Med Phys. 2024 Oct 30. doi: 10.1002/mp.17498. Online ahead of print.

ABSTRACT

BACKGROUND: Accurate preoperative prediction of cervical lymph node metastasis (LNM) for papillary thyroid carcinoma (PTC) patients is essential for disease staging and individualized treatment planning, which can improve prognosis and facilitate better management.

PURPOSE: To establish a fully automated deep learning-enabled model (FADLM) for automated tumor segmentation and cervical LNM prediction in PTC using ultrasound (US) video keyframes.

METHODS: The bicentral study retrospective enrolled 518 PTC patients, who were then randomly divided into the training (Hospital 1, n = 340), internal test (Hospital 1, n = 83), and external test cohorts (Hospital 2, n = 95). The FADLM integrated mask region-based convolutional neural network (Mask R-CNN) for automatic thyroid primary tumor segmentation and ResNet34 with Bayes strategy for cervical LNM diagnosis. A radiomics model (RM) using the same automated segmentation method, a traditional radiomics model (TRM) using manual segmentation, and a clinical-semantic model (CSM) were developed for comparison. The dice similarity coefficient (DSC) was used to evaluate segmentation performance. The prediction performance of the models was validated in terms of discrimination and clinical utility with the area under the receiver operator characteristic curve (AUC), heatmap analysis, and decision curve analysis (DCA). The comparison of the predictive performance among different models was conducted by DeLong test. The performances of two radiologists compared with FADLM and the diagnostic augmentation with FADLM's assistance were analyzed in terms of accuracy, sensitivity and specificity using McNemar's x2 test. The p-value less than 0.05 was defined as a statistically significant difference. The Benjamini-Hochberg procedure was applied for multiple comparisons to deal with Type I error.

RESULTS: The FADLM yielded promising segmentation results in training (DSC: 0.88 ± 0.23), internal test (DSC: 0.88 ± 0.23), and external test cohorts (DSC: 0.85 ± 0.24). The AUCs of FADLM for cervical LNM prediction were 0.78 (95% CI: 0.73, 0.83), 0.83 (95% CI: 0.74, 0.92), and 0.83 (95% CI: 0.75, 0.92), respectively. It all significantly outperformed the RM (AUCs: 0.78 vs. 0.72; 0.83 vs. 0.65; 0.83 vs. 0.68, all adjusted p-values < 0.05) and CSM (AUCs: 0.78 vs. 0.71; 0.83 vs. 0.62; 0.83 vs. 0.68, all adjusted p-values < 0.05) across the three cohorts. The RM offered similar performance to that of TRM (AUCs: 0.61 vs. 0.63, adjusted p-value = 0.60) while significantly reducing the segmentation time (3.3 ± 3.8 vs. 14.1 ± 4.2 s, p-value < 0.001). Under the assistance of FADLM, the accuracies of junior and senior radiologists were improved by 18% and 15% (all adjusted p-values < 0.05) and the sensitivities by 25% and 21% (all adjusted p-values < 0.05) in the external test cohort.

CONCLUSION: The FADLM with elaborately designed automated strategy using US video keyframes holds good potential to provide an efficient and consistent prediction of cervical LNM in PTC. The FADLM displays superior performance to RM, CSM, and radiologists with promising efficacy.

PMID:39475358 | DOI:10.1002/mp.17498

Cereb Cortex. 2024 Oct 3;34(10):bhae427. doi: 10.1093/cercor/bhae427.

ABSTRACT

Research on action-based timing has shed light on the temporal dynamics of sensorimotor coordination. This study investigates the neural mechanisms underlying action-based timing, particularly during finger-tapping tasks involving synchronized and syncopated patterns. Twelve healthy participants completed a continuation task, alternating between tapping in time with an auditory metronome (pacing) and continuing without it (continuation). Electroencephalography data were collected to explore how neural activity changes across these coordination modes and phases. We applied deep learning methods to classify single-trial electroencephalography data and predict behavioral timing conditions. Results showed significant classification accuracy for distinguishing between pacing and continuation phases, particularly during the presence of auditory cues, emphasizing the role of auditory input in motor timing. However, when auditory components were removed from the electroencephalography data, the differentiation between phases became inconclusive. Mean accuracy asynchrony, a measure of timing error, emerged as a superior predictor of performance variability compared to inter-response interval. These findings highlight the importance of auditory cues in modulating motor timing behaviors and present the challenges of isolating motor activation in the absence of auditory stimuli. Our study offers new insights into the neural dynamics of motor timing and demonstrates the utility of deep learning in analyzing single-trial electroencephalography data.

