Eur Radiol Exp. 2024 Jul 15;8(1):80. doi: 10.1186/s41747-024-00478-6.

ABSTRACT

INTRODUCTION: Breast arterial calcifications (BAC) are common incidental findings on routine mammograms, which have been suggested as a sex-specific biomarker of cardiovascular disease (CVD) risk. Previous work showed the efficacy of a pretrained convolutional network (CNN), VCG16, for automatic BAC detection. In this study, we further tested the method by a comparative analysis with other ten CNNs.

MATERIAL AND METHODS: Four-view standard mammography exams from 1,493 women were included in this retrospective study and labeled as BAC or non-BAC by experts. The comparative study was conducted using eleven pretrained convolutional networks (CNNs) with varying depths from five architectures including Xception, VGG, ResNetV2, MobileNet, and DenseNet, fine-tuned for the binary BAC classification task. Performance evaluation involved area under the receiver operating characteristics curve (AUC-ROC) analysis, F1-score (harmonic mean of precision and recall), and generalized gradient-weighted class activation mapping (Grad-CAM++) for visual explanations.

RESULTS: The dataset exhibited a BAC prevalence of 194/1,493 women (13.0%) and 581/5,972 images (9.7%). Among the retrained models, VGG, MobileNet, and DenseNet demonstrated the most promising results, achieving AUC-ROCs > 0.70 in both training and independent testing subsets. In terms of testing F1-score, VGG16 ranked first, higher than MobileNet (0.51) and VGG19 (0.46). Qualitative analysis showed that the Grad-CAM++ heatmaps generated by VGG16 consistently outperformed those produced by others, offering a finer-grained and discriminative localization of calcified regions within images.

CONCLUSION: Deep transfer learning showed promise in automated BAC detection on mammograms, where relatively shallow networks demonstrated superior performances requiring shorter training times and reduced resources.

RELEVANCE STATEMENT: Deep transfer learning is a promising approach to enhance reporting BAC on mammograms and facilitate developing efficient tools for cardiovascular risk stratification in women, leveraging large-scale mammographic screening programs.

KEY POINTS: • We tested different pretrained convolutional networks (CNNs) for BAC detection on mammograms. • VGG and MobileNet demonstrated promising performances, outperforming their deeper, more complex counterparts. • Visual explanations using Grad-CAM++ highlighted VGG16's superior performance in localizing BAC.

PMID:39004645 | DOI:10.1186/s41747-024-00478-6

Histopathology. 2024 Jul 14. doi: 10.1111/his.15274. Online ahead of print.

ABSTRACT

AIMS: Over 50% of breast cancer cases are "Human epidermal growth factor receptor 2 (HER2) low breast cancer (BC)", characterized by HER2 immunohistochemistry (IHC) scores of 1+ or 2+ alongside no amplification on fluorescence in situ hybridization (FISH) testing. The development of new anti-HER2 antibody-drug conjugates (ADCs) for treating HER2-low breast cancers illustrates the importance of accurately assessing HER2 status, particularly HER2-low breast cancer. In this study we evaluated the performance of a deep-learning (DL) model for the assessment of HER2, including an assessment of the causes of discordances of HER2-Null between a pathologist and the DL model. We specifically focussed on aligning the DL model rules with the ASCO/CAP guidelines, including stained cells' staining intensity and completeness of membrane staining.

METHODS AND RESULTS: We trained a DL model on a multicentric cohort of breast cancer cases with HER2-IHC scores (n = 299). The model was validated on two independent multicentric validation cohorts (n = 369 and n = 92), with all cases reviewed by three senior breast pathologists. All cases underwent a thorough review by three senior breast pathologists, with the ground truth determined by a majority consensus on the final HER2 score among the pathologists. In total, 760 breast cancer cases were utilized throughout the training and validation phases of the study. The model's concordance with the ground truth (ICC = 0.77 [0.68-0.83]; Fisher P = 1.32e-10) is higher than the average agreement among the three senior pathologists (ICC = 0.45 [0.17-0.65]; Fisher P = 2e-3). In the two validation cohorts, the DL model identifies 95% [93% - 98%] and 97% [91% - 100%] of HER2-low and HER2-positive tumours, respectively. Discordant results were characterized by morphological features such as extended fibrosis, a high number of tumour-infiltrating lymphocytes, and necrosis, whilst some artefacts such as nonspecific background cytoplasmic stain in the cytoplasm of tumour cells also cause discrepancy.

