bioRxiv [Preprint]. 2025 Jan 3:2024.11.17.624050. doi: 10.1101/2024.11.17.624050.

ABSTRACT

Variant calling using long-read RNA sequencing (lrRNA-seq) can be applied to diverse tasks, such as capturing full-length isoforms and gene expression profiling. It poses challenges, however, due to higher error rates than DNA data, the complexities of transcript diversity, RNA editing events, etc. In this paper, we propose Clair3-RNA, the first deep learning-based variant caller tailored for lrRNA-seq data. Clair3-RNA leverages the strengths of the Clair series pipelines and incorporates several techniques optimized for lrRNA-seq data, such as uneven coverage normalization, refinement of training materials, editing site discovery, and the incorporation of phasing haplotype to enhance variant-calling performance. Clair3-RNA is available for various platforms, including PacBio and ONT complementary DNA sequencing (cDNA), and ONT direct RNA sequencing (dRNA). Our results demonstrated that Clair3-RNA achieved a ~91% SNP F1-score on the ONT platform using the latest ONT SQK-RNA004 kit (dRNA004) and a ~92% SNP F1-score in PacBio Iso-Seq and MAS-Seq for variants supported by at least four reads. The performance reached a ~95% and ~96% F1-score for ONT and PacBio, respectively, with at least ten supporting reads and disregarding the zygosity. With read phased, the performance reached ~97% for ONT and ~98% for PacBio. Extensive evaluation of various GIAB samples demonstrated that Clair3-RNA consistently outperformed existing callers and is capable of distinguishing RNA high-quality editing sites from variants accurately. Clair3-RNA is open-source and available at (https://github.com/HKU-BAL/Clair3-RNA).

PMID:39803537 | PMC:PMC11722298 | DOI:10.1101/2024.11.17.624050

bioRxiv [Preprint]. 2024 Oct 15:2024.10.11.617857. doi: 10.1101/2024.10.11.617857.

ABSTRACT

AlphaFold2 (AF2), a deep-learning based model that predicts protein structures from their amino acid sequences, has recently been used to predict multiple protein conformations. In some cases, AF2 has successfully predicted both dominant and alternative conformations of fold-switching proteins, which remodel their secondary and tertiary structures in response to cellular stimuli. Whether AF2 has learned enough protein folding principles to reliably predict alternative conformations outside of its training set is unclear. Here, we address this question by assessing whether CFold-an implementation of the AF2 network trained on a more limited subset of experimentally determined protein structures- predicts alternative conformations of eight fold switchers from six protein families. Previous work suggests that AF2 predicted these alternative conformations by memorizing them during training. Unlike AF2, CFold's training set contains only one of these alternative conformations. Despite sampling 1300-4400 structures/protein with various sequence sampling techniques, CFold predicted only one alternative structure outside of its training set accurately and with high confidence while also generating experimentally inconsistent structures with higher confidence. Though these results indicate that AF2's current success in predicting alternative conformations of fold switchers stems largely from its training data, results from a sequence pruning technique suggest developments that could lead to a more reliable generative model in the future.

PMID:39803493 | PMC:PMC11722258 | DOI:10.1101/2024.10.11.617857

bioRxiv [Preprint]. 2024 Dec 30:2024.12.30.630782. doi: 10.1101/2024.12.30.630782.

ABSTRACT

The use of microcomputed tomography (Micro-CT) for imaging biological samples has burgeoned in the past decade, due to increased access to scanning platforms, ease of operation, isotropic three-dimensional image information, and the ability to derive accurate quantitative data. However, manual data analysis of Micro-CT images can be laborious and time intensive. Deep learning offers the ability to streamline this process, but historically has included caveats-namely, the need for a large amount of training data, which is often limited in many Micro-CT studies. Here we show that accurate deep learning models can be trained using only 1-3 Micro-CT images of the adult Drosophila melanogaster brain using Dragonfly's pre-trained neural networks and minimal user knowledge. We further demonstrate the power of our model by showing that it can accurately segment the brain across different tissue contrast stains, scanner models, and genotypes. Finally, we show how the model can assist in identifying morphological similarities and differences between mutants based on volumetric quantification, facilitating a rapid assessment of novel phenotypes. Our models are freely available and can be refined based on individual user needs.

