACT

Milk thistle, Silybum marianum (L.), is a well-known medicinal plant used for the treatment of liver diseases due to its high content of silymarin. The seeds contain elaiosome, a fleshy structure attached to the seeds, which is believed to be a rich source of many metabolites including silymarin. Segmentation of elaiosomes using only image analysis is difficult, and this makes it impossible to quantify the elaiosome phenotypes. This study proposes a new approach for semi-automated detection and segmentation of elaiosomes in milk thistle seed using the Detectron2 deep learning algorithm. One hundred manually labeled images were used to train the initial elaiosome detection model. This model was used to predict elaiosome from new datasets, and the precise predictions were manually selected and used as new labeled images for retraining the model. Such semi-automatic image labeling, i.e., using the prediction results of the previous stage for retraining the model, allowed the production of sufficient labeled data for retraining. Finally, a total of 6,000 labeled images were used to train Detectron2 for elaiosome detection and attained a promising result. The results demonstrate the effectiveness of Detectron2 in detecting milk thistle seed elaiosomes with an accuracy of 99.9%. The proposed method automatically detects and segments elaiosome from the milk thistle seed. The predicted mask images of elaiosome were used to analyze its area as one of the seed phenotypic traits along with other seed morphological traits by image-based high-throughput phenotyping in ImageJ. Enabling high-throughput phenotyping of elaiosome and other seed morphological traits will be useful for breeding milk thistle cultivars with desirable traits.

PMID:39129764 | PMC:PMC11310567 | DOI:10.3389/fpls.2024.1395558

Circulation. 2024 Aug 12. doi: 10.1161/CIRCULATIONAHA.124.069047. Online ahead of print.

ABSTRACT

BACKGROUND: Diagnosis of mitral regurgitation (MR) requires careful evaluation by echocardiography with Doppler imaging. This study presents the development and validation of a fully automated deep learning pipeline for identifying apical 4-chamber view videos with color Doppler echocardiography and detecting clinically significant (moderate or severe) MR from transthoracic echocardiograms.

METHODS: A total of 58 614 transthoracic echocardiograms (2 587 538 videos) from Cedars-Sinai Medical Center were used to develop and test an automated pipeline to identify apical 4-chamber view videos with color Doppler across the mitral valve and then assess MR severity. The model was tested internally on a test set of 1800 studies (80 833 videos) from Cedars-Sinai Medical Center and externally evaluated in a geographically distinct cohort of 915 studies (46 890 videos) from Stanford Healthcare.

RESULTS: In the held-out Cedars-Sinai Medical Center test set, the view classifier demonstrated an area under the curve (AUC) of 0.998 (0.998-0.999) and correctly identified 3452 of 3539 echocardiography videos as having color Doppler information across the mitral valve (sensitivity of 0.975 [0.968-0.982] and specificity of 0.999 [0.999-0.999] compared with manually curated videos). In the external test cohort from Stanford Healthcare, the view classifier correctly identified 1051 of 1055 manually curated videos with color Doppler information across the mitral valve (sensitivity of 0.996 [0.990-1.000] and specificity of 0.999 [0.999-0.999]). In the Cedars-Sinai Medical Center test cohort, MR moderate or greater in severity was detected with an AUC of 0.916 (0.899-0.932) and severe MR was detected with an AUC of 0.934 (0.913-0.953). In the Stanford Healthcare test cohort, the model detected MR moderate or greater in severity with an AUC of 0.951 (0.924-0.973) and severe MR with an AUC of 0.969 (0.946-0.987).

CONCLUSIONS: In this study, a novel automated pipeline for identifying clinically significant MR from full transthoracic echocardiography studies demonstrated excellent performance across large numbers of studies and across multiple institutions. Such an approach has the potential for automated screening and surveillance of MR.

PMID:39129623 | DOI:10.1161/CIRCULATIONAHA.124.069047

Brief Bioinform. 2024 Jul 25;25(5):bbae387. doi: 10.1093/bib/bbae387.