PMID:39475113 | DOI:10.1093/cercor/bhae427

Br J Hosp Med (Lond). 2024 Oct 30;85(10):1-14. doi: 10.12968/hmed.2024.0428. Epub 2024 Oct 29.

ABSTRACT

Aims/Background In the treatment of patients with cervical cancer, lymph node metastasis (LNM) is an important indicator for stratified treatment and prognosis of cervical cancer. This study aimed to develop and validate a multimodal model based on contrast-enhanced multiphase computed tomography (CT) images and clinical variables to accurately predict LNM in patients with cervical cancer. Methods This study included 233 multiphase contrast-enhanced CT images of patients with pathologically confirmed cervical malignancies treated at the Affiliated Dongyang Hospital of Wenzhou Medical University. A three-dimensional MedicalNet pre-trained model was used to extract features. Minimum redundancy-maximum correlation, and least absolute shrinkage and selection operator regression were used to screen the features that were ultimately combined with clinical candidate predictors to build the prediction model. The area under the curve (AUC) was used to assess the predictive efficacy of the model. Results The results indicate that the deep transfer learning model exhibited high diagnostic performance within the internal validation set, with an AUC of 0.82, accuracy of 0.88, sensitivity of 0.83, and specificity of 0.89. Conclusion We constructed a comprehensive, multiparameter model based on the concept of deep transfer learning, by pre-training the model with contrast-enhanced multiphase CT images and an array of clinical variables, for predicting LNM in patients with cervical cancer, which could aid the clinical stratification of these patients via a noninvasive manner.

PMID:39475034 | DOI:10.12968/hmed.2024.0428

Ann Surg. 2024 Oct 30. doi: 10.1097/SLA.0000000000006577. Online ahead of print.

ABSTRACT

OBJECTIVE: This study aimed to compare the ability of a deep-learning platform (the MACSSwin-T model) with healthcare professionals in detecting cerebral aneurysms from operative videos. Secondly, we aimed to compare the neurosurgical team's ability to detect cerebral aneurysms with and without AI-assistance.

BACKGROUND: Modern microscopic surgery enables the capture of operative video data on an unforeseen scale. Advances in computer vision, a branch of artificial intelligence (AI), have enabled automated analysis of operative video. These advances are likely to benefit clinicians, healthcare systems, and patients alike, yet such benefits are yet to be realised.

METHODS: In a cross-sectional comparative study, neurosurgeons, anaesthetists, and operating room (OR) nurses, all at varying stages of training and experience, reviewed still frames of aneurysm clipping operations and labelled frames as "aneurysm not in frame" or "aneurysm in frame". Frames then underwent analysis by the AI platform. A second round of data collection was performed whereby the neurosurgical team had AI-assistance. Accuracy of aneurysm detection was calculated for human only, AI only, and AI-assisted human groups.

RESULTS: 5,154 individual frame reviews were collated from 338 healthcare professionals. Healthcare professionals correctly labelled 70% of frames without AI assistance, compared to 78% with AI-assistance (OR 1.77, P<0.001). Neurosurgical Attendings showed the greatest improvement, from 77% to 92% correct predictions with AI-assistance (OR 4.24, P=0.003).

CONCLUSION: AI-assisted human performance surpassed both human and AI alone. Notably, across healthcare professionals surveyed, frame accuracy improved across all subspecialties and experience levels, particularly among the most experienced healthcare professionals. These results challenge the prevailing notion that AI primarily benefits junior clinicians, highlighting its crucial role throughout the surgical hierarchy as an essential component of modern surgical practice.

PMID:39474680 | DOI:10.1097/SLA.0000000000006577

Front Pharmacol. 2024 Oct 15;15:1432592. doi: 10.3389/fphar.2024.1432592. eCollection 2024.

ABSTRACT

BACKGROUND: Aging is marked by the gradual deterioration of cells, tissues, and organs and is a major risk factor for many chronic diseases. Considering the complex mechanisms of aging, traditional Chinese medicine (TCM) could offer distinct advantages. However, due to the complexity and variability of metabolites in TCM, the comprehensive screening of metabolites associated with pharmacology remains a significant issue.

METHODS: A reliable and integrated identification method based on UPLC-Q Exactive-Orbitrap HRMS was established to identify the chemical profiles of Huan Shao Dan (HSD). Then, based on the theory of sequential metabolism, the metabolic sites of HSD in vivo were further investigated. Finally, a deep learning model and a bioactivity assessment assay were applied to screen potential anti-aging metabolites.

RESULTS: This study identified 366 metabolites in HSD. Based on the results of sequential metabolism, 135 metabolites were then absorbed into plasma. A total of 178 peaks were identified from the sample after incubation with artificial gastric juice. In addition, 102 and 91 peaks were identified from the fecal and urine samples, respectively. Finally, based on the results of the deep learning model and bioactivity assay, ginsenoside Rg1, Rg2, and Rc, pseudoginsenoside F11, and jionoside B1 were selected as potential anti-aging metabolites.

