Genome Res. 2024 Nov 19:gr.279694.124. doi: 10.1101/gr.279694.124. Online ahead of print.

ABSTRACT

Polygenic risk score (PRS) is a widely-used approach for predicting individuals' genetic risk of complex diseases, playing a pivotal role in advancing precision medicine. Traditional PRS methods, predominantly following a linear structure, often fall short in capturing the intricate relationships between genotype and phenotype. In this study, we present PRS-Net, an interpretable geometric deep learning-based framework that effectively models the nonlinearity of biological systems for enhanced disease prediction and biological discovery. PRS-Net begins by deconvoluting the genome-wide PRS at the single-gene resolution, and then explicitly encapsulates gene-gene interactions leveraging a graph neural network (GNN) for genetic risk prediction, enabling a systematic characterization of molecular interplay underpinning diseases. An attentive readout module is introduced to facilitate model interpretation. Extensive tests across multiple complex traits and diseases demonstrate the superior prediction performance of PRS-Net compared to conventional PRS methods. The interpretability of PRS-Net further enhances the identification of disease-relevant genes and gene programs. PRS-Net provides a potent tool for concurrent genetic risk prediction and biological discovery for complex diseases.

PMID:39562137 | DOI:10.1101/gr.279694.124

Am J Hum Genet. 2024 Nov 13:S0002-9297(24)00380-X. doi: 10.1016/j.ajhg.2024.10.011. Online ahead of print.

ABSTRACT

There has been considerable progress in building models to predict the effect of missense substitutions in protein-coding genes, fueled in large part by progress in applying deep learning methods to sequence data. These models have the potential to enable clinical variant annotation on a large scale and hence increase the impact of patient sequencing in guiding diagnosis and treatment. To realize this potential, it is essential to provide reliable assessments of model performance, scope of applicability, and robustness. As a response to this need, the ClinGen Sequence Variant Interpretation Working Group, Pejaver et al., recently proposed a strategy for validation and calibration of in-silico predictions in the context of guidelines for variant annotation. While this work marks an important step forward, the strategy presented still has important limitations. We propose core principles and recommendations to overcome these limitations that can enable both more reliable and more impactful use of variant effect prediction models in the future.

PMID:39561772 | DOI:10.1016/j.ajhg.2024.10.011

Talanta. 2024 Nov 8;284:127192. doi: 10.1016/j.talanta.2024.127192. Online ahead of print.

ABSTRACT

LIBS-based analysis has experienced an ever-increasing interest in recent years as a well-suited technique for chemical analysis tasks relying on elemental fingerprinting. This method stands out for its ability to offer rapid, simultaneous multi-element analysis with the advantage of portability. In the context of this research, our aim is to bridge the gap between the analysis of simulated and real data to better account for variations in plasma temperature and electron density, which are typically not considered in LIBS analysis. To achieve this, we employ two distinct methodologies, PLS and CNNs, to construct predictive models for predicting the concentration of the 24 elements within each LIBS spectrum. The initial phase of our investigation concentrates on the training and testing of these models using simulated LIBS data, with results evaluated through RMSEP values. The IQR and median RMSEP values for all the elements demonstrate that CNNs consistently achieved values below 0.01, while PLS results ranged from 0.01 to 0.05, highlighting the superior stability and predictive accuracy of CNNs model. In the next phase, we applied the pre-trained models to analyze the real LIBS spectra, consistently identifying Aluminum (Al), Iron (Fe), and Silicon (Si) as having the highest predicted concentrations. The overall predicted values were approximately 0.5 for Al, 0.6 for Si, and 0.04 for Fe. In the third phase, deliberate adjustments are made to the training parameters and architecture of the proposed CNNs model to force the network to emphasize specific elements, prioritizing them over other components present in each real LIBS spectrum. The generation of the three modified versions of the initially proposed CNNs allows us to explore the impact of regularization, sample weighting, and a customized loss function on prediction outcomes. Some elements emerge during the prediction phase, with Calcium (Ca), Magnesium (Mg), Zinc (Zn), Titanium (Ti), and Gallium (Ga) exhibiting more pronounced patterns.

PMID:39561618 | DOI:10.1016/j.talanta.2024.127192

Comput Biol Chem. 2024 Nov 15;113:108282. doi: 10.1016/j.compbiolchem.2024.108282. Online ahead of print.