ABSTRACT

Enzymatic reaction kinetics are central in analyzing enzymatic reaction mechanisms and target-enzyme optimization, and thus in biomanufacturing and other industries. The enzyme turnover number (kcat) and Michaelis constant (Km), key kinetic parameters for measuring enzyme catalytic efficiency, are crucial for analyzing enzymatic reaction mechanisms and the directed evolution of target enzymes. Experimental determination of kcat and Km is costly in terms of time, labor, and cost. To consider the intrinsic connection between kcat and Km and further improve the prediction performance, we propose a universal pretrained multitask deep learning model, MPEK, to predict these parameters simultaneously while considering pH, temperature, and organismal information. Through testing on the same kcat and Km test datasets, MPEK demonstrated superior prediction performance over the previous models. Specifically, MPEK achieved the Pearson coefficient of 0.808 for predicting kcat, improving ca. 14.6% and 7.6% compared to the DLKcat and UniKP models, and it achieved the Pearson coefficient of 0.777 for predicting Km, improving ca. 34.9% and 53.3% compared to the Kroll_model and UniKP models. More importantly, MPEK was able to reveal enzyme promiscuity and was sensitive to slight changes in the mutant enzyme sequence. In addition, in three case studies, it was shown that MPEK has the potential for assisted enzyme mining and directed evolution. To facilitate in silico evaluation of enzyme catalytic efficiency, we have established a web server implementing this model, which can be accessed at http://mathtc.nscc-tj.cn/mpek.

PMID:39129365 | DOI:10.1093/bib/bbae387

J Mol Biol. 2024 Aug 9:168733. doi: 10.1016/j.jmb.2024.168733. Online ahead of print.

ABSTRACT

Detecting chromosome structural abnormalities in medical genetics is essential for diagnosing genetic disorders and understanding their implications for an individual's health. However, existing computational methods are formulated as a binary-class classification problem trained only on representations of positive/negative chromosome pairs. This paper introduces an innovative framework for detecting chromosome abnormalities with banding resolution, capable of precisely identifying and masking the specific abnormal regions. We highlight a pixel-level abnormal mapping strategy guided by banding features. This approach integrates data from both the original image and banding characteristics, enhancing the interpretability of prediction results for cytogeneticists. Furthermore, we have implemented an ensemble approach that pairs a discriminator with a conditional random field heatmap generator. This combination significantly reduces the false positive rate in abnormality screening. We benchmarked our proposed framework with state-of-the-art (SOTA) methods in abnormal screening and structural abnormal region segmentation. Our results show cutting-edge effectiveness and greatly reduce the high false positive rate. It also shows superior performance in sensitivity and segmentation accuracy. Being able to identify abnormal regions consistently shows that our model has demonstrated significant clinical utility with high model interpretability. BRChromNet is open-sourced and available at https://github.com/frankchen121212/BR-ChromNet.

PMID:39128787 | DOI:10.1016/j.jmb.2024.168733

J Neurosci Methods. 2024 Aug 9:110247. doi: 10.1016/j.jneumeth.2024.110247. Online ahead of print.