CONCLUSION: Deep learning can support pathologists' interpretation of difficult HER2-low cases. Morphological variables and some specific artefacts can cause discrepant HER2-scores between the pathologist and the DL model.

PMID:39004603 | DOI:10.1111/his.15274

Neuroimage. 2024 Jul 12:120737. doi: 10.1016/j.neuroimage.2024.120737. Online ahead of print.

ABSTRACT

Alzheimer's disease (AD) is heterogeneous, but existing methods for capturing this heterogeneity through dimensionality reduction and unsupervised clustering have limitations when it comes to extracting intricate atrophy patterns. In this study, we propose a deep learning based self-supervised framework that characterizes complex atrophy features using latent space representation. It integrates feature engineering, classification, and clustering to synergistically disentangle heterogeneity in Alzheimer's disease. Through this representation learning, we trained a clustered latent space with distinct atrophy patterns and clinical characteristics in AD, and replicated the findings in prodromal Alzheimer's disease. Moreover, we discovered that these clusters are not solely attributed to subtypes but also reflect disease progression in the latent space, representing the core dimensions of heterogeneity, namely progression and subtypes. Furthermore, longitudinal latent space analysis revealed two distinct disease progression pathways: medial temporal and parietotemporal pathways. The proposed approach enables effective latent representations that can be integrated with individual-level cognitive profiles, thereby facilitating a comprehensive understanding of AD heterogeneity.

PMID:39004409 | DOI:10.1016/j.neuroimage.2024.120737

J Biomed Inform. 2024 Jul 12:104685. doi: 10.1016/j.jbi.2024.104685. Online ahead of print.

ABSTRACT

BACKGROUND: Risk prediction plays a crucial role in planning for prevention, monitoring, and treatment. Electronic Health Records (EHRs) offer an expansive repository of temporal medical data encompassing both risk factors and outcome indicators essential for effective risk prediction. However, challenges emerge due to the lack of readily available gold-standard outcomes and the complex effects of various risk factors. Compounding these challenges are the false positives in diagnosis codes, and formidable task of pinpointing the onset timing in annotations.

OBJECTIVE: We develop a Semi-supervised Double Deep Learning Temporal Risk Prediction (SeDDLeR) algorithm based on extensive unlabeled longitudinal Electronic Health Records (EHR) data augmented by a limited set of gold standard labels on the binary status information indicating whether the clinical event of interest occurred during the follow-up period.

METHODS: The SeDDLeR algorithm calculates an individualized risk of developing future clinical events over time using each patient's baseline EHR features via the following steps: (1) construction of an initial EHR-derived surrogate as a proxy for the onset status; (2) deep learning calibration of the surrogate along gold-standard onset status; and (3) semi-supervised deep learning for risk prediction combining calibrated surrogates and gold-standard onset status. To account for missing onset time and heterogeneous follow-up, we introduce temporal kernel weighting. We devise a Gated Recurrent Units (GRUs) module to capture temporal characteristics. We subsequently assess our proposed SeDDLeR method in simulation studies and apply the method to the Massachusetts General Brigham (MGB) Biobank to predict type 2 diabetes (T2D) risk.

RESULTS: SeDDLeR outperforms benchmark risk prediction methods, including Semi-parametric Transformation Model (STM) and DeepHit, with consistently best accuracy across experiments. SeDDLeR achieved the best C-statistics ( 0.815, SE 0.023; vs STM +.084, SE 0.030, P-value .004; vs DeepHit +.055, SE 0.027, P-value .024) and best average time-specific AUC (0.778, SE 0.022; vs STM + 0.059, SE 0.039, P-value .067; vs DeepHit + 0.168, SE 0.032, P-value <0.001) in the MGB T2D study.

CONCLUSION: SeDDLeR can train robust risk prediction models in both real-world EHR and synthetic datasets with minimal requirements of labeling event times. It holds the potential to be incorporated for future clinical trial recruitment or clinical decision-making.

PMID:39004109 | DOI:10.1016/j.jbi.2024.104685

Artif Intell Med. 2024 Jul 6;154:102932. doi: 10.1016/j.artmed.2024.102932. Online ahead of print.