SUMMARY: Micro-CT data can be automatically segmented and quantified using a deep learning model trained on as few as 3 samples, facilitating rapid comparison of developmental phenotypes.

PMID:39803485 | PMC:PMC11722237 | DOI:10.1101/2024.12.30.630782

Kidney Med. 2024 Nov 12;7(1):100927. doi: 10.1016/j.xkme.2024.100927. eCollection 2025 Jan.

ABSTRACT

Artificial intelligence (AI) is increasingly used in many medical specialties. However, nephrology has lagged in adopting and incorporating machine learning techniques. Nephrology is well positioned to capitalize on the benefits of AI. The abundance of structured clinical data, combined with the mathematical nature of this specialty, makes it an attractive option for AI applications. AI can also play a significant role in addressing health inequities, especially in organ transplantation. It has also been used to detect rare diseases such as Fabry disease early. This review article aims to increase awareness on the basic concepts in machine learning and discuss AI applications in nephrology. It also addresses the challenges in integrating AI into clinical practice and the need for creating an AI-competent nephrology workforce. Even though AI will not replace nephrologists, those who are able to incorporate AI into their practice effectively will undoubtedly provide better care to their patients. The integration of AI technology is no longer just an option but a necessity for staying ahead in the field of nephrology. Finally, AI can contribute as a force multiplier in transitioning to a value-based care model.

PMID:39803417 | PMC:PMC11719832 | DOI:10.1016/j.xkme.2024.100927

Eur J Radiol Open. 2024 Dec 17;14:100624. doi: 10.1016/j.ejro.2024.100624. eCollection 2025 Jun.

ABSTRACT

OBJECTIVES: To develop and validate a deep learning model using multimodal PET/CT imaging for detecting and classifying focal liver lesions (FLL).

METHODS: This study included 185 patients who underwent 18F-FDG PET/CT imaging at our institution from March 2022 to February 2023. We analyzed serological data and imaging. Liver lesions were segmented on PET and CT, serving as the "reference standard". Deep learning models were trained using PET and CT images to generate predicted segmentations and classify lesion nature. Model performance was evaluated by comparing the predicted segmentations with the reference segmentations, using metrics such as Dice, Precision, Recall, F1-score, ROC, and AUC, and compared it with physician diagnoses.

RESULTS: This study finally included 150 patients, comprising 46 patients with benign liver nodules, 51 patients with malignant liver nodules, and 53 patients with no FLLs. Significant differences were observed among groups for age, AST, ALP, GGT, AFP, CA19-9and CEA. On the validation set, the Dice coefficient of the model was 0.740. For the normal group, the recall was 0.918, precision was 0.904, F1-score was 0.909, and AUC was 0.976. For the benign group, the recall was 0.869, precision was 0.862, F1-score was 0.863, and AUC was 0.928. For the malignant group, the recall was 0.858, precision was 0.914, F1-score was 0.883, and AUC was 0.979. The model's overall diagnostic performance was between that of junior and senior physician.

CONCLUSION: This deep learning model demonstrated high sensitivity in detecting FLLs and effectively differentiated between benign and malignant lesions.

PMID:39803389 | PMC:PMC11720101 | DOI:10.1016/j.ejro.2024.100624

Phys Imaging Radiat Oncol. 2024 Sep 30;32:100654. doi: 10.1016/j.phro.2024.100654. eCollection 2024 Oct.

ABSTRACT

BACKGROUND AND PURPOSE: Deep-learning-based automatic segmentation is widely used in radiation oncology to delineate organs-at-risk. Dual-energy CT (DECT) allows the reconstruction of enhanced contrast images that could help with manual and auto-delineation. This paper presents a performance evaluation of a commercial auto-segmentation software on image series generated by a DECT.