CONCLUSION: This study provides a valuable reference for future research on the material basis of HSD by describing the chemical profiles both in vivo and in vitro. Moreover, the proposed screening approach may serve as a rapid tool for identifying potential anti-aging metabolites in TCM.

PMID:39474607 | PMC:PMC11518704 | DOI:10.3389/fphar.2024.1432592

Cureus. 2024 Oct 29;16(10):e72646. doi: 10.7759/cureus.72646. eCollection 2024 Oct.

ABSTRACT

Artificial intelligence (AI) is rapidly transforming the field of radiology, offering significant advancements in diagnostic accuracy, workflow efficiency, and patient care. This article explores AI's impact on various subfields of radiology, emphasizing its potential to improve clinical practices and enhance patient outcomes. AI-driven technologies such as machine learning, deep learning, and natural language processing (NLP) are playing a pivotal role in automating routine tasks, aiding in early disease detection, and supporting clinical decision-making, allowing radiologists to focus on more complex diagnostic challenges. Key applications of AI in radiology include improving image analysis through computer-aided diagnosis (CAD) systems, which enhance the detection of abnormalities in imaging, such as tumors. AI tools have demonstrated high accuracy in analyzing medical images, integrating data from multiple imaging modalities such as CT, MRI, and PET to provide comprehensive diagnostic insights. These advancements facilitate personalized treatment planning and complement radiologists' workflows. However, for AI to be fully integrated into radiology workflows, several challenges must be addressed, including ensuring transparency in how AI algorithms work, protecting patient data, and avoiding biases that could affect diverse populations. Developing explainable AI systems that can clearly show how decisions are made is crucial, as is ensuring AI tools can seamlessly fit into existing radiology systems. Collaboration between radiologists, AI developers, and policymakers, alongside strong ethical guidelines and regulatory oversight, will be key to ensuring AI is implemented safely and effectively in clinical practice. Overall, AI holds tremendous promise in revolutionizing radiology. Through its ability to automate complex tasks, enhance diagnostic capabilities, and streamline workflows, AI has the potential to significantly improve the quality and efficiency of radiology practices. Continued research, development, and collaboration will be crucial in unlocking AI's full potential and addressing the challenges that accompany its adoption.

PMID:39474591 | PMC:PMC11521355 | DOI:10.7759/cureus.72646

Chem Biomed Imaging. 2024 May 6;2(6):442-452. doi: 10.1021/cbmi.4c00007. eCollection 2024 Jun 24.

ABSTRACT

The combination of deep learning techniques and Raman spectroscopy shows great potential offering precise and prompt identification of pathogenic bacteria in clinical settings. However, the traditional closed-set classification approaches assume that all test samples belong to one of the known pathogens, and their applicability is limited since the clinical environment is inherently unpredictable and dynamic, unknown, or emerging pathogens may not be included in the available catalogs. We demonstrate that the current state-of-the-art neural networks identifying pathogens through Raman spectra are vulnerable to unknown inputs, resulting in an uncontrollable false positive rate. To address this issue, first we developed an ensemble of ResNet architectures combined with the attention mechanism that achieves a 30-isolate accuracy of 87.8 ± 0.1%. Second, through the integration of feature regularization by the Objectosphere loss function, our model both achieves high accuracy in identifying known pathogens from the catalog and effectively separates unknown samples drastically reducing the false positive rate. Finally, the proposed feature regularization method during training significantly enhances the performance of out-of-distribution detectors during the inference phase improving the reliability of the detection of unknown classes. Our algorithm for Raman spectroscopy empowers the identification of previously unknown, uncataloged, and emerging pathogens ensuring adaptability to future pathogens that may surface. Moreover, it can be extended to enhance open-set medical image classification, bolstering its reliability in dynamic operational settings.

PMID:39474520 | PMC:PMC11503672 | DOI:10.1021/cbmi.4c00007

Lancet Reg Health Am. 2024 Aug 9;37:100852. doi: 10.1016/j.lana.2024.100852. eCollection 2024 Sep.

ABSTRACT

BACKGROUND: Existing research lacks information on the potential impacts of multi-phased coronavirus disease 2019 (COVID-19) vaccine rollouts on population mental health. This study aims to evaluate the impact of various COVID-19 vaccine rollout phases on trends and prevalence of anxiety and depression among US adults at a population level.