ABSTRACT

Freezing of Gait (FOG) is a noticeable symptom of Parkinson's disease, like being stuck in place and increasing the risk of falls. The wearable multi-channel sensor system is an efficient method to predict and monitor the FOG, thus warning the wearer to avoid falls and improving the quality of life. However, the existing approaches for the prediction of FOG mainly focus on a single sensor system and cannot handle the interference between multi-channel wearable sensors. Hence, we propose a novel multi-channel time-series neural network (MCT-Net) approach to merge multi-channel gait features into a comprehensive prediction framework, alerting patients to FOG symptoms in advance. Owing to the causal distributed convolution, MCT-Net is a real-time method available to give optimal prediction earlier and implemented in remote devices. Moreover, intra-channel and inter-channel transformers of MCT-Net extract and integrate different sensor position features into a unified deep learning model. Compared with four other state-of-the-art FOG prediction baselines, the proposed MCT-Net obtains 96.21% in accuracy and 80.46% in F1-score on average 2 s before FOG occurrence, demonstrating the superiority of MCT-Net.

PMID:39004005 | DOI:10.1016/j.artmed.2024.102932

J Contam Hydrol. 2024 Jul 5;265:104394. doi: 10.1016/j.jconhyd.2024.104394. Online ahead of print.

ABSTRACT

Macrodispersivity is critical for predicting solute behaviors with dispersive transport models. Conventional methods of estimating macrodispersivity usually need to solve flow equations and are time-consuming. Convolutional neural networks (CNN) have recently been proven capable of efficiently mapping the hydraulic conductivity field and macrodispersivity. However, the mapping accuracy still needs further improvement. In this paper, we present a new network shortcut connection style called weighted multi-scale connections (WMC) for convolutional neural networks to improve mapping accuracy. We provide empirical evidence showing that the WMC can improve the performance of CNN in macrodispersivity estimation by implementing the WMC in CNNs (CNN without short-cut connections, ResNet, and DenseNet), and evaluating them on datasets of macrodispersivity estimation. For the CNN without short-cut connections, the WMC can improve the estimating R2 by at least 3% on three datasets of conductivity fields. For ResNet18, the WMC improved the estimated R2 by an average of 2.5% on all three datasets. For ResNet34, the WMC improved the estimated R2 by an average of 5.6%. For ResNet50, the WMC improved the estimated R2 by an average of 16%. For ResNet101, the WMC improved the estimating R2 by an average of 30%. For DenseNets, the improved estimated R2 ranges from 0.5% to 5%. The WMC can strengthen feature propagation of different sizes and alleviate the vanishing-gradient issue. Moreover, it can be implemented to any CNN with down-sampling layers or blocks.

PMID:39003944 | DOI:10.1016/j.jconhyd.2024.104394

Int J Med Inform. 2024 Jul 11;190:105544. doi: 10.1016/j.ijmedinf.2024.105544. Online ahead of print.

ABSTRACT

OBJECTIVE: To determine the incidence of patients presenting in pain to a large Australian inner-city emergency department (ED) using a clinical text deep learning algorithm.

MATERIALS AND METHODS: A fine-tuned, domain-specific, transformer-based clinical text deep learning model was used to interpret free-text nursing assessments in the electronic medical records of 235,789 adult presentations to the ED over a three-year period. The model classified presentations according to whether the patient had pain on arrival at the ED. Interrupted time series analysis was used to determine the incidence of pain in patients on arrival over time. We described the changes in the population characteristics and incidence of patients with pain on arrival occurring with the start of the Covid-19 pandemic.

RESULTS: 55.16% (95%CI 54.95%-55.36%) of all patients presenting to this ED had pain on arrival. There were differences in demographics and arrival and departure patterns between patients with and without pain. The Covid-19 pandemic initially precipitated a decrease followed by a sharp, sustained rise in pain on arrival, with concurrent changes to the population arriving in pain and their treatment.

DISCUSSION: Applying a clinical text deep learning model has successfully identified the incidence of pain on arrival. It represents an automated, reproducible mechanism to identify pain from routinely collected medical records. The description of this population and their treatment forms the basis of intervention to improve care for patients with pain. The combination of the clinical text deep learning models and interrupted time series analysis has reported on the effects of the Covid-19 pandemic on pain care in the ED, outlining a methodology to assess the impact of significant events or interventions on pain care in the ED.