MATERIAL AND METHODS: Different types of DECT images from seventy four head-and-neck (HN) patients were retrieved, including polyenergetic images at different voltages [80 kV reconstructed with a kernel corresponding to the commercial algorithm DirectDensity™ (PEI80-DD), 80 kV (PEI80), 120 kV-mixed (PEI120)] and a virtual-monoenergetic image at 40 keV (VMI40). Delineations used for treatment planning were considered as ground truth (GT) and were compared with the auto-segmentations performed on the 4 DECT images. A blinded qualitative evaluation of 3 structures (thyroid, left parotid, left nodes level II) was carried out. Performance metrics were calculated for thirteen HN structures to evaluate the auto-contours including dice similarity coefficient (DSC), 95th percentile Hausdorff distance (95HD) and mean surface distance (MSD).

RESULTS: We observed a high rate of low scores for PEI80-DD and VMI40 auto-segmentations on the thyroid and for GT and VMI40 contours on the nodes level II. All images received excellent scores for the parotid glands. The metrics comparison between GT and auto-segmented contours revealed that PEI80-DD had the highest DSC scores, significantly outperforming other reconstructed images for all organs (p < 0.05).

CONCLUSIONS: The results indicate that the auto-contouring system cannot generalize to images derived from DECT acquisition. It is therefore crucial to identify which organs benefit from these acquisitions to adapt the training datasets accordingly.

PMID:39803347 | PMC:PMC11718415 | DOI:10.1016/j.phro.2024.100654

Korean J Neurotrauma. 2024 Dec 16;20(4):303-304. doi: 10.13004/kjnt.2024.20.e40. eCollection 2024 Dec.

NO ABSTRACT

PMID:39803342 | PMC:PMC11711028 | DOI:10.13004/kjnt.2024.20.e40

Brief Bioinform. 2024 Nov 22;26(1):bbae719. doi: 10.1093/bib/bbae719.

ABSTRACT

Spatial transcriptomics technologies have been extensively applied in biological research, enabling the study of transcriptome while preserving the spatial context of tissues. Paired with spatial transcriptomics data, platforms often provide histology and (or) chromatin images, which capture cellular morphology and chromatin organization. Additionally, single-cell RNA sequencing (scRNA-seq) data from matching tissues often accompany spatial data, offering a transcriptome-wide gene expression profile of individual cells. Integrating such additional data from other modalities can effectively enhance spatial transcriptomics data, and, conversely, spatial transcriptomics data can supplement scRNA-seq with spatial information. Moreover, the rapid development of spatial multi-omics technology has spurred the demand for the integration of spatial multi-omics data to present a more detailed molecular landscape within tissues. Numerous deep learning (DL) methods have been developed for integrating spatial transcriptomics with other modalities. However, a comprehensive review of DL approaches for integrating spatial transcriptomics data with other modalities remains absent. In this study, we systematically review the applications of DL in integrating spatial transcriptomics data with other modalities. We first delineate the DL techniques applied in this integration and the key tasks involved. Next, we detail these methods and categorize them based on integrated modality and key task. Furthermore, we summarize the integration strategies of these integration methods. Finally, we discuss the challenges and future directions in integrating spatial transcriptomics with other modalities, aiming to facilitate the development of robust computational methods that more comprehensively exploit multimodal information.

PMID:39800876 | DOI:10.1093/bib/bbae719

Brief Bioinform. 2024 Nov 22;26(1):bbae717. doi: 10.1093/bib/bbae717.