METHODS: We performed a US population-based multi-intervention interrupted time series analysis through Deep Learning and autoregressive integrated moving average (ARIMA) approaches, analyzing 4 waves of US CDC's Behavioral Risk Factor Surveillance System (BRFSS) data (January 2019-February 2023) to assess changes in the weekly prevalence of anxiety and depression following interruptions, including all major COVID-19 vaccine rollout phases from 2020 to early 2023 while considering pandemic-related events.

FINDINGS: Among 1,615,643 US adults (1,011,300 [76.4%] aged 18-64 years, 867,826 [51.2%] female, 126,594 [16.9%] Hispanic, 120,380 [11.9%] non-Hispanic Black, 1,191,668 [61.7%] non-Hispanic White, and 113,461 [9.5%] other non-Hispanic people of color), we found that three COVID-19 vaccine rollout phases (ie, prioritization for educational/childcare workers, boosters for all US adults, authorization for young children) were associated with a 0.93 percentage-point (95% CI -1.81 to -0.04, p = 0.041), 1.28 percentage-point (95% CI -2.32 to -0.24, p = 0.017), and 0.89 percentage-point (95% CI -1.56 to -0.22, p = 0.010) reduction, respectively, in anxiety and depression prevalence among the general US adult population despite an upward trend in the prevalence of anxiety and depression from 2019 to early 2023. Among different population groups, Phase 1 was associated with increases in anxiety and depression prevalence among Black/African Americans (2.26 percentage-point, 95% CI 0.24-4.28, p = 0.029), other non-Hispanic people of color (2.68 percentage-point, 95% CI 0.36-5.00, p = 0.024), and lower-income individuals (3.95 percentage-point, 95% CI 2.20-5.71, p < 0.0001).

INTERPRETATION: Our findings suggest disparate effects of phased COVID-19 vaccine rollout on mental health across US populations, underlining the need for careful planning in future strategies for phased disease prevention and interventions.

FUNDING: None.

PMID:39474466 | PMC:PMC11519686 | DOI:10.1016/j.lana.2024.100852

Eur Heart J Imaging Methods Pract. 2024 Oct 28;2(4):qyae086. doi: 10.1093/ehjimp/qyae086. eCollection 2024 Oct.

ABSTRACT

AIMS: To address the limitations of traditional diagnostic methods for mitral valve prolapse (MVP), specifically fibroelastic deficiency (FED) and Barlow's disease (BD), by introducing an automated diagnostic approach utilizing multi-view echocardiographic sequences and deep learning.

METHODS AND RESULTS: An echocardiographic data set, collected from Zhongshan Hospital, Fudan University, containing apical 2 chambers (A2C), apical 3 chambers (A3C), and apical 4 chambers (A4C) views, was employed to train the deep learning models. We separately trained view-specific and view-agnostic deep neural network models, which were denoted as MVP-VS and MVP view-agonistic (VA), for MVP diagnosis. Diagnostic accuracy, precision, sensitivity, F1-score, and specificity were evaluated for both BD and FED phenotypes. MVP-VS demonstrated an overall diagnostic accuracy of 0.94 for MVP. In the context of BD diagnosis, precision, sensitivity, F1-score, and specificity were 0.83, 1.00, 0.90, and 0.92, respectively. For FED diagnosis, the metrics were 1.00, 0.83, 0.91, and 1.00. MVP-VA exhibited an overall accuracy of 0.95, with BD-specific metrics of 0.85, 1.00, 0.92, and 0.94 and FED-specific metrics of 1.00, 0.83, 0.91, and 1.00. In particular, the MVP-VA model using mixed views for training demonstrated efficient diagnostic performance, eliminating the need for repeated development of MVP-VS models and improving the efficiency of the clinical pipeline by using arbitrary views in the deep learning model.

CONCLUSION: This study pioneers the integration of artificial intelligence into MVP diagnosis and demonstrates the effectiveness of deep neural networks in overcoming the challenges of traditional diagnostic methods. The efficiency and accuracy of the proposed automated approach suggest its potential for clinical applications in the diagnosis of valvular heart disease.

PMID:39474265 | PMC:PMC11519029 | DOI:10.1093/ehjimp/qyae086

Phys Imaging Radiat Oncol. 2024 Mar 22;30:100569. doi: 10.1016/j.phro.2024.100569. eCollection 2024 Apr.

ABSTRACT

BACKGROUND AND PURPOSE: Daily cone-beam computed tomography (CBCT) in image-guided radiotherapy administers radiation exposure and subjects patients to secondary cancer risk. Reducing imaging dose remains challenging as image quality deteriorates. We investigated three imaging dose levels by reducing projections and correcting images using two deep learning algorithms, aiming at identifying the lowest achievable imaging dose.