ABSTRACT

The prevalence of brain tumor disorders is currently a global issue. In general, radiography, which includes a large number of images, is an efficient method for diagnosing these life-threatening disorders. The biggest issue in this area is that it takes a radiologist a long time and is physically strenuous to look at all the images. As a result, research into developing systems based on machine learning to assist radiologists in diagnosis continues to rise daily. Convolutional neural networks (CNNs), one type of deep learning approach, have been pivotal in achieving state-of-the-art results in several medical imaging applications, including the identification of brain tumors. CNN hyperparameters are typically set manually for segmentation and classification, which might take a while and increase the chance of using suboptimal hyperparameters for both tasks. Bayesian optimization is a useful method for updating the deep CNN's optimal hyperparameters. The CNN network, however, can be considered a "black box" model because of how difficult it is to comprehend the information it stores because of its complexity. Therefore, this problem can be solved by using Explainable Artificial Intelligence (XAI) tools, which provide doctors with a realistic explanation of CNN's assessments. Implementation of deep learning-based systems in real-time diagnosis is still rare. One of the causes could be that these methods don't quantify the Uncertainty in the predictions, which could undermine trust in the AI-based diagnosis of diseases. To be used in real-time medical diagnosis, CNN-based models must be realistic and appealing, and uncertainty needs to be evaluated. So, a novel three-phase strategy is proposed for segmenting and classifying brain tumors. Segmentation of brain tumors using the DeeplabV3+ model is first performed with tuning of hyperparameters using Bayesian optimization. For classification, features from state-of-the-art deep learning models Darknet53 and mobilenetv2 are extracted and fed to SVM for classification, and hyperparameters of SVM are also optimized using a Bayesian approach. The second step is to understand whatever portion of the images CNN uses for feature extraction using XAI algorithms. Using confusion entropy, the Uncertainty of the Bayesian optimized classifier is finally quantified. Based on a Bayesian-optimized deep learning framework, the experimental findings demonstrate that the proposed method outperforms earlier techniques, achieving a 97% classification accuracy and a 0.98 global accuracy.

PMID:39128599 | DOI:10.1016/j.jneumeth.2024.110247

Sci Total Environ. 2024 Aug 9:175434. doi: 10.1016/j.scitotenv.2024.175434. Online ahead of print.

ABSTRACT

Pollution fluxes from rivers into the sea is currently the main source of pollutants in nearshore areas. Based on the source-sink process of the basin-estuary-coastal waters system, the pollution fluxes into the sea and its spatiotemporal heterogeneity were estimated. A deep learning-based model was established to simplify the estimation of pollution fluxes into the sea, with socio-economic drivers and meteorological data as input variables. A method for estimating the contribution rate of pollution fluxes from different spatial gradient was proposed. The study found that (1) the pollution fluxes into the sea of total nitrogen (TN) and total phosphorus (TP) in the Bohai Sea Rim Basin (BSRB) in 1980, 1990, 2000, 2010, and 2020 were 25.38 × 104, 26.12 × 104, 27.27 × 104, 29.82 × 104, 25.31 × 104 and 1.32 × 104, 2.14 × 104, 2.09 × 104, 1.87 × 104, 1.68 × 104 tons, respectively. (2) The proportion of rural life and livestock to the TN was the highest, accounting for 39.18 % and 21.19 %, respectively. The proportion of livestock to the TP was the highest, accounting for 39.20 %, followed by rural life, accounting for 24.72 %. The results indicated that the pollution fluxes in the BSRB was related to human economic activities and relevant environmental protection measures. (3) The deep learning-based model established to estimate runoff pollution fluxes into the sea had the accuracy of over 90 %. (4) As for contribution rate, in terms of elevation, the range of 0-100 m had the highest proportion, accounting for 39.65 %. The range of 50-100 km from coastline had the highest proportion, accounting for 18.11 %. In terms of district, coastal area has the highest proportion, accounting for 38.00 %. This study revealed the changing trends and driving mechanisms of pollution fluxes into the sea over the past 40 years and established a simplified deep learning-based model for estimating pollution fluxes into the sea. Then, we identified regions with high pollution contribution rates. The results can provide scientific references for the adaptive management of the nearshore areas based on the ecosystem.

PMID:39128526 | DOI:10.1016/j.scitotenv.2024.175434

J Electrocardiol. 2024 Jul 22;86:153761. doi: 10.1016/j.jelectrocard.2024.153761. Online ahead of print.