CONCLUSION: Applying a novel deep learning approach to identifying pain guides methodological approaches to evaluating pain care interventions in the ED, giving previously unavailable population-level insights.

PMID:39003790 | DOI:10.1016/j.ijmedinf.2024.105544

J Agric Food Chem. 2024 Jul 9. doi: 10.1021/acs.jafc.4c03023. Online ahead of print.

ABSTRACT

Currently, allosteric inhibitors have emerged as an effective strategy in the development of preservatives against the drug-resistant Botrytis cinerea (B. cinerea). However, their passively driven development efficiency has proven challenging to meet the practical demands. Here, leveraging the deep learning Neural Relational Inference (NRI) framework, we actively identified an allosteric inhibitor targeting B. cinerea Chitinase, namely, 2-acetonaphthone. 2-Acetonaphthone binds to the crucial domain of Chitinase, forming the strong interaction with the allosteric sites. Throughout the interaction process, 2-acetonaphthone diminished the overall connectivity of the protein, inducing conformational changes. These findings align with the results obtained from Chitinase activity experiments, revealing an IC50 value of 67.6 μg/mL. Moreover, 2-acetonaphthone exhibited outstanding anti-B. cinerea activity by inhibiting Chitinase. In the gray mold infection model, 2-acetonaphthone significantly extended the preservation time of cherry tomatoes, positioning it as a promising preservative for fruit storage.

PMID:39003764 | DOI:10.1021/acs.jafc.4c03023

BJU Int. 2024 Jul 14. doi: 10.1111/bju.16412. Online ahead of print.

ABSTRACT

OBJECTIVES: To assess artificial intelligence (AI) ability to evaluate intraprostatic prostate cancer (PCa) on prostate-specific membrane antigen positron emission tomography (PSMA PET) scans prior to active treatment (radiotherapy or prostatectomy).

MATERIALS AND METHODS: This systematic review was registered on the International Prospective Register of Systematic Reviews (PROSPERO identifier: CRD42023438706). A search was performed on Medline, Embase, Web of Science, and Engineering Village with the following terms: 'artificial intelligence', 'prostate cancer', and 'PSMA PET'. All articles published up to February 2024 were considered. Studies were included if patients underwent PSMA PET scan to evaluate intraprostatic lesions prior to active treatment. The two authors independently evaluated titles, abstracts, and full text. The Prediction model Risk Of Bias Assessment Tool (PROBAST) was used.

RESULTS: Our search yield 948 articles, of which 14 were eligible for inclusion. Eight studies met the primary endpoint of differentiating high-grade PCa. Differentiating between International Society of Urological Pathology (ISUP) Grade Group (GG) ≥3 PCa had an accuracy between 0.671 to 0.992, sensitivity of 0.91, specificity of 0.35. Differentiating ISUP GG ≥4 PCa had an accuracy between 0.83 and 0.88, sensitivity was 0.89, specificity was 0.87. AI could identify non-PSMA-avid lesions with an accuracy of 0.87, specificity of 0.85, and specificity of 0.89. Three studies demonstrated ability of AI to detect extraprostatic extensions with an area under curve between 0.70 and 0.77. Lastly, AI can automate segmentation of intraprostatic lesion and measurement of gross tumour volume.

CONCLUSION: Although the current state of AI differentiating high-grade PCa is promising, it remains experimental and not ready for routine clinical application. Benefits of using AI to assess intraprostatic lesions on PSMA PET scans include: local staging, identifying otherwise radiologically occult lesions, standardisation and expedite reporting of PSMA PET scans. Larger, prospective, multicentre studies are needed.

PMID:39003625 | DOI:10.1111/bju.16412

Sci Rep. 2024 Jul 13;14(1):16212. doi: 10.1038/s41598-024-67109-3.

