Biol Methods Protoc. 2025 Jan 6;10(1):bpae097. doi: 10.1093/biomethods/bpae097. eCollection 2025.

ABSTRACT

A mixture-of-experts (MoE) approach has been developed to mitigate the poor out-of-distribution (OOD) generalization of deep learning (DL) models for single-sequence-based prediction of RNA secondary structure. The main idea behind this approach is to use DL models for in-distribution (ID) test sequences to leverage their superior ID performances, while relying on physics-based models for OOD sequences to ensure robust predictions. One key ingredient of the pipeline, named MoEFold2D, is automated ID/OOD detection via consensus analysis of an ensemble of DL model predictions without requiring access to training data during inference. Specifically, motivated by the clustered distribution of known RNA structures, a collection of distinct DL models is trained by iteratively leaving one cluster out. Each DL model hence serves as an expert on all but one cluster in the training data. Consequently, for an ID sequence, all but one DL model makes accurate predictions consistent with one another, while an OOD sequence yields highly inconsistent predictions among all DL models. Through consensus analysis of DL predictions, test sequences are categorized as ID or OOD. ID sequences are subsequently predicted by averaging the DL models in consensus, and OOD sequences are predicted using physics-based models. Instead of remediating generalization gaps with alternative approaches such as transfer learning and sequence alignment, MoEFold2D circumvents unpredictable ID-OOD gaps and combines the strengths of DL and physics-based models to achieve accurate ID and robust OOD predictions.

PMID:39811444 | PMC:PMC11729747 | DOI:10.1093/biomethods/bpae097

Heliyon. 2024 Dec 16;11(1):e40855. doi: 10.1016/j.heliyon.2024.e40855. eCollection 2025 Jan 15.

ABSTRACT

INTRODUCTION: The COVID-19 pandemic has created an urgent demand for research, which has spurred the development of enhanced biosafety protocols in biosafety level (BSL)-3 laboratories to safeguard against the risks associated with handling highly contagious pathogens. Laboratory management failures can pose significant hazards.

METHODS: An external system captured images of personnel entering a laboratory, which were then analyzed by an AI-based system to verify their compliance with personal protective equipment (PPE) regulations, thereby introducing an additional layer of protection. A deep learning model was trained to detect the presence of essential PPE items, such as clothing, masks, hoods, double-layer gloves, shoe covers, and respirators, ensuring adherence to World Health Organization (WHO) standards. The internal laboratory management system used a deep learning model to delineate alert zones and monitor compliance with the imposed safety protocols.

RESULTS: The external detection system was trained on a dataset consisting of 4112 images divided into 15 PPE compliance classes. The model achieved an accuracy of 97.52 % and a recall of 97.03 %. The identification results were presented in real time via a visual interface and simultaneously stored on the administrator's dashboard for future reference. We trained the internal management system on 3347 images, achieving 90 % accuracy and 85 % recall. The results were transmitted in JSON format to the internal monitoring system, which triggered alerts in response to violations of safe practices or alert zones. Real-time notifications were sent to the administrators when the safety thresholds were met.

CONCLUSION: The BSL-3 laboratory monitoring system significantly reduces the risk of exposure to pathogens for personnel during laboratory operations. By ensuring the correct use of PPE and enhancing adherence to the imposed safety protocols, this system contributes to maintaining the integrity of BSL-3 facilities and mitigates the risk of personnel becoming infection vectors.

PMID:39811271 | PMC:PMC11730239 | DOI:10.1016/j.heliyon.2024.e40855

Ophthalmol Sci. 2024 Nov 12;5(2):100653. doi: 10.1016/j.xops.2024.100653. eCollection 2025 Mar-Apr.

ABSTRACT

PURPOSE: The diagnosis of fungal keratitis using potassium hydroxide (KOH) smears of corneal scrapings enables initiation of the correct antimicrobial therapy at the point-of-care but requires time-consuming manual examination and expertise. This study evaluates the efficacy of a deep learning framework, dual stream multiple instance learning (DSMIL), in automating the analysis of whole slide imaging (WSI) of KOH smears for rapid and accurate detection of fungal infections.

DESIGN: Retrospective observational study.

PARTICIPANTS: Corneal scrapings from 568 patients with suspected fungal keratitis; 51% contained filamentous fungi according to human expert interpretation.