MATERIALS AND METHODS: CBCTs were reconstructed with 100%, 25%, 15% and 10% projections. Models were trained (30), validated (3) and tested (8) with prostate cancer patient datasets. We optimized and compared the performance of 1) a cycle generative adversarial network (cycleGAN) with residual connection and 2) a contrastive unpaired translation network (CUT) to generate synthetic computed tomography (sCT) from reduced imaging dose CBCTs. Volumetric modulated arc therapy plans were optimized on a reference intensity-corrected full dose CBCTcor and recalculated on sCTs. Hounsfield unit (HU) and positioning accuracy were evaluated. Bladder and rectum were manually delineated to determine anatomical fidelity.

RESULTS: All sCTs achieved average mean absolute mean absolute error/structural similarity index measure/peak signal-to-noise ratio of ⩽ 59HU/ ⩾ 0.94/ ⩾ 33 dB. All dose-volume histogram parameter differences were within 2 Gy or 2 % . Positioning differences were ⩽ 0.30 mm or 0.30°. cycleGAN with Dice similarity coefficients (DSC) for bladder/rectum of ⩾ 0.85/ ⩾ 0.81 performed better than CUT ( ⩾ 0.83/ ⩾ 0.76). A significantly lower DSC accuracy was observed for 15 % and 10 % sCTs. cycleGAN performed better than CUT for contouring, however both yielded comparable outcomes in other evaluations.

CONCLUSION: sCTs based on different CBCT doses using cycleGAN and CUT were investigated. Based on segmentation accuracy, 25 % is the minimum imaging dose.

PMID:39474260 | PMC:PMC11519690 | DOI:10.1016/j.phro.2024.100569

Osteology (Basel). 2024 Jun;4(2):98-110. doi: 10.3390/osteology4020008. Epub 2024 Jun 11.

ABSTRACT

Evaluation of 3D bone morphology of the glenohumeral joint is necessary for pre-surgical planning. Zero echo time (ZTE) magnetic resonance imaging (MRI) provides excellent bone contrast and can potentially be used in place of computed tomography. Segmentation of shoulder anatomy, particularly humeral head and acetabulum, is needed for detailed assessment of each anatomy and for pre-surgical preparation. In this study we compared performance of two popular deep learning models based on Google's DeepLab and U-Net to perform automated segmentation on ZTE MRI of human shoulders. Axial ZTE images of normal shoulders (n=31) acquired at 3-Tesla were annotated for training with a DeepLab and 2D U-Net, and the trained model was validated with testing data (n=13). While both models showed visually satisfactory results for segmenting the humeral bone, U-Net slightly over-estimated while DeepLab under-estimated the segmented area compared to the ground truth. Testing accuracy quantified by Dice score was significantly higher (p<0.05) for U-Net (88%) than DeepLab (81%) for the humeral segmentation. We have also implemented the U-Net model onto an MRI console for a push-button DL segmentation processing. Although this is an early work with limitations, our approach has the potential to improve shoulder MR evaluation hindered by manual post-processing and may provide clinical benefit for quickly visualizing bones of the glenohumeral joint.

PMID:39474235 | PMC:PMC11520815 | DOI:10.3390/osteology4020008

Front Surg. 2024 Oct 15;11:1329085. doi: 10.3389/fsurg.2024.1329085. eCollection 2024.

ABSTRACT

INTRODUCTION: This study presents the development and validation of a Deep Learning Convolutional Neural Network (CNN) model for estimating acetabular version (AV) from native hip plain radiographs.

METHODS: Utilizing a dataset comprising 300 participants with unrelated pelvic complaints, the CNN model was trained and evaluated against CT-Scans, considered the gold standard, using a 5-fold cross-validation.

RESULTS: Notably, the CNN model exhibited a robust performance, demonstrating a strong Pearson correlation with CT-Scans (right hip: r = 0.70, p < 0.001; left hip: r = 0.71, p < 0.001) and achieving a mean absolute error of 2.95°. Remarkably, over 83% of predictions yielded errors ≤5°, highlighting the model's high precision in AV estimation.

DISCUSSION: The model holds promise in preoperative planning for hip arthroplasty, potentially reducing complications like recurrent dislocation and component wear. Future directions include further refinement of the CNN model, with ongoing investigations aimed at enhancing preoperative planning potential and ensuring comprehensive assessment across diverse patient populations, particularly in diseased cases. Additionally, future research could explore the model's potential value in scenarios necessitating minimized ionizing radiation exposure, such as post-operative evaluations.

PMID:39474228 | PMC:PMC11518832 | DOI:10.3389/fsurg.2024.1329085

Neurophotonics. 2024 Oct;11(4):045007. doi: 10.1117/1.NPh.11.4.045007. Epub 2024 Oct 29.