ABSTRACT

Combination therapies have emerged as a promising approach for treating complex diseases, particularly cancer. However, predicting the efficacy and safety profiles of these therapies remains a significant challenge, primarily because of the complex interactions among drugs and their wide-ranging effects. To address this issue, we introduce DD-PRiSM (Decomposition of Drug-Pair Response into Synergy and Monotherapy effect), a deep-learning pipeline that predicts the effects of combination therapy. DD-PRiSM consists of two predictive models. The first is the Monotherapy model, which predicts parameters of the drug response curve based on drug structure and cell line gene expression. This reconstructed curve is then used to predict cell viability at the given drug dosage. The second is the Combination therapy model, which predicts the efficacy of drug combinations by analyzing individual drug effects and their synergistic interactions with a specific dosage level of individual drugs. The efficacy of DD-PRiSM is demonstrated through its performance metrics, achieving a root mean square error of 0.0854, a Pearson correlation coefficient of 0.9063, and an R2 of 0.8209 for unseen pairs. Furthermore, DD-PRiSM distinguishes itself by its capability to decompose combination therapy efficacy, successfully identifying synergistic drug pairs. We demonstrated synergistic responses vary across cancer types and identified hub drugs that trigger synergistic effects. Finally, we suggested a promising drug pair through our case study.

PMID:39800875 | DOI:10.1093/bib/bbae717

Brief Bioinform. 2024 Nov 22;26(1):bbae711. doi: 10.1093/bib/bbae711.

ABSTRACT

Single-cell multi-omics techniques, which enable the simultaneous measurement of multiple modalities such as RNA gene expression and Assay for Transposase-Accessible Chromatin (ATAC) within individual cells, have become a powerful tool for deciphering the intricate complexity of cellular systems. Most current methods rely on motif databases to establish cross-modality relationships between genes from RNA-seq data and peaks from ATAC-seq data. However, these approaches are constrained by incomplete database coverage, particularly for novel or poorly characterized relationships. To address these limitations, we introduce single-cell Multi-omics Integration (scMI), a heterogeneous graph embedding method that encodes both cells and modality features from single-cell RNA-seq and ATAC-seq data into a shared latent space by learning cross-modality relationships. By modeling cells and modality features as distinct node types, we design an inter-type attention mechanism to effectively capture long-range cross-modality interactions between genes and peaks. Benchmark results demonstrate that embeddings learned by scMI preserve more biological information and achieve comparable or superior performance in downstream tasks including modality prediction, cell clustering, and gene regulatory network inference compared to methods that rely on databases. Furthermore, scMI significantly improves the alignment and integration of unmatched multi-omics data, enabling more accurate embedding and improved outcomes in downstream tasks.

PMID:39800872 | DOI:10.1093/bib/bbae711

Environ Sci Pollut Res Int. 2025 Jan 13. doi: 10.1007/s11356-024-35764-8. Online ahead of print.

ABSTRACT

Human-induced global warming, primarily attributed to the rise in atmospheric CO2, poses a substantial risk to the survival of humanity. While most research focuses on predicting annual CO2 emissions, which are crucial for setting long-term emission mitigation targets, the precise prediction of daily CO2 emissions is equally vital for setting short-term targets. This study examines the performance of 14 models in predicting daily CO2 emissions data from 1/1/2022 to 30/9/2023 across the top four polluting regions (China, India, the USA, and the EU27&UK). The 14 models used in the study include four statistical models (ARMA, ARIMA, SARMA, and SARIMA), three machine learning models (support vector machine (SVM), random forest (RF), and gradient boosting (GB)), and seven deep learning models (artificial neural network (ANN), recurrent neural network variations such as gated recurrent unit (GRU), long short-term memory (LSTM), bidirectional-LSTM (BILSTM), and three hybrid combinations of CNN-RNN). Performance evaluation employs four metrics (R2, MAE, RMSE, and MAPE). The results show that the machine learning (ML) and deep learning (DL) models, with higher R2 (0.714-0.932) and lower RMSE (0.480-0.247) values, respectively, outperformed the statistical model, which had R2 (- 0.060-0.719) and RMSE (1.695-0.537) values, in predicting daily CO2 emissions across all four regions. The performance of the ML and DL models was further enhanced by differencing, a technique that improves accuracy by ensuring stationarity and creating additional features and patterns from which the model can learn. Additionally, applying ensemble techniques such as bagging and voting improved the performance of the ML models by approximately 9.6%, whereas hybrid combinations of CNN-RNN enhanced the performance of the RNN models. In summary, the performance of both the ML and DL models was relatively similar. However, due to the high computational requirements associated with DL models, the recommended models for daily CO2 emission prediction are ML models using the ensemble technique of voting and bagging. This model can assist in accurately forecasting daily emissions, aiding authorities in setting targets for CO2 emission reduction.