ABSTRACT

Although bioinformatics-based methods accurately identify SEs (Super-enhancers), the results depend on feature design. It is foundational to representing biological sequences and automatically extracting their key features for improving SE identification. We propose a deep learning model MuSE (Multi-Feature Fusion for Super-Enhancer), based on multi-feature fusion. This model utilizes two encoding methods, one-hot and DNA2Vec, to signify DNA sequences. Specifically, one-hot encoding reflects single nucleotide information, while k-mer representations based on DNA2Vec capture both local sequence fragment information and global sequence characteristics. These types of feature vectors are conducted and combined by neural networks, which aim at SE prediction. To validate the effectiveness of MuSE, we design extensive experiments on human and mouse species datasets. Compared to baselines such as SENet, MuSE improves the prediction of F1 score to a maximum improvement exceeding 0.05 on mouse species. The k-mer representations based on DNA2Vec among the given features have the most important impact on predictions. This feature effectively captures context semantic knowledge and positional information of DNA sequences. However, its representation of the individuality of each species negatively affects MuSE's generalization ability. Nevertheless, the cross-species prediction results of MuSE improve again to reach an AUC of nearly 0.8, after removing this type of feature. Source codes are available at https://github.com/15831959673/MuSE.

PMID:39561516 | DOI:10.1016/j.compbiolchem.2024.108282

J Med Internet Res. 2024 Nov 19;26:e58892. doi: 10.2196/58892.

ABSTRACT

BACKGROUND: Current blood glucose monitoring (BGM) methods are often invasive and require repetitive pricking of a finger to obtain blood samples, predisposing individuals to pain, discomfort, and infection. Noninvasive blood glucose monitoring (NIBGM) is ideal for minimizing discomfort, reducing the risk of infection, and increasing convenience.

OBJECTIVE: This review aimed to map the use cases of artificial intelligence (AI) in NIBGM.

METHODS: A systematic scoping review was conducted according to the Arksey O'Malley five-step framework. Eight electronic databases (CINAHL, Embase, PubMed, Web of Science, Scopus, The Cochrane-Central Library, ACM Digital Library, and IEEE Xplore) were searched from inception until February 8, 2023. Study selection was conducted by 2 independent reviewers, descriptive analysis was conducted, and findings were presented narratively. Study characteristics (author, country, type of publication, study design, population characteristics, mean age, types of noninvasive techniques used, and application, as well as characteristics of the BGM systems) were extracted independently and cross-checked by 2 investigators. Methodological quality appraisal was conducted using the Checklist for assessment of medical AI.

RESULTS: A total of 33 papers were included, representing studies from Asia, the United States, Europe, the Middle East, and Africa published between 2005 and 2023. Most studies used optical techniques (n=19, 58%) to estimate blood glucose levels (n=27, 82%). Others used electrochemical sensors (n=4), imaging (n=2), mixed techniques (n=2), and tissue impedance (n=1). Accuracy ranged from 35.56% to 94.23% and Clarke error grid (A+B) ranged from 86.91% to 100%. The most popular machine learning algorithm used was random forest (n=10) and the most popular deep learning model was the artificial neural network (n=6). The mean overall checklist for assessment of medical AI score on the included papers was 33.5 (SD 3.09), suggesting an average of medium quality. The studies reviewed demonstrate that some AI techniques can accurately predict glucose levels from noninvasive sources while enhancing comfort and ease of use for patients. However, the overall range of accuracy was wide due to the heterogeneity of models and input data.

CONCLUSIONS: Efforts are needed to standardize and regulate the use of AI technologies in BGM, as well as develop consensus guidelines and protocols to ensure the quality and safety of AI-assisted monitoring systems. The use of AI for NIBGM is a promising area of research that has the potential to revolutionize diabetes management.

PMID:39561353 | DOI:10.2196/58892

JCO Clin Cancer Inform. 2024 Nov;8:e2400110. doi: 10.1200/CCI.24.00110. Epub 2024 Nov 19.

ABSTRACT

PURPOSE: Population-based cancer registries (PBCRs) collect data on all new cancer diagnoses in a defined population. Data are sourced from pathology reports, and the PBCRs rely on manual and rule-based solutions. This study presents a state-of-the-art natural language processing (NLP) pipeline, built by fine-tuning pretrained language models (LMs). The pipeline is deployed at the British Columbia Cancer Registry (BCCR) to detect reportable tumors from a population-based feed of electronic pathology.