ABSTRACT

SIGNIFICANCE: Voltage imaging is a powerful tool for studying the dynamics of neuronal activities in the brain. However, voltage imaging data are fundamentally corrupted by severe Poisson noise in the low-photon regime, which hinders the accurate extraction of neuronal activities. Self-supervised deep learning denoising methods have shown great potential in addressing the challenges in low-photon voltage imaging without the need for ground-truth but usually suffer from the trade-off between spatial and temporal performances.

AIM: We present DeepVID v2, a self-supervised denoising framework with decoupled spatial and temporal enhancement capability to significantly augment low-photon voltage imaging.

APPROACH: DeepVID v2 is built on our original DeepVID framework, which performs frame-based denoising by utilizing a sequence of frames around the central frame targeted for denoising to leverage temporal information and ensure consistency. Similar to DeepVID, the network further integrates multiple blind pixels in the central frame to enrich the learning of local spatial information. In addition, DeepVID v2 introduces a new spatial prior extraction branch to capture fine structural details to learn high spatial resolution information. Two variants of DeepVID v2 are introduced to meet specific denoising needs: an online version tailored for real-time inference with a limited number of frames and an offline version designed to leverage the full dataset, achieving optimal temporal and spatial performances.

RESULTS: We demonstrate that DeepVID v2 is able to overcome the trade-off between spatial and temporal performances and achieve superior denoising capability in resolving both high-resolution spatial structures and rapid temporal neuronal activities. We further show that DeepVID v2 can generalize to different imaging conditions, including time-series measurements with various signal-to-noise ratios and extreme low-photon conditions.

CONCLUSIONS: Our results underscore DeepVID v2 as a promising tool for enhancing voltage imaging. This framework has the potential to generalize to other low-photon imaging modalities and greatly facilitate the study of neuronal activities in the brain.

PMID:39474199 | PMC:PMC11519979 | DOI:10.1117/1.NPh.11.4.045007

Acad Radiol. 2024 Oct 28:S1076-6332(24)00767-0. doi: 10.1016/j.acra.2024.10.007. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: Accurate prediction of microvascular invasion (MVI) in hepatocellular carcinoma (HCC) is crucial for guiding treatment. This study evaluates and compares the performance of clinicoradiologic, traditional radiomics, deep-learning radiomics, feature fusion, and decision fusion models based on multi-region MR habitat imaging using six machine-learning classifiers.

MATERIALS AND METHODS: We retrospectively included 300 HCC patients. The intratumoral and peritumoral regions were segmented into distinct habitats, from which radiomics and deep-learning features were extracted using arterial phase MR images. To reduce feature dimensionality, we applied intra-class correlation coefficient (ICC) analysis, Pearson correlation coefficient (PCC) filtering, and recursive feature elimination (RFE). Based on the selected optimal features, prediction models were constructed using decision tree (DT), K-nearest neighbors (KNN), logistic regression (LR), random forest (RF), support vector machine (SVM), and XGBoost (XGB) classifiers. Additionally, fusion models were developed utilizing both feature fusion and decision fusion strategies. The performance of these models was validated using the area under the receiver operating characteristic curve (ROC AUC), calibration curves, and decision curve analysis.

RESULTS: The decision fusion model (VOI-Peri10-1) using LR and integrating clinicoradiologic, radiomics, and deep-learning features achieved the highest AUC of 0.808 (95% confidence interval [CI]: 0.807-0.912) in the test cohort, with good calibration (Hosmer-Lemeshow test, P > 0.050) and clinical net benefit.

CONCLUSION: The LR-based decision fusion model integrating clinicoradiologic, radiomics, and deep-learning features shows promise for preoperative prediction of MVI in HCC, aiding in patient outcome predictions and personalized treatment planning.

PMID:39472207 | DOI:10.1016/j.acra.2024.10.007

Genome Res. 2024 Oct 29:gr.279307.124. doi: 10.1101/gr.279307.124. Online ahead of print.

ABSTRACT

The critical stage of every de novo genome assembler is identifying paths in assembly graphs that correspond to the reconstructed genomic sequences. The existing algorithmic methods struggle with this, primarily due to repetitive regions causing complex graph tangles, leading to fragmented assemblies. Here, we introduce GNNome, a framework for path identification based on geometric deep learning that enables training models on assembly graphs without relying on existing assembly strategies. By leveraging only the symmetries inherent to the problem, GNNome reconstructs assemblies from PacBio HiFi reads with contiguity and quality comparable to those of the state-of-the-art tools across several species. With every new genome assembled telomere-to-telomere, the amount of reliable training data at our disposal increases. Combining the straightforward generation of abundant simulated data for diverse genomic structures with the AI approach makes the proposed framework a plausible cornerstone for future work on reconstructing complex genomes with different ploidy and aneuploidy degrees. To facilitate such developments, we make the framework and the best-performing model publicly available, provided as a tool that can directly be used to assemble new haploid genomes.