METHODS: Dual stream multiple instance learning was employed to analyze WSI of KOH smears. Due to the extensive size of these images, often exceeding 100 000 pixels, conventional computer vision methods (e.g., convolutional neural networks) are not feasible. Dual stream multiple instance learning segments the WSI into patches for analysis, extracting relevant features from each patch and aggregating these to make a comprehensive slide-level diagnosis while generating heat maps to visualize areas contributing most to the prediction. Fivefold cross-validation was used for training and validation, with a hold-out test set comprising 15% of the total samples.

MAIN OUTCOME MEASURES: Accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC), F1 score, positive predictive value (PPV), and negative predictive value (NPV) in distinguishing fungal from nonfungal slides.

RESULTS: Dual stream multiple instance learning demonstrated an overall AUC of 0.88 with an accuracy of 79% and an F1 score of 0.79 in distinguishing fungal from nonfungal slides, with sensitivity of 85%, specificity of 71%, PPV of 80%, and NPV of 79%. For "consensus cases," where 2 human graders agreed on the slide interpretation, the model achieved an accuracy of 85% and an F1 score of 0.85. For "discrepant cases," the accuracy was 71% with an F1 score of 0.71. The generated heatmaps highlighted regions corresponding to fungal elements. Code and models are open-sourced and available at https://github.com/Redd-Cornea-AI/KOH-Smear-DSMIL.

CONCLUSIONS: The DSMIL framework shows significant promise in automating interpretation of KOH smears. Its capability to handle large, high-resolution WSI data and accurately detect fungal infections, while providing visual explanations through heatmaps, could enhance the scalability of KOH smear interpretation, ultimately reducing the global burden of blindness from infectious keratitis.

FINANCIAL DISCLOSURES: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

PMID:39811263 | PMC:PMC11731208 | DOI:10.1016/j.xops.2024.100653

Cancer Cell Int. 2025 Jan 14;25(1):13. doi: 10.1186/s12935-025-03642-z.

ABSTRACT

BACKGROUND: Patients with lung adenocarcinoma (LUAD) receiving drug treatment often have an unpredictive response and there is a lack of effective methods to predict treatment outcome for patients. Dendritic cells (DCs) play a significant role in the tumor microenvironment and the DCs-related gene signature may be used to predict treatment outcome. Here, we screened for DC-related genes to construct a prognostic signature to predict prognosis and response to immunotherapy in LUAD patients.

METHODS: DC-related biological functions and genes were identified using single-cell RNA sequencing (scRNA-seq) and bulk RNA sequencing. DCs-related gene signature (DCRGS) was constructed using integrated machine learning algorithms. Expression of key genes in clinical samples was examined by real-time q-PCR. Performance of the prognostic model, DCRGS, for the prognostic evaluation, was assessed using a multiple time-dependent receiver operating characteristic (ROC) curve, the R package, "timeROC", and validated using GEO datasets.

RESULTS: Analysis of scRNA-seq data showed that there is a significant upregulation of LGALS9 expression in DCs isolated from malignant pleural effusion samples. Leveraging the Coxboost and random survival forest combination algorithm, we filtered out six DC-related genes on which a prognostic prediction model, DCRGS, was established. A high predictive capability nomogram was constructed by combining DCRGS with clinical features. We found that patients with a high-DCRGS score had immunosuppression, activated tumor-associated pathways, and elevated somatic mutational load and copy number variant load. In contrast, patients in the low-DCRGS subgroup were resistant to chemotherapy but sensitive to the CTLA-4 immune checkpoint inhibitor and targeted therapy.

CONCLUSION: We have innovatively established a deep learning-based prediction model, DCRGS, for the prediction of the prognosis of patients with LUAD. The model possesses a strong prognostic prediction performance with high accuracy and sensitivity and could be clinically useful to guide the management of LUAD. Furthermore, the findings of this study could provide an important reference for individualized clinical treatment and prognostic prediction of patients with LUAD.

PMID:39810206 | DOI:10.1186/s12935-025-03642-z

J Transl Med. 2025 Jan 14;23(1):65. doi: 10.1186/s12967-024-06044-3.

ABSTRACT

BACKGROUND: Parkinson's Disease (PD) is a neurodegenerative disorder, and eye movement abnormalities are a significant symptom of its diagnosis. In this paper, we developed a multi-task driven by eye movement in a virtual reality (VR) environment to elicit PD-specific eye movement abnormalities. The abnormal features were subsequently modeled by using the proposed deep learning algorithm to achieve an auxiliary diagnosis of PD.