PMID:39800837 | DOI:10.1007/s11356-024-35764-8

Breast Cancer Res. 2025 Jan 12;27(1):6. doi: 10.1186/s13058-025-01959-1.

ABSTRACT

Recent evidence indicates that endocrine resistance in estrogen receptor-positive (ER+) breast cancer is closely correlated with phenotypic characteristics of epithelial-to-mesenchymal transition (EMT). Nonetheless, identifying tumor tissues with a mesenchymal phenotype remains challenging in clinical practice. In this study, we validated the correlation between EMT status and resistance to endocrine therapy in ER+ breast cancer from a transcriptomic perspective. To confirm the presence of morphological discrepancies in tumor tissues of ER+ breast cancer classified as epithelial- and mesenchymal-phenotypes according to EMT-related transcriptional features, we trained deep learning algorithms based on EfficientNetV2 architecture to assign the phenotypic status for each patient utilizing hematoxylin & eosin (H&E)-stained slides from The Cancer Genome Atlas database. Our classifier model accurately identified the precise phenotypic status, achieving an area under the curve (AUC) of 0.886 at the tile-level and an AUC of 0.910 at the slide-level. Furthermore, we evaluated the efficacy of the classifier in predicting endocrine response using data from an independent ER+ breast cancer patient cohort. Our classifier achieved a predicting accuracy of 81.25%, and 88.7% slides labeled as endocrine resistant were predicted as the mesenchymal-phenotype, while 75.6% slides labeled as sensitive were predicted as the epithelial-phenotype. Our work introduces an H&E-based framework capable of accurately predicting EMT phenotype and endocrine response for ER+ breast cancer, demonstrating its potential for clinical application and benefit.

PMID:39800743 | DOI:10.1186/s13058-025-01959-1

Global Spine J. 2025 Jan 12:21925682251314379. doi: 10.1177/21925682251314379. Online ahead of print.

ABSTRACT

STUDY DESIGN: Systematic review.

OBJECTIVE: Artificial intelligence (AI) and deep learning (DL) models have recently emerged as tools to improve fracture detection, mainly through imaging modalities such as computed tomography (CT) and radiographs. This systematic review evaluates the diagnostic performance of AI and DL models in detecting cervical spine fractures and assesses their potential role in clinical practice.

METHODS: A systematic search of PubMed/Medline, Embase, Scopus, and Web of Science was conducted for studies published between January 2000 and July 2024. Studies that evaluated AI models for cervical spine fracture detection were included. Diagnostic performance metrics were extracted and included sensitivity, specificity, accuracy, and area under the curve. The PROBAST tool assessed bias, and PRISMA criteria were used for study selection and reporting.

RESULTS: Eleven studies published between 2021 and 2024 were included in the review. AI models demonstrated variable performance, with sensitivity ranging from 54.9% to 100% and specificity from 72% to 98.6%. Models applied to CT imaging generally outperformed those applied to radiographs, with convolutional neural networks (CNN) and advanced architectures such as MobileNetV2 and Vision Transformer (ViT) achieving the highest accuracy. However, most studies lacked external validation, raising concerns about the generalizability of their findings.

CONCLUSIONS: AI and DL models show significant potential in improving fracture detection, particularly in CT imaging. While these models offer high diagnostic accuracy, further validation and refinement are necessary before they can be widely integrated into clinical practice. AI should complement, rather than replace, human expertise in diagnostic workflows.

PMID:39800538 | DOI:10.1177/21925682251314379

Open Heart. 2025 Jan 11;12(1):e002998. doi: 10.1136/openhrt-2024-002998.