PMID:39472021 | DOI:10.1101/gr.279307.124

Radiol Technol. 2024 Nov;96(2):99-107.

ABSTRACT

PURPOSE: To investigate the effects of deep learning reconstruction on depicting arteries and providing suitable images for the evaluation of hemorrhages with abdominopelvic contrast-enhanced computed tomography (CT) compared with hybrid iterative reconstruction.

METHODS: This retrospective study included 16 patients (mean age: 54.2 ± 22.1 years; 8 men and 8 women) with acute hemorrhage who underwent contrast-enhanced CT. Unenhanced axial, arterial phase axial, arterial phase coronal, and delayed phase axial images were reconstructed with deep learning reconstruction, hybrid iterative reconstruction, and filtered back projection, which was used as a control in qualitative analyses. Circular and line regions of interest were placed on the aorta and superior mesenteric artery (SMA), respectively, in quantitative analyses. Using a blind process, 2 radiologists independently evaluated image noise, depiction of arteries, and suitability for the evaluation of hemorrhage in qualitative image analyses.

RESULTS: Image noise in deep learning reconstruction was significantly reduced compared with hybrid iterative reconstruction in the quantitative (P < .001) and qualitative analyses (Reader 1, P ≤ .001 for all series; Reader 2, P = .002, .001, and < .001). The slope at the half maximum in deep learning reconstruction (123.8 ± 63.2 HU/mm) significantly improved compared with hybrid iterative reconstruction (105.3 ± 51.0 HU/mm) in the CT attenuation profile of the SMA (P < .001). Qualitative analyses revealed a significantly improved depiction of arteries (Reader 1, P < .001 for all series; Reader 2, P = .037, .008, and < .001) and suitability for evaluating acute hemorrhage in the arterial phase image (Reader 1, P < .001 for both series; Reader 2, P = .041 and .004) with deep learning reconstruction compared with hybrid iterative reconstruction.

DISCUSSION: Deep learning reconstruction provided images with a significantly better depiction of arteries and more suitable quality arterial phase images for the evaluation of abdominopelvic hemorrhage compared with hybrid iterative reconstruction.

CONCLUSION: Deep learning reconstruction is better for reconstructing abdominopelvic contrast-enhanced CT images when evaluating hemorrhages; however, a prospective study including a large number of patients is needed to strengthen the findings of this study.

PMID:39472011

Sci Total Environ. 2024 Oct 27:177183. doi: 10.1016/j.scitotenv.2024.177183. Online ahead of print.

ABSTRACT

Deep learning models are widely used for PM2.5 prediction. However, neglecting temporal and spatial characteristics leads to low prediction accuracy. In this work, a new deep learning model (RCG - Attention model) was developed, which combines the residual neural network (ResNet) and the convolution gated recurrent network (ConvGRU) and is applied to extract the spatio - temporal features for predicting PM2.5 concentration over the subsequent 24 h. The ResNet extracts the spatial distribution features of pollutants, and the ConvGRU extracts temporal features. The spatial and temporal features are fused by the multi - head attention mechanism to obtain multi - dimensional features. These features are finally fed into a series of fully connected layers to predict the future results. Incorporating these chemical components enhances the scientific validity of the dataset and strengthens the inherent logical connections among variables. The Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and R - squared (R2) results indicate that the prediction performance of the RCG - Attention model surpasses that of other baseline models. The model demonstrates superior prediction performance across multiple monitoring stations, suggesting robust generalization capabilities and adaptability for various regions in one city. The SHAP results show that PM10, NO2, RH, NO3-, OC and NH4+ are significant influencing features. The RCG - Attention model provides a comprehensive solution for PM2.5 concentration prediction by integrating spatial and temporal feature extraction with chemical components.

PMID:39471939 | DOI:10.1016/j.scitotenv.2024.177183

Caries Res. 2024 Oct 29:1-21. doi: 10.1159/000542289. Online ahead of print.

ABSTRACT

INTRODUCTION: Despite the notable progress in developing artificial intelligence (AI)-based tools for caries detection in bitewings, limited research has addressed the detection and staging of secondary caries. Therefore, we aimed to develop a Convolutional neural network (CNN)-based algorithm for these purposes using a novel approach for determining lesion severity.