METHODS: We recruited 114 PD patients and 125 healthy controls and collected their eye-tracking data in a VR environment. Participants completed a series of specific VR tasks, including gaze stability, pro-saccades, anti-saccades, and smooth pursuit. After the tasks, eye movement features were extracted from the behaviors of fixations, saccades, and smooth pursuit to establish a PD diagnostic model.

RESULTS: The performance of the models was evaluated through cross-validation, revealing a recall of 97.65%, an accuracy of 92.73%, and a receiver operator characteristic area under the curve (ROC-AUC) of 97.08% for the proposed model.

CONCLUSION: We extracted PD-specific eye movement features from the behaviors of fixations, saccades, and smooth pursuit in a VR environment to create a model with high accuracy and recall for PD diagnosis. Our method provides physicians with a new auxiliary tool to improve the prognosis and quality of life of PD patients.

PMID:39810187 | DOI:10.1186/s12967-024-06044-3

BMC Med Educ. 2025 Jan 14;25(1):66. doi: 10.1186/s12909-025-06648-3.

ABSTRACT

BACKGROUND: Reflection fosters self-regulated learning by enabling learners to critically evaluate their performance, identify gaps, and make plans to improve. Feedback, in turn, provides external insights that complement reflection, helping learners recognize their strengths and weaknesses, adjust their learning strategies, and enhance clinical reasoning and decision-making skills. However, reflection alone may not produce the desirable effects unless coupled with feedback. This study aimed to investigate the impact of feedback integrated reflection on meaningful learning and higher order MCQ score among under-grade medical students.

OBJECTIVE: To evaluate the impact of feedback-integrated reflection versus reflection alone on higher-order MCQ scores among undergraduate medical students in a gynecology clinical setting.

METHODS: A randomized controlled trial was conducted with 68 final-year medical students randomly assigned to a study group (feedback-integrated reflection) and a control group (reflection alone). Both groups completed a pre-test, followed by six daily teaching sessions on gynecology topics. Participants engaged in written reflections after each session, and the study group additionally received individualized feedback. Independent sample t-tests were used to compare pre and post-test scores between the groups, while paired t-tests assessed within-group improvements.

RESULTS: Pre-test scores were comparable between the study group (11.68 ± 2.60, 38.93%) and the control group (11.29 ± 2.38, 37.15%; P = 0.52). Post-test scores showed a significant improvement in the study group (20.88 ± 2.98, 69.32%) compared to the control group (15.29 ± 2.66, 51.00%; P = 0.0001). The percentage gain in learning was 35.43% for the control group (reflection alone) and 78.77% for the study group (feedback-integrated reflection). The normalized learning gain (NLG) was calculated to compare the effectiveness of the intervention (feedback-integrated reflection) with that of the control (reflection alone). The study group demonstrated a mean normalized learning gain of 69.07%, compared to 29.18% in the control group. The net learning gain, calculated as the difference in normalized learning gains between the study and control groups, was found to be 39.89%.

CONCLUSION: The findings highlight the effectiveness of feedback-integrated reflection versus reflection alone in fostering deeper learning by improving higher-order MCQ scores in a gynecologic setting in the undergraduate medical education.

TRIAL REGISTRATION: This trial was registered retrospectively on 27th July 2024. Trial registration no is CTU/07/2024/010/RMU.

PMID:39810114 | DOI:10.1186/s12909-025-06648-3

BMC Genom Data. 2025 Jan 14;26(1):4. doi: 10.1186/s12863-024-01293-z.

ABSTRACT

BACKGROUND: miRNAs (microRNAs) are endogenous RNAs with lengths of 18 to 24 nucleotides and play critical roles in gene regulation and disease progression. Although traditional wet-lab experiments provide direct evidence for miRNA-disease associations, they are often time-consuming and complicated to analyze by current bioinformatics tools. In recent years, machine learning (ML) and deep learning (DL) techniques are powerful tools to analyze large-scale biological data. Hence, developing a model to predict, identify, and rank connections in miRNAs and diseases can significantly enhance the precision and efficiency in investigating the relationships between miRNAs and diseases.