ABSTRACT

PURPOSE: We examined whether end-to-end deep-learning models could detect moderate (≥50%) or severe (≥70%) stenosis in the left anterior descending artery (LAD), right coronary artery (RCA) or left circumflex artery (LCX) in iodine contrast-enhanced ECG-gated coronary CT angiography (CCTA) scans.

METHODS: From a database of 6293 CCTA scans, we used pre-existing curved multiplanar reformations (CMR) images of the LAD, RCA and LCX arteries to create end-to-end deep-learning models for the detection of moderate or severe stenoses. We preprocessed the images by exploiting domain knowledge and employed a transfer learning approach using EfficientNet, ResNet, DenseNet and Inception-ResNet, with a class-weighted strategy optimised through cross-validation. Heatmaps were generated to indicate critical areas identified by the models, aiding clinicians in understanding the model's decision-making process.

RESULTS: Among the 900 CMR cases, 279 involved the LAD artery, 259 the RCA artery and 253 the LCX artery. EfficientNet models outperformed others, with EfficientNetB3 and EfficientNetB0 demonstrating the highest accuracy for LAD, EfficientNetB2 for RCA and EfficientNetB0 for LCX. The area under the curve for receiver operating characteristic (AUROC) reached 0.95 for moderate and 0.94 for severe stenosis in the LAD. For the RCA, the AUROC was 0.92 for both moderate and severe stenosis detection. The LCX achieved an AUROC of 0.88 for the detection of moderate stenoses, though the calibration curve exhibited significant overestimation. Calibration curves matched probabilities for the LAD but showed discrepancies for the RCA. Heatmap visualisations confirmed the models' precision in delineating stenotic lesions. Decision curve analysis and net reclassification index assessments reinforced the efficacy of EfficientNet models, confirming their superior diagnostic capabilities.

CONCLUSION: Our end-to-end deep-learning model demonstrates, for the LAD artery, excellent discriminatory ability and calibration during internal validation, despite a small dataset used to train the network. The model reliably produces precise, highly interpretable images.

PMID:39800435 | DOI:10.1136/openhrt-2024-002998

Int J Radiat Oncol Biol Phys. 2025 Jan 10:S0360-3016(25)00002-1. doi: 10.1016/j.ijrobp.2024.12.026. Online ahead of print.

ABSTRACT

PURPOSE: High dose rate (HDR) prostate brachytherapy (BT) procedure requires image-guided needle insertion. Given that general anesthesia is often employed during the procedure, minimizing overall planning time is crucial. In this study, we explore the clinical feasibility and time-saving potential of artificial intelligence (AI)-driven auto-reconstruction of transperineal needles in the context of US-guided prostate BT planning.

MATERIALS AND METHODS: This study included a total of 102 US-planned BT images from a single institution and split into three groups: 50 for model training and validation, 11 to evaluate reconstruction accuracy (test set), and 41 to evaluate the AI tool in a clinical implementation (clinical set). Reconstruction accuracy for the test set was evaluated by comparing the performance of AI-derived and manually reconstructed needles from 5 medical physicists on the 3D-US scans after treatment. The needle total reconstruction time for the clinical set was defined as the timestamp difference from scan acquisition to the start of dose calculations and was compared to values recorded before the clinical implementation of the AI-assisted tool.

RESULTS: A mean error of (0.44±0.32)mm was found between the AI-reconstructed and the human consensus needle positions in the test set, with 95.0% of AI needle points falling below 1mm from their human-made counterparts. Post-hoc analysis showed only one of the human observers' reconstructions were significantly different from the others including the AI's. In the clinical set, the AI algorithm achieved a true positive reconstruction rate of 93.4% with only 4.5% of these needles requiring manual corrections from the planner before dosimetry. Total time required to perform AI-assisted catheter reconstruction on clinical cases was on average 15.2min lower (p < 0.01) compared to procedure without AI assistance.