ABSTRACT

To obtain seasonable and precise crop yield information with fine resolution is very important for ensuring the food security. However, the quantity and quality of available images and the selection of prediction variables often limit the performance of yield prediction. In our study, the synthesized images of Landsat and MODIS were used to provide remote sensing (RS) variables, which can fill the missing values of Landsat images well and cover the study area completely. The deep learning (DL) was used to combine different vegetation index (VI) with climate data to build wheat yield prediction model in Hebei Province (HB). The results showed that kernel NDVI (kNDVI) and near-infrared reflectance (NIRv) slightly outperform normalized difference vegetation index (NDVI) in yield prediction. And the regression algorithm had a more prominent effect on yield prediction, while the yield prediction model using Long Short-Term Memory (LSTM) outperformed the yield prediction model using Light Gradient Boosting Machine (LGBM). The model combining LSTM algorithm and NIRv had the best prediction effect and relatively stable performance in single year. The optimal model was then used to generate 30 m resolution wheat yield maps in the past 20 years, with higher overall accuracy. In addition, we can define the optimum prediction time at April, which can consider simultaneously the performance and lead time. In general, we expect that this prediction model can provide important information to understand and ensure food security.

PMID:39003342 | DOI:10.1038/s41598-024-67109-3

Sci Rep. 2024 Jul 13;14(1):16166. doi: 10.1038/s41598-024-65954-w.

ABSTRACT

Machine learning is increasingly applied to Earth Observation (EO) data to obtain datasets that contribute towards international accords. However, these datasets contain inherent uncertainty that needs to be quantified reliably to avoid negative consequences. In response to the increased need to report uncertainty, we bring attention to the promise of conformal prediction within the domain of EO. Unlike previous uncertainty quantification methods, conformal prediction offers statistically valid prediction regions while concurrently supporting any machine learning model and data distribution. To support the need for conformal prediction, we reviewed EO datasets and found that only 22.5% of the datasets incorporated a degree of uncertainty information, with unreliable methods prevalent. Current open implementations require moving large amounts of EO data to the algorithms. We introduced Google Earth Engine native modules that bring conformal prediction to the data and compute, facilitating the integration of uncertainty quantification into existing traditional and deep learning modelling workflows. To demonstrate the versatility and scalability of these tools we apply them to valued EO applications spanning local to global extents, regression, and classification tasks. Subsequently, we discuss the opportunities arising from the use of conformal prediction in EO. We anticipate that accessible and easy-to-use tools, such as those provided here, will drive wider adoption of rigorous uncertainty quantification in EO, thereby enhancing the reliability of downstream uses such as operational monitoring and decision-making.

PMID:39003341 | DOI:10.1038/s41598-024-65954-w

J Formos Med Assoc. 2024 Jul 12:S0929-6646(24)00313-9. doi: 10.1016/j.jfma.2024.07.010. Online ahead of print.

ABSTRACT

BACKGROUND/PURPOSE: The global incidence of lip and oral cavity cancer continues to rise, necessitating improved early detection methods. This study leverages the capabilities of computer vision and deep learning to enhance the early detection and classification of oral mucosal lesions.

METHODS: A dataset initially consisting of 6903 white-light macroscopic images collected from 2006 to 2013 was expanded to over 50,000 images to train the YOLOv7 deep learning model. Lesions were categorized into three referral grades: benign (green), potentially malignant (yellow), and malignant (red), facilitating efficient triage.

RESULTS: The YOLOv7 models, particularly the YOLOv7-E6, demonstrated high precision and recall across all lesion categories. The YOLOv7-D6 model excelled at identifying malignant lesions with notable precision, recall, and F1 scores. Enhancements, including the integration of coordinate attention in the YOLOv7-D6-CA model, significantly improved the accuracy of lesion classification.

CONCLUSION: The study underscores the robust comparison of various YOLOv7 model configurations in the classification to triage oral lesions. The overall results highlight the potential of deep learning models to contribute to the early detection of oral cancers, offering valuable tools for both clinical settings and remote screening applications.

PMID:39003230 | DOI:10.1016/j.jfma.2024.07.010

Sci Rep. 2024 Jul 13;14(1):16204. doi: 10.1038/s41598-024-67217-0.