RESULTS: In this study, we utilized miRNA-disease association data obtained by biotechnological experiments to develop a DL model for miRNA-disease associations. To improve the accuracy of prediction in this model, we introduced two labeling strategies, weight-based and majority-based definitions, to classify miRNA-disease associations. After preprocessing, data was trained with a novel model combining gated recurrent units (GRU) and graph convolutional network (GCN) to predict the level of miRNA-disease associations. The miRNA-disease association datasets were from HMDD (the Human miRNA Disease Database) and categorized by two distinct labeling approaches, weight-based definitions and majority-based definitions. We classified the miRNA-disease associations into three groups, "upregulated", "downregulated" and "nonspecific", by regression analysis and multiclass classification. This GRU-GCN coordinated model achieved a robust area under the curve (AUC) score of 0.8 in all datasets, demonstrating the efficacy in predicting potential miRNA-disease relationships.

CONCLUSIONS: By introducing innovative label-preprocessing methods, this study addressed the relationships between miRNAs and diseases, and improved the ambiguity of the results in different experiments. Based on these refined label definitions, we developed a DL-based model to refine and predict the results of associations between miRNAs and diseases. This model offers a valuable tool for complementing traditional experimental methods and enhancing our understanding of miRNA-related disease mechanisms.

PMID:39810100 | DOI:10.1186/s12863-024-01293-z

J Imaging Inform Med. 2025 Jan 14. doi: 10.1007/s10278-024-01375-5. Online ahead of print.

ABSTRACT

Optical coherence tomography angiography (OCTA) is an emerging, non-invasive technique increasingly utilized for retinal vasculature imaging. Analysis of OCTA images can effectively diagnose retinal diseases, unfortunately, complex vascular structures within OCTA images possess significant challenges for automated segmentation. A novel, fully convolutional dense connected residual network is proposed to effectively segment the vascular regions within OCTA images. Firstly, a dual-branch structure Recurrent Residual Convolutional Neural Network (RRCNN) block is constructed utilizing RecurrentBlock and convolutional operations. Subsequently, the ResConvNeXt V2 Block is built as the backbone structure of the network. The output from the ResConvNeXt V2 Block is then fed into the side branch and the next ResConvNeXt V2 Block. Within the side branch, the Group Receptive Field Block (GRFB) processes the results from the previous and current layers. Ultimately, the side branch results are added to the backbone network outputs to produce the final segmentation. The model achieves superior performance. Experiments were conducted on the ROSSA and OCTA-500 datasets, yielding Dice scores of 91.88%, 91.72%, and 89.18% for the respective datasets, and accuracies of 98.31%, 99.02%, and 98.02%.

PMID:39810043 | DOI:10.1007/s10278-024-01375-5

Psychon Bull Rev. 2025 Jan 14. doi: 10.3758/s13423-024-02616-y. Online ahead of print.

ABSTRACT

We examined the intricate mechanisms underlying visual processing of complex motion stimuli by measuring the detection sensitivity to contraction and expansion patterns and the discrimination sensitivity to the location of the center of motion (CoM) in various real and unreal optic flow stimuli. We conducted two experiments (N = 20 each) and compared responses to both "real" optic flow stimuli containing information about self-movement in a three-dimensional scene and "unreal" optic flow stimuli lacking such information. We found that detection sensitivity to contraction surpassed that to expansion patterns for unreal optic flow stimuli, whereas this trend was reversed for real optic flow stimuli. Furthermore, while discrimination sensitivity to the CoM location was not affected by stimulus duration for unreal optic flow stimuli, it showed a significant improvement when stimulus duration increased from 100 to 400 ms for real optic flow stimuli. These findings provide compelling evidence that the visual system employs distinct processing approaches for real versus unreal optic flow even when they are perfectly matched for two-dimensional global features and local motion signals. These differences reveal influences of self-movement in natural environments, enabling the visual system to uniquely process stimuli with significant survival implications.

PMID:39810018 | DOI:10.3758/s13423-024-02616-y

Sci Rep. 2025 Jan 14;15(1):1962. doi: 10.1038/s41598-025-86455-4.