METHODS: We fine-tune two publicly available LMs, GatorTron and BlueBERT, which are pretrained on clinical text. Fine-tuning is done using BCCR's pathology reports. For the final decision making, we combine both models' output using an OR approach. The fine-tuning data set consisted of 40,000 reports from the diagnosis year of 2021, and the test data sets consisted of 10,000 reports from the diagnosis year 2021, 20,000 reports from diagnosis year 2022, and 400 reports from diagnosis year 2023.

RESULTS: The retrospective evaluation of our proposed approach showed boosted reportable accuracy, maintaining the true reportable threshold of 98%.

CONCLUSION: Disadvantages of rule-based NLP in cancer surveillance include manual effort in rule design and sensitivity to language change. Deep learning approaches demonstrate superior performance in classification. PBCRs distinguish reportability status of incoming electronic cancer pathology reports. Deep learning methods provide significant advantages over rule-based NLP.

PMID:39561305 | DOI:10.1200/CCI.24.00110

Clin Cancer Res. 2024 Nov 19. doi: 10.1158/1078-0432.CCR-24-2599. Online ahead of print.

ABSTRACT

PURPOSE: Necrosis quantification in the neoadjuvant setting using pathology slide review is the most important validated prognostic marker in conventional osteosarcoma. Herein, we explored three deep learning strategies on histology samples to predict outcome for OSA in the neoadjuvant setting.

EXPERIMENTAL DESIGN: Our study relies on a training cohort from New York University (New York, NY) and an external cohort from Charles university (Prague, Czechia). We trained and validated the performance of a supervised approach that integrates neural network predictions of necrosis/tumor content, and compared predicted overall survival (OS) using Kaplan-Meier curves. Furthermore, we explored morphology-based supervised and self-supervised approaches to determine whether intrinsic histomorphological features could serve as a potential marker for OS in the setting of neoadjuvant.

RESULTS: Excellent correlation between the trained network and the pathologists was obtained for the quantification of necrosis content (R2=0.899, r=0.949, p < 0.0001). OS prediction cutoffs were consistent between pathologists and the neural network (22% and 30% of necrosis, respectively). Morphology-based supervised approach predicted OS with p-value=0.0028, HR=2.43 [1.10-5.38]. The self-supervised approach corroborated the findings with clusters enriched in necrosis, fibroblastic stroma, and osteoblastic morphology associating with better OS (lg2HR; -2.366; -1.164; -1.175; 95% CI=[-2.996; -0.514]). Viable/partially viable tumor and fat necrosis were associated with worse OS (lg2HR;1.287;0.822;0.828; 95% CI=[0.38-1.974]).

CONCLUSIONS: Neural networks can be used to automatically estimate the necrosis to tumor ratio, a quantitative metric predictive of survival. Furthermore, we identified alternate histomorphological biomarkers specific to the necrotic and tumor regions themselves which can be used as predictors.

PMID:39561274 | DOI:10.1158/1078-0432.CCR-24-2599

PLoS Comput Biol. 2024 Nov 19;20(11):e1012597. doi: 10.1371/journal.pcbi.1012597. Online ahead of print.

ABSTRACT

Predicting virus-host associations is essential to determine the specific host species that viruses interact with, and discover if new viruses infect humans and animals. Currently, the host of the majority of viruses is unknown, particularly in microbiomes. To address this challenge, we introduce EvoMIL, a deep learning method that predicts the host species for viruses from viral sequences only. It also identifies important viral proteins that significantly contribute to host prediction. The method combines a pre-trained large protein language model (ESM) and attention-based multiple instance learning to allow protein-orientated predictions. Our results show that protein embeddings capture stronger predictive signals than sequence composition features, including amino acids, physiochemical properties, and DNA k-mers. In multi-host prediction tasks, EvoMIL achieves median F1 score improvements of 10.8%, 16.2%, and 4.9% in prokaryotic hosts, and 1.7%, 6.6% and 11.5% in eukaryotic hosts. EvoMIL binary classifiers achieve impressive AUC over 0.95 for all prokaryotic hosts and range from roughly 0.8 to 0.9 for eukaryotic hosts. Furthermore, EvoMIL identifies important proteins in the prediction task. We found them capturing key functions in virus-host specificity.