ABSTRACT

The electrocardiogram (ECG) signal is susceptible to interference from unknown noises during the acquisition process due to their low frequency and amplitude, resulting in the loss of significant information in the signals. Recent advancements in deep learning models have shown promising results in signal processing. However, these models lack robustness against various types of noise and often overlook the gradient difference between denoised and original signals. In this study, we propose a novel deep learning denoising method based on an attention half instance normalization block (AHIN block) and a gradient difference max loss function (GDM Loss). Our approach consists of two stages: firstly, we input the noisy ECG signal to obtain a denoised version; secondly, we reconstruct the denoised signal by fusing preliminary results from the first stage while correcting waveform distortions caused by initial denoising to minimize information loss. Additionally, we introduce a new loss function that considers differences between slopes of the denoised ECG signal and clean ECG signal. To validate our proposed method's effectiveness, extensive experiments were conducted on both our model architecture and loss function compared with other state-of-the-art methods. Results demonstrate that our approach achieves excellent performance in terms of both signal-to-noise ratio (SNR) and root-mean-square error (RMSE). The proposed noise reduction method improves 8.86%, 12.05% and 10.50% respectively under BW, MA and EM noise.

PMID:39128171 | DOI:10.1016/j.jelectrocard.2024.153761

Sci Rep. 2024 Aug 11;14(1):18630. doi: 10.1038/s41598-024-69153-5.

ABSTRACT

The importance of 3D reconstruction of coronary arteries using multiple coronary angiography (CAG) images has been increasingly recognized in the field of cardiovascular disease management. This process relies on the camera matrix's optimization, needing correspondence info for identical point positions across two images. Therefore, an automatic method for determining correspondence between two CAG images is highly desirable. Despite this need, there is a paucity of research focusing on image matching in the CAG images. Additionally, standard deep learning image matching techniques often degrade due to unique features and noise in CAG images. This study aims to fill this gap by applying a deep learning-based image matching method specifically tailored for the CAG images. We have improved the structure of our point detector and redesigned loss function to better handle sparse labeling and indistinct local features specific to CAG images. Our method include changes to training loss and introduction of a multi-head descriptor structure leading to an approximate 6% improvement. We anticipate that our work will provide valuable insights into adapting techniques from general domains to more specialized ones like medical imaging and serve as an improved benchmark for future endeavors in X-ray image-based correspondence matching.

PMID:39128936 | DOI:10.1038/s41598-024-69153-5

Eurasian J Med. 2023 Dec;55(1):114-119. doi: 10.5152/eurasianjmed.2023.23360.

ABSTRACT

This review explores the transformative impact of artificial intelligence (AI) in breast imaging, driven by a global rise in breast cancer cases. Propelled by deep learning techniques, AI shows promise in refining diagnostic processes, yet adoption rates vary. Its ability to manage extensive datasets and process multidimensional information holds potential for advancing precision medicine in breast cancer research. However, integration faces challenges, from data-related obstacles to ensuring transparency and trust in decision-making. Legal considerations, including the formation of AI teams and intellectual property protection, influence health care's adoption of AI. Ethical dimensions underscore the need for responsible AI implementation, emphasizing autonomy, well-being, safety, transparency, and accessibility. Establishing a robust legal and ethical framework is crucial for conscientiously deploying AI, ensuring positive impacts on patient safety and treatment efcacy. As nations and organizations aspire to engage in global competition, not merely as consumers, the review highlights the critical importance of developing legal regulations. A comprehensive approach, from AI team formation to end-user processes, is essential for navigating the complex terrain of AI applications in breast imaging. Legal experts play a key role in ensuring compliance, managing risks, and fostering resilient integration. The ultimate goal is a harmonious synergy between technological advancements and ethical considerations, ushering in enhanced breast cancer diagnostics through responsible AI utilization.

PMID:39128072 | DOI:10.5152/eurasianjmed.2023.23360

Orv Hetil. 2024 Aug 11;165(32):1242-1251. doi: 10.1556/650.2024.33098. Print 2024 Aug 11.