CONCLUSIONS: This study demonstrates the feasibility of an AI-assisted needle reconstructing tool for 3D-US based HDR prostate BT. This is a step toward treatment planning automation and increased efficiency in HDR prostate BT.

PMID:39800329 | DOI:10.1016/j.ijrobp.2024.12.026

Neuroimage. 2025 Jan 10:121002. doi: 10.1016/j.neuroimage.2025.121002. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: The RANO-BM criteria, which employ a one-dimensional measurement of the largest diameter, are imperfect due to the fact that the lesion volume is neither isotropic nor homogeneous. Furthermore, this approach is inherently time-consuming. Consequently, in clinical practice, monitoring patients in clinical trials in compliance with the RANO-BM criteria is rarely achieved. The objective of this study was to develop and validate an AI solution capable of delineating brain metastases (BM) on MRI to easily obtain, using an in-house solution, RANO-BM criteria as well as BM volume in a routine clinical setting.

MATERIALS (PATIENTS) AND METHODS: A total of 27456 post-Gadolinium-T1 MRI from 132 patients with BM were employed in this study. A deep learning (DL) model was constructed using the PyTorch and PyTorch Lightning frameworks, and the UNETR transfer learning method was employed to segment BM from MRI.

RESULTS: A visual analysis of the AI model results demonstrates confident delineation of the BM lesions. The model shows 100% accuracy in predicting RANO-BM criteria in comparison to that of an expert medical doctor. There was a high degree of overlap between the AI and the doctor's segmentation, with a mean DICE score of 0.77. The diameter and volume of the BM lesions were found to be concordant between the AI and the reference segmentation. The user interface developed in this study can readily provide RANO-BM criteria following AI BM segmentation.

CONCLUSION: The in-house deep learning solution is accessible to everyone without expertise in AI and offers effective BM segmentation and substantial time savings.

PMID:39800174 | DOI:10.1016/j.neuroimage.2025.121002

Heart Rhythm. 2025 Jan 10:S1547-5271(25)00019-0. doi: 10.1016/j.hrthm.2024.12.048. Online ahead of print.

ABSTRACT

BACKGROUND: Although smartphone-based devices have been developed to record 1-lead ECG, existing solutions for automatic atrial fibrillation (AF) detection often has poor positive predictive value.

OBJECTIVE: This study aimed to validate a cloud-based deep learning platform for automatic AF detection in a large cohort of patients using 1-lead ECG records.

METHODS: We analyzed 8,528 patients with 30-second ECG records from a single-lead handheld ECG device. Ground truth for AF presence was established through a benchmark algorithm and expert manual labeling. The Willem Artificial Intelligence (AI) platform, not trained on these ECGs, was used for automatic arrhythmia detection, including AF. A rules-based algorithm was also used for comparison. An expert cardiology committee reviewed false positives and negatives and performance metrics were computed.

RESULTS: The AI platform achieved an accuracy of 96.1% (initial labels) and 96.4% (expert review), with sensitivities of 83.3% and 84.2%, and specificities of 97.3% and 97.6%, respectively. The positive predictive value was 75.2% and 78.0%, and the negative predictive value was 98.4%. Performance of the AI platform largely exceeded the performance of the rules-based algorithm for all metrics. The AI also detected other arrhythmias, such as premature ventricular complexes, premature atrial complexes along with 1-degree atrioventricular blocks.

CONCLUSIONS: The result of this external validation indicates that the AI platform can match cardiologist-level accuracy in AF detection from 1-lead ECGs. Such tools are promising for AF screening and has the potential to improve accuracy in non-cardiology expert healthcare professional interpretation and trigger further tests for effective patient management.

PMID:39800092 | DOI:10.1016/j.hrthm.2024.12.048

Int J Comput Assist Radiol Surg. 2025 Jan 12. doi: 10.1007/s11548-024-03306-9. Online ahead of print.

ABSTRACT

PURPOSE: This paper presents a deep learning approach to recognize and predict surgical activity in robot-assisted minimally invasive surgery (RAMIS). Our primary objective is to deploy the developed model for implementing a real-time surgical risk monitoring system within the realm of RAMIS.