PMID:39561204 | DOI:10.1371/journal.pcbi.1012597

PLoS One. 2024 Nov 19;19(11):e0313145. doi: 10.1371/journal.pone.0313145. eCollection 2024.

ABSTRACT

Ship detection over long distances is crucial for the visual perception of intelligent ships. Since traditional image processing-based methods are not robust, deep learning-based image recognition methods can automatically obtain the features of small ships. However, due to the limited pixels of ships over long distances, accurate features of such ships are difficult to obtain. To address this, a two-stage object detection method that combines the advantages of traditional and deep-learning methods is proposed. In the first stage, an object detection model for the sea-sky line (SSL) region is trained to select a potential region of ships. In the second stage, another object detection model for ships is trained using sliced patches containing ships. When testing, the SSL region is first detected using the trained 8th version of You Only Look Once (YOLOv8). Then, the SSL region detected is divided into several overlapping patches using the slicing technique, and another trained YOLOv8 is applied to detect ships. The experimental results showed that our method achieved 85% average precision when the intersection over union is 0.5 (AP50), and a detection speed of 75 ms per image with a pixel size of 1080×640. The code is available at https://github.com/gongyanfeng/PaperCode.

PMID:39561153 | DOI:10.1371/journal.pone.0313145

Med Biol Eng Comput. 2024 Nov 19. doi: 10.1007/s11517-024-03235-4. Online ahead of print.

ABSTRACT

The current clinical gold standard to assess the condition and detect loosening of pedicle screw implants is radiation-emitting medical imaging. However, solely based on medical imaging, clinicians are not able to reliably identify loose implants in a substantial amount of cases. To complement medical imaging for pedicle screw loosening detection, we propose a new methodology and paradigm for the radiation-free, non-destructive, and easy-to-integrate loosening detection based on vibroacoustic sensing. For the detection of a loose implant, we excite the vertebra of interest with a sine sweep vibration at the spinous process and use a custom highly sensitive piezo vibration sensor attached directly at the screw head to capture the propagated vibration characteristics which are analyzed using a detection pipeline based on spectrogram features and a SE-ResNet-18. To validate the proposed approach, we propose a novel, biomechanically validated simulation technique for pedicle screw loosening, conduct experiments using four human cadaveric lumbar spine specimens, and evaluate our algorithm in a cross-validation experiment. The proposed method reaches a sensitivity of 91.50 ± 6.58 % and a specificity of 91.10 ± 2.27 % for pedicle screw loosening detection.

PMID:39560918 | DOI:10.1007/s11517-024-03235-4

Med Biol Eng Comput. 2024 Nov 19. doi: 10.1007/s11517-024-03245-2. Online ahead of print.

ABSTRACT

Sepsis poses a major global health threat, accounting for millions of deaths annually and significant economic costs. Accurately predicting the risk of mortality in sepsis patients enables early identification, promotes the efficient allocation of medical resources, and facilitates timely interventions, thereby improving patient outcomes. Current methods typically utilize only one type of data-either constant, temporal, or ICD codes. This study introduces a novel approach, the Time-Constant Kolmogorov-Arnold Network (TCKAN), which uniquely integrates temporal data, constant data, and ICD codes within a single predictive model. Unlike existing methods that typically rely on one type of data, TCKAN leverages a multi-modal data integration strategy, resulting in superior predictive accuracy and robustness in identifying high-risk sepsis patients. Validated against the MIMIC-III and MIMIC-IV datasets, TCKAN surpasses existing machine learning and deep learning methods in accuracy, sensitivity, and specificity. Notably, TCKAN achieved AUCs of 87.76% and 88.07%, demonstrating superior capability in identifying high-risk patients. Additionally, TCKAN effectively combats the prevalent issue of data imbalance in clinical settings, improving the detection of patients at elevated risk of mortality and facilitating timely interventions. These results confirm the model's effectiveness and its potential to transform patient management and treatment optimization in clinical practice. Although the TCKAN model has already incorporated temporal, constant, and ICD code data, future research could include more diverse medical data types, such as imaging and laboratory test results, to achieve a more comprehensive data integration and further improve predictive accuracy.