ABSTRACT

Bevezetés: A ’cone-beam’ (kúpsugaras) számítógépes tomográfiás (CBCT) felvételek szegmentációja során a síkbeli képekből álló adatokat három dimenzióban (3D) rekonstruáljuk. A szájsebészetben és a parodontológiában a digitális adatfeldolgozás lehetővé teszi a műtéti beavatkozások 3D tervezését. A leggyakrabban alkalmazott határérték-alapú szegmentáció gyors, de pontatlan, míg a félautomatikus módszerek megfelelő pontosságúak, de rendkívül időigényesek. Az utóbbi években a mesterséges intelligencián alapuló technológiák elterjedésével azonban mostanra lehetőség van a CBCT-felvételek automatikus szegmentációjára. Célkitűzés: A klinikai gyakorlatból vett CBCT-felvételeken betanított mélytanulási szegmentációs modell bemutatása és hatékonyságának vizsgálata. Módszer: A vizsgálat három fő fázisa volt: a tanuló adatbázis felállítása, a mélytanulási modell betanítása és ezen architektúra pontosságának tesztelése. A tanuló adatbázis felállításához 70, részlegesen fogatlan páciens CBCT-felvételeit alkalmaztuk. A SegResNet hálózati architektúrára épülő szegmentációs modellt a MONAI rendszer segítségével fejlesztettük ki. A mélytanulási modell pontosságának ellenőrzéséhez 15 CBCT-felvételt használtunk. Ezeket a felvételeket a mélytanulási modell segítségével, valamint félautomatikus szegmentációval is feldolgoztuk, és összehasonlítottuk a két szegmentáció eredményét. Eredmények: A mélytanulásos szegmentáció és a félautomatikus szegmentáció közötti hasonlóság a Jaccard-index szerint átlagosan 0,91 ± 0,02, a Dice hasonlósági együttható átlagos értéke 0,95 ± 0,01, míg a két modell közötti átlagos Hausdorff- (95%) távolság 0,67 mm ± 0,22 mm volt. A mélytanulásos architektúra által szegmentált és a félautomatikus szegmentációval létrehozott 3D modellek térfogata nem mutatott statisztikailag szignifikáns különbséget (p = 0,31). Megbeszélés: A vizsgálatunkban használt mélytanulási modell az irodalomban található mesterségesintelligencia-rendszerekhez hasonló pontossággal végezte el a CBCT-felvételek szegmentációját, és mivel a CBCT-felvételek a rutin klinikai gyakorlatból származtak, a mélytanulási modell relatíve nagy megbízhatósággal szegmentálta a parodontalis csonttopográfiát és az alveolaris gerincdefektusokat. Következtetés: A mélytanulási modell nagy pontossággal szegmentálta az alsó állcsontot dentális CBCT-felvételeken. Ezek alapján megállapítható, hogy a mélytanulásos szegmentációval előállított 3D modell alkalmas lehet rekonstruktív szájsebészeti és parodontalis sebészeti beavatkozások digitális tervezésére. Orv Hetil. 2024; 165(32): 1242–1251.

PMID:39127997 | DOI:10.1556/650.2024.33098

Ophthalmol Ther. 2024 Aug 11. doi: 10.1007/s40123-024-01009-7. Online ahead of print.

ABSTRACT

INTRODUCTION: The aim of this work is to develop a deep learning (DL) system for rapidly and accurately screening for intraocular tumor (IOT), retinal detachment (RD), vitreous hemorrhage (VH), and posterior scleral staphyloma (PSS) using ocular B-scan ultrasound images.

METHODS: Ultrasound images from five clinically confirmed categories, including vitreous hemorrhage, retinal detachment, intraocular tumor, posterior scleral staphyloma, and normal eyes, were used to develop and evaluate a fine-grained classification system (the Dual-Path Lesion Attention Network, DPLA-Net). Images were derived from five centers scanned by different sonographers and divided into training, validation, and test sets in a ratio of 7:1:2. Two senior ophthalmologists and four junior ophthalmologists were recruited to evaluate the system's performance.