ABSTRACT

The NLRP3 inflammasome, regulated by TLR4, plays a pivotal role in periodontitis by mediating inflammatory cytokine release and bone loss induced by Porphyromonas gingivalis. Periodontal disease creates a hypoxic environment, favoring anaerobic bacteria survival and exacerbating inflammation. The NLRP3 inflammasome triggers pyroptosis, a programmed cell death that amplifies inflammation and tissue damage. This study evaluates the efficacy of Variational Graph Autoencoders (VGAEs) in reconstructing gene data related to NLRP3-mediated pyroptosis in periodontitis. The NCBI GEO dataset GSE262663, containing three samples with and without hypoxia exposure, was analyzed using unsupervised K-means clustering. This method identifies natural groupings within biological data without prior labels. VGAE, a deep learning model, captures complex graph relationships for tasks like link prediction and edge detection. The VGAE model demonstrated exceptional performance with an accuracy of 99.42% and perfect precision. While it identified 5,820 false negatives, indicating a conservative approach, it accurately predicted 4,080 out of 9,900 positive samples. The model's latent space distribution differed significantly from the original data, suggesting a tightly clustered representation of the gene expression patterns. K-means clustering and VGAE show promise in gene expression analysis and graph structure reconstruction for periodontitis research.

PMID:39809940 | DOI:10.1038/s41598-025-86455-4

Nat Biotechnol. 2025 Jan 14. doi: 10.1038/s41587-024-02528-1. Online ahead of print.

ABSTRACT

Efficient and accurate nanocarrier development for targeted drug delivery is hindered by a lack of methods to analyze its cell-level biodistribution across whole organisms. Here we present Single Cell Precision Nanocarrier Identification (SCP-Nano), an integrated experimental and deep learning pipeline to comprehensively quantify the targeting of nanocarriers throughout the whole mouse body at single-cell resolution. SCP-Nano reveals the tissue distribution patterns of lipid nanoparticles (LNPs) after different injection routes at doses as low as 0.0005 mg kg-1-far below the detection limits of conventional whole body imaging techniques. We demonstrate that intramuscularly injected LNPs carrying SARS-CoV-2 spike mRNA reach heart tissue, leading to proteome changes, suggesting immune activation and blood vessel damage. SCP-Nano generalizes to various types of nanocarriers, including liposomes, polyplexes, DNA origami and adeno-associated viruses (AAVs), revealing that an AAV2 variant transduces adipocytes throughout the body. SCP-Nano enables comprehensive three-dimensional mapping of nanocarrier distribution throughout mouse bodies with high sensitivity and should accelerate the development of precise and safe nanocarrier-based therapeutics.

PMID:39809933 | DOI:10.1038/s41587-024-02528-1

Sci Rep. 2025 Jan 14;15(1):1924. doi: 10.1038/s41598-024-84869-0.

ABSTRACT

To address the challenges of unbalanced class labels with varying maturity levels of tomato fruits and low recognition accuracy for both fruits and stems in intelligent harvesting, we propose the YOLOX-SE-GIoU model for identifying tomato fruit maturity and stems. The SE focus module was incorporated into YOLOX to improve the identification accuracy, addressing the imbalance in the number of tomato fruits and stems. Additionally, we optimized the loss function to GIoU loss to minimize discrepancies across different scales of fruits and stems. The mean average precision (mAP) of the improved YOLOX-SE-GIoU model reaches 92.17%. Compared to YOLOv4, YOLOv5, YOLOv7, and YOLOX models, the improved model shows an improvement of 1.17-22.21%. The average precision (AP) for unbalanced semi-ripe tomatoes increased by 1.68-26.66%, while the AP for stems increased by 3.78-45.03%. Experimental results demonstrate that the YOLOX-SE-GIoU model exhibits superior overall recognition performance for unbalanced and scale-variant samples compared to the original model and other models in the same series. It effectively reduces false and missed detections during tomato harvesting, improving the identification accuracy of tomato fruits and stems. The findings of this work provide a technical foundation for developing advanced fruit harvesting techniques.

PMID:39809915 | DOI:10.1038/s41598-024-84869-0

Sci Rep. 2025 Jan 14;15(1):1926. doi: 10.1038/s41598-024-81244-x.

ABSTRACT

The rapid expansion of the coal mining industry has introduced significant safety risks, particularly within the harsh environments of open-pit coal mines. The safe and stable operation of belt conveyor idlers is crucial not only for ensuring efficient coal production but also for safeguarding the lives of coal mine workers. Therefore, this paper proposes a method based on deep learning for real-time detection of conveyor idler faults. The selected YOLOv5 network is analyzed and improved based on the training results. First, the coordinate attention mechanism is integrated into the model to reassign the weights across different channels. Subsequently, the α-CIoU localization loss function replaces the traditional CIoU to enhance the model's regression accuracy. Experimental results demonstrate that the enhanced YOLOv5 algorithm achieves a 95.3% mAP on the self-constructed infrared image dataset, surpassing the original algorithm by 2.7%. Moreover, with a processing speed of 285 FPS, it accurately performs the defect detection of conveyor idlers while satisfying real-time operational requirements.