PMID:39560917 | DOI:10.1007/s11517-024-03245-2

CJEM. 2024 Nov 19. doi: 10.1007/s43678-024-00807-z. Online ahead of print.

ABSTRACT

STUDY OBJECTIVE: This study investigates the potential to improve emergency department (ED) triage using machine learning models by comparing their predictive performance with the Canadian Triage Acuity Scale (CTAS) in identifying the need for critical care within 12 h of ED arrival.

METHODS: Three machine learning models (LASSO regression, gradient-boosted trees, and a deep learning model with embeddings) were developed using retrospective data from 670,841 ED visits to the Jewish General Hospital from June 2012 to Jan 2021. The model outcome was the need for critical care within the first 12 h of ED arrival. Metrics, including the areas under the receiver-operator characteristic curve (ROC) and precision-recall curve (PRC) were used for performance evaluation. Shapley additive explanation scores were used to compare predictor importance.

RESULTS: The three machine learning models (deep learning, gradient-boosted trees and LASSO regression) had areas under the ROC of 0.926 ± 0.003, 0.912 ± 0.003 and 0.892 ± 0.004 respectively, and areas under the PRC of 0.27 ± 0.01, 0.24 ± 0.01 and 0.23 ± 0.01 respectively. In comparison, the CTAS score had an area under the ROC of 0.804 ± 0.006 and under the PRC of 0.11 ± 0.01. The predictors of most importance were similar between the models.

CONCLUSIONS: Machine learning models outperformed CTAS in identifying, at the point of ED triage, patients likely to need early critical care. If validated in future studies, machine learning models such as the ones developed here may be considered for incorporation in future revisions of the CTAS triage algorithm, potentially improving discrimination and reliability.

PMID:39560909 | DOI:10.1007/s43678-024-00807-z

Interdiscip Sci. 2024 Nov 19. doi: 10.1007/s12539-024-00670-7. Online ahead of print.

ABSTRACT

Yinchenhao Decoction (YCHD), a classic formula in traditional Chinese medicine, is believed to have the potential to treat liver diseases by modulating the Toll-like receptor 4 (TLR4) target. Therefore, a thorough exploration of the effective components and therapeutic mechanisms targeting TLR4 in YCHD is a promising strategy for liver diseases. In this study, the AIGO-DTI deep learning framework was proposed to predict the targeting probability of major components in YCHD for TLR4. Comparative evaluations with four machine learning models (RF, SVM, KNN, XGBoost) and two deep learning models (GCN, GAT) demonstrated that the AIGO-DTI framework exhibited the best overall performance, with Recall and AUC reaching 0.968 and 0.991, respectively.This study further utilized the AIGO-DTI model to identify the potential impact of Isoscopoletin, a major component of YCHD, on TLR4. Subsequent wet experiments revealed that Isoscopoletin could influence the maturation of Dendritic Cells (DCs) induced by Lipopolysaccharide (LPS) through TLR4, suggesting its therapeutic potential for liver diseases, especially hepatitis. Additionally, based on the AIGO-DTI framework, this study established an online platform named TLR4-Predict to facilitate domain experts in discovering more compounds related to TLR4. Overall, the proposed AIGO-DTI framework accurately predicts unique compounds in YCHD that interact with TLR4, providing new insights for identifying and screening lead compounds targeting TLR4.

PMID:39560852 | DOI:10.1007/s12539-024-00670-7

Eur Radiol Exp. 2024 Nov 19;8(1):130. doi: 10.1186/s41747-024-00529-y.

ABSTRACT

Feature extraction and selection from medical data are the basis of radiomics and image biomarker discovery for various architectures, including convolutional neural networks (CNNs). We herein describe the typical radiomics steps and the components of a CNN for both deep feature extraction and end-to-end approaches. We discuss the curse of dimensionality, along with dimensionality reduction techniques. Despite the outstanding performance of deep learning (DL) approaches, the use of handcrafted features instead of deep learned features needs to be considered for each specific study. Dataset size is a key factor: large-scale datasets with low sample diversity could lead to overfitting; limited sample sizes can provide unstable models. The dataset must be representative of all the "facets" of the clinical phenomenon/disease investigated. The access to high-performance computational resources from graphics processing units is another key factor, especially for the training phase of deep architectures. The advantages of multi-institutional federated/collaborative learning are described. When large language models are used, high stability is needed to avoid catastrophic forgetting in complex domain-specific tasks. We highlight that non-DL approaches provide model explainability superior to that provided by DL approaches. To implement explainability, the need for explainable AI arises, also through post hoc mechanisms. RELEVANCE STATEMENT: This work aims to provide the key concepts for processing the imaging features to extract reliable and robust image biomarkers. KEY POINTS: The key concepts for processing the imaging features to extract reliable and robust image biomarkers are provided. The main differences between radiomics and representation learning approaches are highlighted. The advantages and disadvantages of handcrafted versus learned features are given without losing sight of the clinical purpose of artificial intelligence models.