RESULTS: This multi-center cross-sectional study was conducted in six hospitals in China. A total of 6054 ultrasound images were collected; 4758 images were used for the training and validation of the system, and 1296 images were used as a testing set. DPLA-Net achieved a mean accuracy of 0.943 in the testing set, and the area under the curve was 0.988 for IOT, 0.997 for RD, 0.994 for PSS, 0.988 for VH, and 0.993 for normal. With the help of DPLA-Net, the accuracy of the four junior ophthalmologists improved from 0.696 (95% confidence interval [CI] 0.684-0.707) to 0.919 (95% CI 0.912-0.926, p < 0.001), and the time used for classifying each image reduced from 16.84 ± 2.34 s to 10.09 ± 1.79 s.

CONCLUSIONS: The proposed DPLA-Net showed high accuracy for screening and classifying multiple ophthalmic diseases using B-scan ultrasound images across mutiple centers. Moreover, the system can promote the efficiency of classification by ophthalmologists.

PMID:39127983 | DOI:10.1007/s40123-024-01009-7

Sci Rep. 2024 Aug 10;14(1):18579. doi: 10.1038/s41598-024-69584-0.

ABSTRACT

The effective detection of safflower in the field is crucial for implementing automated visual navigation and harvesting systems. Due to the small physical size of safflower clusters, their dense spatial distribution, and the complexity of field scenes, current target detection technologies face several challenges in safflower detection, such as insufficient accuracy and high computational demands. Therefore, this paper introduces an improved safflower target detection model based on YOLOv5, termed Safflower-YOLO (SF-YOLO). This model employs Ghost_conv to replace traditional convolution blocks in the backbone network, significantly enhancing computational efficiency. Furthermore, the CBAM attention mechanism is integrated into the backbone network, and a combined L C I O U + N W D loss function is introduced to allow for more precise feature extraction, enhanced adaptive fusion capabilities, and accelerated loss convergence. Anchor boxes, updated through K-means clustering, are used to replace the original anchors, enabling the model to better adapt to the multi-scale information of safflowers in the field. Data augmentation techniques such as Gaussian blur, noise addition, sharpening, and channel shuffling are applied to the dataset to maintain robustness against variations in lighting, noise, and visual angles. Experimental results demonstrate that SF-YOLO surpasses the original YOLOv5s model, with reductions in GFlops and Params from 15.8 to 13.2 G and 7.013 to 5.34 M, respectively, representing decreases of 16.6% and 23.9%. Concurrently, SF-YOLO's mAP0.5 increases by 1.3%, reaching 95.3%. This work enhances the accuracy of safflower detection in complex agricultural environments, providing a reference for subsequent autonomous visual navigation and automated non-destructive harvesting technologies in safflower operations.

PMID:39127852 | DOI:10.1038/s41598-024-69584-0

Sci Rep. 2024 Aug 10;14(1):18599. doi: 10.1038/s41598-024-69462-9.

ABSTRACT

Fault detection and isolation in unmanned aerial vehicle (UAV) propellers are critical for operational safety and efficiency. Most existing fault diagnosis techniques rely basically on traditional statistical-based methods that necessitate better approaches. This study explores the application of untraditional feature extraction methodologies, namely Permutation Entropy (PE), Lempel-Ziv Complexity (LZC), and Teager-Kaiser Energy Operator (TKEO), on the PADRE dataset, which encapsulates various rotor fault configurations. The extracted features were subjected to a Chi-Square (χ2) feature selection process to identify the most significant features for input into a Deep Neural Network. The Taguchi method was utilized to test the performance of the recorded features, correspondingly. Performance metrics, including Accuracy, F1-Score, Precision, and Recall, were employed to evaluate the model's effectiveness before and after the feature selection. The achieved accuracy has increased by 0.9% when compared with results utilizing traditional statistical methods. Comparative analysis with prior research reveals that the proposed untraditional features surpass traditional methods in diagnosing UAV propeller faults. It resulted in improved performance metrics with Accuracy, F1-Score, Precision, and Recall reaching 99.6%, 99.5%, 99.5%, and 99.5%, respectively. The results suggest promising directions for future research in UAV maintenance and safety protocols.

PMID:39127843 | DOI:10.1038/s41598-024-69462-9

Nat Commun. 2024 Aug 10;15(1):6859. doi: 10.1038/s41467-024-50911-y.