METHODS: We used a dataset from a Dutch dental practice-based research network containing 2,612 restored teeth in 413 bitewings from 383 patients aged 15 to 88 years and trained the Mask R-CNN architecture with a Swin Transformer backbone. Two-stage training fine-tuned caries detection accuracy and severity assessment. Annotations of caries around restorations were made by two evaluators and checked by two other experts. Aggregated accuracy metrics (mean ± Standard Deviation - SD) in detecting teeth with secondary caries were calculated considering two thresholds: detecting all lesions and dentine lesions. The correlation between the lesion severity scores obtained with the algorithm and the annotators' consensus was determined using the Pearson correlation coefficient and Bland-Altman plots.

RESULTS: Our refined algorithm showed high specificity in detecting all lesions (0.966 ± 0.025) and dentine lesions (0.964 ± 0.019). Sensitivity values were lower: 0.737 ± 0.079 for all lesions and 0.808 ± 0.083 for dentine lesions. The areas under ROC curves (SD) were 0.940 (0.025) for all lesions and 0.946 (0.023) for dentine lesions. The correlation coefficient for severity scores was 0.802.

CONCLUSION: We developed an improved algorithm to support clinicians in detecting and staging secondary caries in bitewing, incorporating an innovative approach for annotation, considering the lesion severity as a continuous outcome.

PMID:39471790 | DOI:10.1159/000542289

Comput Biol Med. 2024 Oct 28;183:109274. doi: 10.1016/j.compbiomed.2024.109274. Online ahead of print.

ABSTRACT

Automatic segmentation of breast tumor ultrasound images can provide doctors with objective and efficient references for lesions and regions of interest. Both dataset optimization and model structure optimization are crucial for achieving optimal image segmentation performance, and it can be challenging to satisfy the clinical needs solely through model structure enhancements in the context of insufficient breast tumor ultrasound datasets for model training. While significant research has focused on enhancing the architecture of deep learning models to improve tumor segmentation performance, there is a relative paucity of work dedicated to dataset augmentation. Current data augmentation techniques, such as rotation and transformation, often yield insufficient improvements in model accuracy. The deep learning methods used for generating synthetic images, such as GANs is primarily applied to produce visually natural-looking images. Nevertheless, the accuracy of the labels for these generated images still requires manual verification, and the images exhibit a lack of diversity. Therefore, they are not suitable for the training datasets augmentation of image segmentation models. This study introduces a novel dataset augmentation approach that generates synthetic images by embedding tumor regions into normal images. We explore two synthetic methods: one using identical backgrounds and another with varying backgrounds. Through experimental validation, we demonstrate the efficiency of the synthetic datasets in enhancing the performance of image segmentation models. Notably, the synthetic method utilizing different backgrounds exhibits superior improvement compared to the identical background approach. Our findings contribute to medical image analysis, particularly in tumor segmentation, by providing a practical and effective dataset augmentation strategy that can significantly improve the accuracy and reliability of segmentation models.

PMID:39471661 | DOI:10.1016/j.compbiomed.2024.109274

Prev Vet Med. 2024 Oct 9;233:106354. doi: 10.1016/j.prevetmed.2024.106354. Online ahead of print.

ABSTRACT

Bovine Respiratory Disease (BRD) is a prevalent infectious disease of respiratory tract in cattle, presenting challenges in accurate diagnosis and forecasting due to the complex interactions of multiple risk factors. Common methods, including mathematical epidemiological models and data-driven approaches such as machine learning models, face limitations such as difficult parameter estimation or the need for data across all potential outcomes, which is challenging given the scarcity and noise in observing BRD processes. In response to these challenges, we introduce a novel approach known as the Bayesian Deep Mechanistic method. This method couples a data-driven model with a mathematical epidemiological model while accounting for uncertainties within the processes. By utilising 265 lung ultrasound videos as sensor data from 163 animals across 9 farms in France, we trained a Bayesian deep learning model to predict the infection status (infected or non-infected) of an entire batch of 12 animals, also providing associated confidence levels. These predictions, coupled with their confidence levels, were used to filter out highly uncertain diagnoses and diffuse uncertainties into the parameterisation of a mathematical epidemiological model to forecast the progression of infected animals. Our findings highlight that considering the confidence levels (or uncertainties) of predictions enhances both the diagnosis and forecasting of BRD. Uncertainty-aware diagnosis reduced errors to 32 %, outperforming traditional automatic diagnosis. Forecast relying on veterinarian diagnoses, considered the most confident, had a 23 % error, whilst forecast taking into account diagnosis uncertainties had a close 27.2 % error. Building upon uncertainty-awareness, our future research could explore integrating multiple types of sensor data, such as video surveillance, audio recordings, and environmental parameters, to provide a comprehensive evaluation of animal health, employing multi-modal methods for processing this diverse data.

PMID:39471650 | DOI:10.1016/j.prevetmed.2024.106354