PMID:39560820 | DOI:10.1186/s41747-024-00529-y

Abdom Radiol (NY). 2024 Nov 19. doi: 10.1007/s00261-024-04686-x. Online ahead of print.

ABSTRACT

OBJECTIVES: To evaluate the effectiveness of super-resolution deep learning reconstruction (SR-DLR) in low-dose abdominal computed tomography (CT) imaging compared with hybrid iterative reconstruction (HIR) and conventional deep learning reconstruction (cDLR) algorithms.

METHODS: We retrospectively analyzed abdominal CT scans performed using a low-dose protocol. Three different image reconstruction algorithms-HIR, cDLR, and SR-DLR-were applied to the same raw image data. Objective evaluations included noise magnitude and contrast-to-noise ratio (CNR), as well as noise power spectrum (NPS) and edge rise slope (ERS). Subjective evaluations were performed by radiologists, who assessed image quality in terms of noise, artifacts, sharpness, and overall diagnostic utility.

RESULTS: Raw CT image data were obtained from 35 patients (mean CTDIvol 11.0 mGy; mean DLP 344.8 mGy/cm). cDLR yielded the lowest noise levels and highest CNR (p < 0.001). However, SR-DLR outperformed cDLR in terms of noise texture and resolution, achieving the lowest NPS peak and highest ERS (p < 0.001 and p = 0.005, respectively). Subjectively, SR-DLR was rated highest across all categories, including noise, artifacts, sharpness, and overall image quality, with statistically significant differences compared to cDLR and HIR (p < 0.001).

CONCLUSION: SR-DLR was the most effective reconstruction algorithm for low-dose abdominal CT imaging, offering superior image quality and noise reduction compared to cDLR and HIR. This suggests that SR-DLR can enhance the reliability and diagnostic accuracy of abdominal imaging, particularly in low-dose settings, making it a valuable tool in clinical practice.

PMID:39560744 | DOI:10.1007/s00261-024-04686-x

Eur Radiol. 2024 Nov 19. doi: 10.1007/s00330-024-11167-8. Online ahead of print.

ABSTRACT

BACKGROUND: Segmentations are crucial in medical imaging for morphological, volumetric, and radiomics biomarkers. Manual segmentation is accurate but not feasible in clinical workflow, while automatic segmentation generally performs sub-par.

PURPOSE: To develop a minimally interactive deep learning-based segmentation method for soft-tissue tumors (STTs) on CT and MRI.

MATERIAL AND METHODS: The interactive method requires the user to click six points near the tumor's extreme boundaries in the image. These six points are transformed into a distance map and serve, with the image, as input for a convolutional neural network. A multi-center public dataset with 514 patients and nine STT phenotypes in seven anatomical locations, with CT or T1-weighted MRI, was used for training and internal validation. For external validation, another public dataset was employed, which included five unseen STT phenotypes in extremities on CT, T1-weighted MRI, and T2-weighted fat-saturated (FS) MRI.

RESULTS: Internal validation resulted in a dice similarity coefficient (DSC) of 0.85 ± 0.11 (mean ± standard deviation) for CT and 0.84 ± 0.12 for T1-weighted MRI. External validation resulted in DSCs of 0.81 ± 0.08 for CT, 0.84 ± 0.09 for T1-weighted MRI, and 0.88 ± 0.08 for T2-weighted FS MRI. Volumetric measurements showed consistent replication with low error internally (volume: 1 ± 28 mm3, r = 0.99; diameter: - 6 ± 14 mm, r = 0.90) and externally (volume: - 7 ± 23 mm3, r = 0.96; diameter: - 3 ± 6 mm, r = 0.99). Interactive segmentation time was considerably shorter (CT: 364 s, T1-weighted MRI: 258s) than manual segmentation (CT: 1639s, T1-weighted MRI: 1895s).