PMID:39809903 | DOI:10.1038/s41598-024-81244-x

PLoS One. 2025 Jan 14;20(1):e0316548. doi: 10.1371/journal.pone.0316548. eCollection 2025.

ABSTRACT

This study introduces a high-resolution wind nowcasting model designed for aviation applications at Madeira International Airport, a location known for its complex wind patterns. By using data from a network of six meteorological stations and deep learning techniques, the produced model is capable of predicting wind speed and direction up to 30-minute ahead with 1-minute temporal resolution. The optimized architecture demonstrated robust predictive performance across all forecast horizons. For the most challenging task, the 30-minute ahead forecasts, the model achieved a wind speed Mean Absolute Error (MAE) of 0.78 m/s and a wind direction MAE of 33.06°. Furthermore, the use of Gaussian noise concatenation to both input and label training data yielded the most consistent results. A case study further validated the model's efficacy, with MAE values below 0.43 m/s for wind speed and between 33.93° and 35.03° for wind direction across different forecast horizons. This approach shows that combining strategically deployed sensor networks with machine learning techniques offers improvements in wind nowcasting for airports in complex environments, possibly enhancing operational efficiency and safety.

PMID:39808682 | DOI:10.1371/journal.pone.0316548

PLoS One. 2025 Jan 14;20(1):e0317585. doi: 10.1371/journal.pone.0317585. eCollection 2025.

ABSTRACT

Urban waterfront areas, which are essential natural resources and highly perceived public areas in cities, play a crucial role in enhancing urban environment. This study integrates deep learning with human perception data sourced from street view images to study the relationship between visual landscape features and human perception of urban waterfront areas, employing linear regression and random forest models to predict human perception along urban coastal roads. Based on aesthetic and distinctiveness perception, urban coastal roads in Xiamen were classified into four types with different emphasis and priorities for improvement. The results showed that: 1) the degree of coastal openness had the greatest influence on human perception while the coastal landscape with a high green visual index decreases the distinctiveness perception; 2) the random forest model can effectively predict human perception on urban coastal roads with an accuracy rate of 87% and 77%; 3) The proportion of low perception road sections with potential for improvement is 60.6%, among which the proportion of low aesthetic perception and low distinctiveness perception road sections is 10.5%. These findings offer crucial evidence regarding human perception of urban coastal roads, and can provide targeted recommendations for enhancing the visual environment of urban coastal road landscapes.

PMID:39808675 | DOI:10.1371/journal.pone.0317585

J Thorac Imaging. 2024 Sep 20. doi: 10.1097/RTI.0000000000000808. Online ahead of print.

ABSTRACT

PURPOSE: Radiological follow-up of oncology patients requires the detection of metastatic lung lesions and the quantitative analysis of their changes in longitudinal imaging studies. Our aim was to evaluate SimU-Net, a novel deep learning method for the automatic analysis of metastatic lung lesions and their temporal changes in pairs of chest CT scans.

MATERIALS AND METHODS: SimU-Net is a simultaneous multichannel 3D U-Net model trained on pairs of registered prior and current scans of a patient. It is part of a fully automatic pipeline for the detection, segmentation, matching, and classification of metastatic lung lesions in longitudinal chest CT scans. A data set of 5040 metastatic lung lesions in 344 pairs of 208 prior and current chest CT scans from 79 patients was used for training/validation (173 scans, 65 patients) and testing (35 scans, 14 patients) of a standalone 3D U-Net models and 3 simultaneous SimU-Net models. Outcome measures were the lesion detection and segmentation precision, recall, Dice score, average symmetric surface distance (ASSD), lesion matching, and classification of lesion changes from computed versus manual ground-truth annotations by an expert radiologist.

RESULTS: SimU-Net achieved a mean lesion detection recall and precision of 0.93±0.13 and 0.79±0.24 and a mean lesion segmentation Dice and ASSD of 0.84±0.09 and 0.33±0.22 mm. These results outperformed the standalone 3D U-Net model by 9.4% in the recall, 2.4% in Dice, and 15.4% in ASSD, with a minor 3.6% decrease in precision. The SimU-Net pipeline achieved perfect precision and recall (1.0±0.0) for lesion matching and classification of lesion changes.