ABSTRACT

To retrospectively assess the effectiveness of deep learning (DL) model, based on breast magnetic resonance imaging (MRI), in predicting preoperative lymphovascular invasion (LVI) status in patients diagnosed with invasive breast cancer who have negative axillary lymph nodes (LNs). Data was gathered from 280 patients, including 148 with LVI-positive and 141 with LVI-negative lesions. These patients had undergone preoperative breast MRI and were histopathologically confirmed to have invasive breast cancer without axillary LN metastasis. The cohort was randomly split into training and validation groups in a 7:3 ratio. Radiomics features for each lesion were extracted from the first post-contrast dynamic contrast-enhanced (DCE)-MRI. The Least Absolute Shrinkage and Selection Operator (LASSO) regression method and logistic regression analyses were employed to identify significant radiomic features and clinicoradiological variables. These models were established using four machine learning (ML) algorithms and one DL algorithm. The predictive performance of the models (radiomics, clinicoradiological, and combination) was assessed through discrimination and compared using the DeLong test. Four clinicoradiological parameters and 10 radiomic features were selected by LASSO for model development. The Multilayer Perceptron (MLP) model, constructed using both radiomic and clinicoradiological features, demonstrated excellent performance in predicting LVI, achieving a high area under the curve (AUC) of 0.835 for validation. The DL model (MLP-radiomic) achieved the highest accuracy (AUC = 0.896), followed by DL model (MLP-combination) with an AUC of 0.835. Both DL models were significantly superior to the ML model (RF-clinical) with an AUC of 0.720. The DL model (MLP), which integrates radiomic features with clinicoradiological information, effectively aids in the preoperative determination of LVI status in patients with invasive breast cancer and negative axillary LNs. This is beneficial for making informed clinical decisions.

PMID:39003325 | DOI:10.1038/s41598-024-67217-0

Sci Rep. 2024 Jul 13;14(1):16165. doi: 10.1038/s41598-024-65795-7.

ABSTRACT

When conducting spine-related diagnosis and surgery, the three-dimensional (3D) upright posture of the spine under natural weight bearing is of significant clinical value for physicians to analyze the force on the spine. However, existing medical imaging technologies cannot meet current requirements of medical service. On the one hand, the mainstream 3D volumetric imaging modalities (e.g. CT and MRI) require patients to lie down during the imaging process. On the other hand, the imaging modalities conducted in an upright posture (e.g. radiograph) can only realize 2D projections, which lose the valid information of spinal anatomy and curvature. Developments of deep learning-based 3D reconstruction methods bring potential to overcome the limitations of the existing medical imaging technologies. To deal with the limitations of current medical imaging technologies as is described above, in this paper, we propose a novel deep learning framework, ReVerteR, which can realize automatic 3D Reconstruction of Vertebrae from orthogonal bi-planar Radiographs. With the utilization of self-attention mechanism and specially designed loss function combining Dice, Hausdorff, Focal, and MSE, ReVerteR can alleviate the sample-imbalance problem during the reconstruction process and realize the fusion of the centroid annotation and the focused vertebra. Furthermore, aiming at automatic and customized 3D spinal reconstruction in real-world scenarios, we extend ReVerteR to a clinical deployment-oriented framework, and develop an interactive interface with all functions in the framework integrated so as to enhance human-computer interaction during clinical decision-making. Extensive experiments and visualization conducted on our constructed datasets based on two benchmark datasets of spinal CT, VerSe 2019 and VerSe 2020, demonstrate the effectiveness of our proposed ReVerteR. In this paper, we propose an automatic 3D reconstruction method of vertebrae based on orthogonal bi-planar radiographs. With the 3D upright posture of the spine under natural weight bearing effectively constructed, our proposed method is expected to better support doctors make clinical decision during spine-related diagnosis and surgery.

PMID:39003269 | DOI:10.1038/s41598-024-65795-7

Acad Radiol. 2024 Jul 12:S1076-6332(24)00422-7. doi: 10.1016/j.acra.2024.06.035. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: To assess the efficacy of a preoperative contrast-enhanced CT (CECT)-based deep learning radiomics nomogram (DLRN) for predicting murine double minute 2 (MDM2) gene amplification as a means of distinguishing between retroperitoneal well-differentiated liposarcomas (WDLPS) and lipomas.

METHODS: This retrospective multi-center study included 167 patients (training/external test cohort, 104/63) with MDM2-positive WDLPS or MDM2-negative lipomas. Clinical data and CECT features were independently measured and analyzed by two radiologists. A clinico-radiological model, radiomics signature (RS), deep learning and radiomics signature (DLRS), and a DLRN incorporating radiomics and deep learning features were developed to differentiate between WDLPS and lipoma. The model utility was evaluated based on the area under the receiver operating characteristic curve (AUC), accuracy, calibration curve, and decision curve analysis (DCA).