CONCLUSION: The minimally interactive segmentation method effectively segments STT phenotypes on CT and MRI, with robust generalization to unseen phenotypes and imaging modalities.

KEY POINTS: Question Can this deep learning-based method segment soft-tissue tumors faster than can be done manually and more accurately than other automatic methods? Findings The minimally interactive segmentation method achieved accurate segmentation results in internal and external validation, and generalized well across soft-tissue tumor phenotypes and imaging modalities. Clinical relevance This minimally interactive deep learning-based segmentation method could reduce the burden of manual segmentation, facilitate the integration of imaging-based biomarkers (e.g., radiomics) into clinical practice, and provide a fast, semi-automatic solution for volume and diameter measurements (e.g., RECIST).

PMID:39560714 | DOI:10.1007/s00330-024-11167-8

J Neurooncol. 2024 Nov 19. doi: 10.1007/s11060-024-04875-0. Online ahead of print.

ABSTRACT

PURPOSE: Lower-grade gliomas typically exhibit 5-aminolevulinic acid (5-ALA)-induced fluorescence in only 20-30% of cases, a rate that can be increased by doubling the administered dose of 5-ALA. Fluorescence can depict anaplastic foci, which can be precisely sampled to avoid undergrading. We aimed to analyze whether a deep learning model could predict intraoperative fluorescence based on preoperative magnetic resonance imaging (MRI).

METHODS: We evaluated a cohort of 163 glioma patients categorized intraoperatively as fluorescent (n = 83) or non-fluorescent (n = 80). The preoperative MR images of gliomas lacking high-grade characteristics (e.g., necrosis or irregular ring contrast-enhancement) consisted of T1, T1-post gadolinium, and FLAIR sequences. The preprocessed MRIs were fed into an encoder-decoder convolutional neural network (U-Net), pre-trained for tumor segmentation using those three MRI sequences. We used the outputs of the bottleneck layer of the U-Net in the Variational Autoencoder (VAE) as features for classification. We identified and utilized the most effective features in a Random Forest classifier using the principal component analysis (PCA) and the partial least square discriminant analysis (PLS-DA) algorithms. We evaluated the performance of the classifier using a tenfold cross-validation procedure.

RESULTS: Our proposed approach's performance was assessed using mean balanced accuracy, mean sensitivity, and mean specificity. The optimal results were obtained by employing top-performing features selected by PCA, resulting in a mean balanced accuracy of 80% and mean sensitivity and specificity of 84% and 76%, respectively.

CONCLUSIONS: Our findings highlight the potential of a U-Net model, coupled with a Random Forest classifier, for pre-operative prediction of intraoperative fluorescence. We achieved high accuracy using the features extracted by the U-Net model pre-trained for brain tumor segmentation. While the model can still be improved, it has the potential for evaluating when to administer 5-ALA to gliomas lacking typical high-grade radiographic features.

PMID:39560696 | DOI:10.1007/s11060-024-04875-0

Radiology. 2024 Nov;313(2):e233147. doi: 10.1148/radiol.233147.

ABSTRACT

Background Deep learning (DL) algorithms have shown promising results in mammographic screening either compared to a single reader or, when deployed in conjunction with a human reader, compared with double reading. Purpose To externally validate the performance of three DL algorithms as mammographic screen readers in an independent UK data set. Materials and Methods Three commercial DL algorithms (DL-1, DL-2, and DL-3) were retrospectively investigated from January 2022 to June 2022 using consecutive full-field digital mammograms collected at two UK sites during 1 year (2017). Normal cases with 3-year follow-up and histopathologically proven cancer cases detected either at screening (that round or next) or within the 3-year interval were included. A preset specificity threshold equivalent to a single reader was applied. Performance was evaluated for stand-alone DL reading compared with single human reading, and for DL reading combined with human reading compared with double reading, using sensitivity and specificity as the primary metrics. P < .025 was considered to indicate statistical significance for noninferiority testing. Results A total of 26 722 cases (median patient age, 59.0 years [IQR, 54.0-63.0 years]) with mammograms acquired using machines from two vendors were included. Cases included 332 screen-detected, 174 interval, and 254 next-round cancers. Two of three stand-alone DL algorithms achieved noninferior sensitivity (DL-1: 64.8%, P < .001; DL-2: 56.7%, P = .03; DL-3: 58.9%, P < .001) compared with the single first reader (62.8%), and specificity was noninferior for DL-1 (92.8%; P < .001) and DL-2 (96.8%; P < .001) and superior for DL-3 (97.9%; P < .001) compared with the single first reader (96.5%). Combining the DL algorithms with human readers achieved noninferior sensitivity (67.0%, 65.6%, and 65.4% for DL-1, DL-2, and DL-3, respectively; P < .001 for all) compared with double reading (67.4%), and superior specificity (97.4%, 97.6%, and 97.6%; P < .001 for all) compared with double reading (97.1%). Conclusion Use of stand-alone DL algorithms in combination with a human reader could maintain screening accuracy while reducing workload. Published under a CC BY 4.0 license. Supplemental material is available for this article.