ABSTRACT

Disruption of retinal vasculature is linked to various diseases, including diabetic retinopathy and macular degeneration, leading to vision loss. We present here a novel algorithmic approach that generates highly realistic digital models of human retinal blood vessels, based on established biophysical principles, including fully-connected arterial and venous trees with a single inlet and outlet. This approach, using physics-informed generative adversarial networks (PI-GAN), enables the segmentation and reconstruction of blood vessel networks with no human input and which out-performs human labelling. Segmentation of DRIVE and STARE retina photograph datasets provided near state-of-the-art vessel segmentation, with training on only a small (n = 100) simulated dataset. Our findings highlight the potential of PI-GAN for accurate retinal vasculature characterization, with implications for improving early disease detection, monitoring disease progression, and improving patient care.

PMID:39127778 | DOI:10.1038/s41467-024-50911-y

Sci Rep. 2024 Aug 10;14(1):18619. doi: 10.1038/s41598-024-69827-0.

ABSTRACT

Adenoid hypertrophy can lead to adenoidal mouth breathing, which can result in "adenoid face" and, in severe cases, can even lead to respiratory tract obstruction. The Fujioka ratio method, which calculates the ratio of adenoid (A) to nasopharyngeal (N) space in an adenoidal-cephalogram (A/N), is a well-recognized and effective technique for detecting adenoid hypertrophy. However, this process is time-consuming and relies on personal experience, so a fully automated and standardized method needs to be designed. Most of the current deep learning-based methods for automatic diagnosis of adenoids are CNN-based methods, which are more sensitive to features similar to adenoids in lateral views and can affect the final localization results. In this study, we designed a local attention-based method for automatic diagnosis of adenoids, which takes AdeBlock as the basic module, fuses the spatial and channel information of adenoids through two-branch local attention computation, and combines the downsampling method without losing spatial information. Our method achieved mean squared error (MSE) 0.0023, mean radial error (MRE) 1.91, and SD (standard deviation) 7.64 on the three hospital datasets, outperforming other comparative methods.

PMID:39127777 | DOI:10.1038/s41598-024-69827-0

Plant Methods. 2024 Aug 10;20(1):121. doi: 10.1186/s13007-024-01250-y.

ABSTRACT

BACKGROUND: Structural genomic variants (SVs) are prevalent in plant genomes and have played an important role in evolution and domestication, as they constitute a significant source of genomic and phenotypic variability. Nevertheless, most methods in quantitative genetics focusing on crop improvement, such as genomic prediction, consider only Single Nucleotide Polymorphisms (SNPs). Deep Learning (DL) is a promising strategy for genomic prediction, but its performance using SVs and SNPs as genetic markers remains unknown.

RESULTS: We used rice to investigate whether combining SVs and SNPs can result in better trait prediction over SNPs alone and examine the potential advantage of Deep Learning (DL) networks over Bayesian Linear models. Specifically, the performances of BayesC (considering additive effects) and a Bayesian Reproducible Kernel Hilbert space (RKHS) regression (considering both additive and non-additive effects) were compared to those of two different DL architectures, the Multilayer Perceptron, and the Convolution Neural Network, to explore their prediction ability by using various marker input strategies. We found that exploiting structural and nucleotide variation slightly improved prediction ability on complex traits in 87% of the cases. DL models outperformed Bayesian models in 75% of the studied cases, considering the four traits and the two validation strategies used. Finally, DL systematically improved prediction ability of binary traits against the Bayesian models.

CONCLUSIONS: Our study reveals that the use of structural genomic variants can improve trait prediction in rice, independently of the methodology used. Also, our results suggest that Deep Learning (DL) networks can perform better than Bayesian models in the prediction of binary traits, and in quantitative traits when the training and target sets are not closely related. This highlights the potential of DL to enhance crop improvement in specific scenarios and the importance to consider SVs in addition to SNPs in genomic selection.

PMID:39127715 | DOI:10.1186/s13007-024-01250-y