CONCLUSIONS: Simultaneous deep learning analysis of metastatic lung lesions in prior and current chest CT scans with SimU-Net yields superior accuracy compared with individual analysis of each scan. Implementation of SimU-Net in the radiological workflow may enhance efficiency by automatically computing key metrics used to evaluate metastatic lung lesions and their temporal changes.

PMID:39808543 | DOI:10.1097/RTI.0000000000000808

J Neuroophthalmol. 2024 Dec 1;44(4):462-468. doi: 10.1097/WNO.0000000000002229. Epub 2024 Aug 1.

ABSTRACT

BACKGROUND: Optic neuritis (ON) is a complex clinical syndrome that has diverse etiologies and treatments based on its subtypes. Notably, ON associated with multiple sclerosis (MS ON) has a good prognosis for recovery irrespective of treatment, whereas ON associated with other conditions including neuromyelitis optica spectrum disorders or myelin oligodendrocyte glycoprotein antibody-associated disease is often associated with less favorable outcomes. Delay in treatment of these non-MS ON subtypes can lead to irreversible vision loss. It is important to distinguish MS ON from other ON subtypes early, to guide appropriate management. Yet, identifying ON and differentiating subtypes can be challenging as MRI and serological antibody test results are not always readily available in the acute setting. The purpose of this study is to develop a deep learning artificial intelligence (AI) algorithm to predict subtype based on fundus photographs, to aid the diagnostic evaluation of patients with suspected ON.

METHODS: This was a retrospective study of patients with ON seen at our institution between 2007 and 2022. Fundus photographs (1,599) were retrospectively collected from a total of 321 patients classified into 2 groups: MS ON (262 patients; 1,114 photographs) and non-MS ON (59 patients; 485 photographs). The dataset was divided into training and holdout test sets with an 80%/20% ratio, using stratified sampling to ensure equal representation of MS ON and non-MS ON patients in both sets. Model hyperparameters were tuned using 5-fold cross-validation on the training dataset. The overall performance and generalizability of the model was subsequently evaluated on the holdout test set.

RESULTS: The receiver operating characteristic (ROC) curve for the developed model, evaluated on the holdout test dataset, yielded an area under the ROC curve of 0.83 (95% confidence interval [CI], 0.72-0.92). The model attained an accuracy of 76.2% (95% CI, 68.4-83.1), a sensitivity of 74.2% (95% CI, 55.9-87.4) and a specificity of 76.9% (95% CI, 67.6-85.0) in classifying images as non-MS-related ON.

CONCLUSIONS: This study provides preliminary evidence supporting a role for AI in differentiating non-MS ON subtypes from MS ON. Future work will aim to increase the size of the dataset and explore the role of combining clinical and paraclinical measures to refine deep learning models over time.

PMID:39808513 | DOI:10.1097/WNO.0000000000002229

Insights Imaging. 2025 Jan 14;16(1):17. doi: 10.1186/s13244-025-01898-7.

ABSTRACT

OBJECTIVES: To evaluate the performance of a 3D V-Net-based segmentation model of adrenal lesions in characterizing adrenal glands as normal or abnormal.

METHODS: A total of 1086 CT image series with focal adrenal lesions were retrospectively collected, annotated, and used for the training of the adrenal lesion segmentation model. The dice similarity coefficient (DSC) of the test set was used to evaluate the segmentation performance. The other cohort, consisting of 959 patients with pathologically confirmed adrenal lesions (external validation dataset 1), was included for validation of the classification performance of this model. Then, another consecutive cohort of patients with a history of malignancy (N = 479) was used for validation in the screening population (external validation dataset 2). Parameters of sensitivity, accuracy, etc., were used, and the performance of the model was compared to the radiology report in these validation scenes.

RESULTS: The DSC of the test set of the segmentation model was 0.900 (0.810-0.965) (median (interquartile range)). The model showed sensitivities and accuracies of 99.7%, 98.3% and 87.2%, 62.2% in external validation datasets 1 and 2, respectively. It showed no significant difference comparing to radiology reports in external validation datasets 1 and lesion-containing groups of external validation datasets 2 (p = 1.000 and p > 0.05, respectively).

CONCLUSION: The 3D V-Net-based segmentation model of adrenal lesions can be used for the binary classification of adrenal glands.