PMID:39560480 | DOI:10.1148/radiol.233147

Int J Neural Syst. 2024 Nov 19:2550003. doi: 10.1142/S0129065725500030. Online ahead of print.

ABSTRACT

Seizures have a serious impact on the physical function and daily life of epileptic patients. The automated detection of seizures can assist clinicians in taking preventive measures for patients during the diagnosis process. The combination of deep learning (DL) model with convolutional neural network (CNN) and transformer network can effectively extract both local and global features, resulting in improved seizure detection performance. In this study, an enhanced transformer network named Inresformer is proposed for seizure detection, which is combined with Inception and Residual network extracting different scale features of electroencephalography (EEG) signals to enrich the feature representation. In addition, the improved transformer network replaces the existing Feedforward layers with two half-step Feedforward layers to enhance the nonlinear representation of the model. The proposed architecture utilizes discrete wavelet transform (DWT) to decompose the original EEG signals, and the three sub-bands are selected for signal reconstruction. Then, the Co-MixUp method is adopted to solve the problem of data imbalance, and the processed signals are sent to the Inresformer network for seizure information capture and recognition. Finally, discriminant fusion is performed on the results of three-scale EEG sub-signals to achieve final seizure recognition. The proposed network achieves the best accuracy of 100% on Bonn dataset and the average accuracy of 98.03%, sensitivity of 95.65%, and specificity of 98.57% on the long-term CHB-MIT dataset. Compared to the existing DL networks, the proposed method holds significant potential for clinical research and diagnosis applications with competitive performance.

PMID:39560448 | DOI:10.1142/S0129065725500030

IEEE J Transl Eng Health Med. 2024 Nov 4;12:697-710. doi: 10.1109/JTEHM.2024.3491612. eCollection 2024.

ABSTRACT

To achieve precise Couinaud liver segmentation in preoperative planning for hepatic surgery, accommodating the complex anatomy and significant variations, optimizing surgical approaches, reducing postoperative complications, and preserving liver function.This research presents a novel approach to automating liver segmentation by identifying seven key anatomical landmarks using portal venous phase images from contrast-enhanced magnetic resonance imaging (CE-MRI). By employing a multi-task learning framework, we synchronized the detection of these landmarks with the segmentation process, resulting in accurate and robust delineation of the Couinaud segments.To comprehensively validate our model, we included multiple patient types in our test set-those with normal livers, diffuse liver diseases, and localized liver lesions-under varied imaging conditions, including two field strengths, two devices, and two contrast agents. Our model achieved an average Dice Similarity Coefficient (DSC) of 85.29%, surpassing the next best-performing models by 3.12%.Our research presents a pioneering automated approach for segmenting Couinaud segments using CE-MRI. By correlating landmark detection with segmentation, we enhance surgical planning precision. This method promises improved clinical outcomes by accurately adapting to anatomical variability and reducing potential postoperative complications.Clinical impact: The application of this technique in clinical settings is poised to enhance the precision of liver surgical planning. This could lead to more tailored surgical interventions, minimization of operative risks, and preservation of healthy liver tissue, culminating in improved patient outcomes and potentially lowering the incidence of postoperative complications.Clinical and Translational Impact Statement: This research offers a novel automated liver segmentation technique, enhancing preoperative planning and potentially reducing complications, which may translate into better postoperative outcomes in hepatic surgery.

PMID:39559826 | PMC:PMC11573409 | DOI:10.1109/JTEHM.2024.3491612