METHODS: We propose a modified Transformer model with the architecture comprising no positional encoding, 5 fully connected layers, 1 encoder, and 3 decoders. This model is specifically designed to address 3 primary tasks in surgical robotics: gesture recognition, prediction, and end-effector trajectory prediction. Notably, it operates solely on kinematic data obtained from the joints of robotic arm.

RESULTS: The model's performance was evaluated on JHU-ISI Gesture and Skill Assessment Working Set dataset, achieving highest accuracy of 94.4% for gesture recognition, 84.82% for gesture prediction, and significantly low distance error of 1.34 mm with a prediction of 1 s in advance. Notably, the computational time per iteration was minimal recorded at only 4.2 ms.

CONCLUSION: The results demonstrated the excellence of our proposed model compared to previous studies highlighting its potential for integration in real-time systems. We firmly believe that our model could significantly elevate realms of surgical activity recognition and prediction within RAS and make a substantial and meaningful contribution to the healthcare sector.

PMID:39799528 | DOI:10.1007/s11548-024-03306-9

Bioinformatics. 2025 Jan 12:btaf019. doi: 10.1093/bioinformatics/btaf019. Online ahead of print.

ABSTRACT

MOTIVATION: The knowledge of protein stability upon residue variation is an important step for functional protein design and for understanding how protein variants can promote disease onset. Computational methods are important to complement experimental approaches and allow a fast screening of large datasets of variations.

RESULTS: In this work we present DDGemb, a novel method combining protein language model embeddings and transformer architectures to predict protein ΔΔG upon both single- and multi-point variations. DDGemb has been trained on a high-quality dataset derived from literature and tested on available benchmark datasets of single- and multi-point variations. DDGemb performs at the state of the art in both single- and multi-point variations.

AVAILABILITY: DDGemb is available as web server at https://ddgemb.biocomp.unibo.it. Datasets used in this study are available at https://ddgemb.biocomp.unibo.it/datasets.

PMID:39799516 | DOI:10.1093/bioinformatics/btaf019

J Orthop Surg Res. 2025 Jan 11;20(1):41. doi: 10.1186/s13018-024-05334-2.

ABSTRACT

PURPOSE: The study aimed to develop a deep learning model for rapid, automated measurement of full-spine X-rays in adolescents with Adolescent Idiopathic Scoliosis (AIS). A significant challenge in this field is the time-consuming nature of manual measurements and the inter-individual variability in these measurements. To address these challenges, we utilized RTMpose deep learning technology to automate the process.

METHODS: We conducted a retrospective multicenter diagnostic study using 560 full-spine sagittal plane X-ray images from five hospitals in Inner Mongolia. The model was trained and validated using 500 images, with an additional 60 images for independent external validation. We evaluated the consistency of keypoint annotations among different physicians, the accuracy of model-predicted keypoints, and the accuracy of model measurement results compared to manual measurements.

RESULTS: The consistency percentages of keypoint annotations among different physicians and the model were 90-97% within the 4-mm range. The model's prediction accuracies for key points were 91-100% within the 4-mm range compared to the reference standards. The model's predictions for 15 anatomical parameters showed high consistency with experienced physicians, with intraclass correlation coefficients ranging from 0.892 to 0.991. The mean absolute error for SVA was 1.16 mm, and for other parameters, it ranged from 0.22° to 3.32°. A significant challenge we faced was the variability in data formats and specifications across different hospitals, which we addressed through data augmentation techniques. The model took an average of 9.27 s to automatically measure the 15 anatomical parameters per X-ray image.

CONCLUSION: The deep learning model based on RTMpose can effectively enhance clinical efficiency by automatically measuring the sagittal plane parameters of the spine in X-rays of patients with AIS. The model's performance was found to be highly consistent with manual measurements by experienced physicians, offering a valuable tool for clinical diagnostics.

PMID:39799363 | DOI:10.1186/s13018-024-05334-2