CRITICAL RELEVANCE STATEMENT: A 3D V-Net-based segmentation model of adrenal lesions can be used for the detection of abnormalities of adrenal glands, with a high accuracy in the pre-surgical scene as well as a high sensitivity in the screening scene.

KEY POINTS: Adrenal lesions may be prone to inter-observer variability in routine diagnostic workflow. The study developed a 3D V-Net-based segmentation model of adrenal lesions with DSC 0.900 in the test set. The model showed high sensitivity and accuracy of abnormalities detection in different scenes.

PMID:39808346 | DOI:10.1186/s13244-025-01898-7

Brief Bioinform. 2024 Nov 22;26(1):bbaf001. doi: 10.1093/bib/bbaf001.

ABSTRACT

This study addresses the challenging task of identifying viruses within metagenomic data, which encompasses a broad array of biological samples, including animal reservoirs, environmental sources, and the human body. Traditional methods for virus identification often face limitations due to the diversity and rapid evolution of viral genomes. In response, recent efforts have focused on leveraging artificial intelligence (AI) techniques to enhance accuracy and efficiency in virus detection. However, existing AI-based approaches are primarily binary classifiers, lacking specificity in identifying viral types and reliant on nucleotide sequences. To address these limitations, VirDetect-AI, a novel tool specifically designed for the identification of eukaryotic viruses within metagenomic datasets, is introduced. The VirDetect-AI model employs a combination of convolutional neural networks and residual neural networks to effectively extract hierarchical features and detailed patterns from complex amino acid genomic data. The results demonstrated that the model has outstanding results in all metrics, with a sensitivity of 0.97, a precision of 0.98, and an F1-score of 0.98. VirDetect-AI improves our comprehension of viral ecology and can accurately classify metagenomic sequences into 980 viral protein classes, hence enabling the identification of new viruses. These classes encompass an extensive array of viral genera and families, as well as protein functions and hosts.

PMID:39808116 | DOI:10.1093/bib/bbaf001

Radiology. 2025 Jan;314(1):e240238. doi: 10.1148/radiol.240238.

ABSTRACT

Background Multiparametric MRI, including contrast-enhanced sequences, is recommended for evaluating suspected prostate cancer, but concerns have been raised regarding potential contrast agent accumulation and toxicity. Purpose To evaluate the feasibility of generating simulated contrast-enhanced MRI from noncontrast MRI sequences using deep learning and to explore their potential value for assessing clinically significant prostate cancer using Prostate Imaging Reporting and Data System (PI-RADS) version 2.1. Materials and Methods Male patients with suspected prostate cancer who underwent multiparametric MRI were retrospectively included from three centers from April 2020 to April 2023. A deep learning model (pix2pix algorithm) was trained to synthesize contrast-enhanced MRI scans from four noncontrast MRI sequences (T1-weighted imaging, T2-weighted imaging, diffusion-weighted imaging, and apparent diffusion coefficient maps) and then tested on an internal and two external datasets. The reference standard for model training was the second postcontrast phase of the dynamic contrast-enhanced sequence. Similarity between simulated and acquired contrast-enhanced images was evaluated using the multiscale structural similarity index. Three radiologists independently scored T2-weighted and diffusion-weighted MRI with either simulated or acquired contrast-enhanced images using PI-RADS, version 2.1; agreement was assessed with Cohen κ. Results A total of 567 male patients (mean age, 66 years ± 11 [SD]) were divided into a training test set (n = 244), internal test set (n = 104), external test set 1 (n = 143), and external test set 2 (n = 76). Simulated and acquired contrast-enhanced images demonstrated high similarity (multiscale structural similarity index: 0.82, 0.71, and 0.69 for internal test set, external test set 1, and external test set 2, respectively) with excellent reader agreement of PI-RADS scores (Cohen κ, 0.96; 95% CI: 0.94, 0.98). When simulated contrast-enhanced imaging was added to biparametric MRI, 34 of 323 (10.5%) patients were upgraded to PI-RADS 4 from PI-RADS 3. Conclusion It was feasible to generate simulated contrast-enhanced prostate MRI using deep learning. The simulated and acquired contrast-enhanced MRI scans exhibited high similarity and demonstrated excellent agreement in assessing clinically significant prostate cancer based on PI-RADS, version 2.1. © RSNA, 2025 Supplemental material is available for this article. See also the editorial by Neji and Goh in this issue.

PMID:39807983 | DOI:10.1148/radiol.240238