RESULTS: The DLRN showed good performance for distinguishing retroperitoneal lipomas and WDLPS in the training (AUC, 0.981; accuracy, 0.933) and external validation group (AUC, 0.861; accuracy, 0.810). The DeLong test revealed the DLRN was noticeably better than clinico-radiological and RS models (training: 0.981 vs. 0.890 vs. 0.751; validation: 0.861 vs. 0.724 vs. 0.700; both P < 0.05); however, no discernible difference in performance was seen between the DLRN and DLRS (training: 0.981 vs. 0.969; validation: 0.861 vs. 0.837; both P > 0.05). The calibration curve analysis and DCA demonstrated that the nomogram exhibited good calibration and offered substantial clinical advantages.

CONCLUSION: The DLRN exhibited strong predictive capability in predicting WDLPS and retroperitoneal lipomas preoperatively, making it a promising imaging biomarker that can facilitate personalized management and precision medicine.

PMID:39003228 | DOI:10.1016/j.acra.2024.06.035

J Biomed Inform. 2024 Jul 11:104688. doi: 10.1016/j.jbi.2024.104688. Online ahead of print.

ABSTRACT

OBJECTIVE: Survival analysis is widely utilized in healthcare to predict the timing of disease onset. Traditional methods of survival analysis are usually based on Cox Proportional Hazards model and assume proportional risk for all subjects. However, this assumption is rarely true for most diseases, as the underlying factors have complex, non-linear, and time-varying relationships. This concern is especially relevant for pregnancy, where the risk for pregnancy-related complications, such as preeclampsia, varies across gestation. Recently, deep learning survival models have shown promise in addressing the limitations of classical models, as the novel models allow for non-proportional risk handling, capturing nonlinear relationships, and navigating complex temporal dynamics.

METHODS: We present a methodology to model the temporal risk of preeclampsia during pregnancy and investigate the associated clinical risk factors. We utilized a retrospective dataset including 66,425 pregnant individuals who delivered in two tertiary care centers from 2015 to 2023. We modeled the preeclampsia risk by modifying DeepHit, a deep survival model, which leverages neural network architecture to capture time-varying relationships between covariates in pregnancy. We applied time series k-means clustering to DeepHit's normalized output and investigated interpretability using Shapley values.

RESULTS: We demonstrate that DeepHit can effectively handle high-dimensional data and evolving risk hazards over time with performance similar to the Cox Proportional Hazards model, achieving an area under the curve (AUC) of 0.78 for both models. The deep survival model outperformed traditional methodology by identifying time-varied risk trajectories for preeclampsia, providing insights for early and individualized intervention. K-means clustering resulted in patients delineating into low-risk, early-onset, and late-onset preeclampsia groups-notably, each of those has distinct risk factors.

CONCLUSION: This work demonstrates a novel application of deep survival analysis in time-varying prediction of preeclampsia risk. Our results highlight the advantage of deep survival models compared to Cox Proportional Hazards models in providing personalized risk trajectory and demonstrating the potential of deep survival models to generate interpretable and meaningful clinical applications in medicine.

PMID:39002866 | DOI:10.1016/j.jbi.2024.104688

Schizophr Res. 2024 Jul 12;271:28-35. doi: 10.1016/j.schres.2024.07.015. Online ahead of print.

ABSTRACT

This paper proposes a high-accuracy EEG-based schizophrenia (SZ) detection approach. Unlike comparable literature studies employing conventional machine learning algorithms, our method autonomously extracts the necessary features for network training from EEG recordings. The proposed model is a ten-layered CNN that contains a max pooling layer, a Global Average Pooling layer, four convolution layers, two dropout layers for overfitting prevention, and two fully connected layers. The efficiency of the suggested method was assessed using the ten-fold-cross validation technique and the EEG records of 14 healthy subjects and 14 SZ patients. The obtained mean accuracy score was 99.18 %. To confirm the high mean accuracy attained, we tested the model on unseen data with a near-perfect accuracy score (almost 100 %). In addition, the results we obtained outperform numerous other comparable works.

PMID:39002527 | DOI:10.1016/j.schres.2024.07.015

Accid Anal Prev. 2024 Jul 12;206:107712. doi: 10.1016/j.aap.2024.107712. Online ahead of print.

ABSTRACT

Urban arterial and collector roads, while interconnected within the urban transportation network, serve distinct purposes, leading to different driving risk profiles. Investigating these differences using advanced methods is of paramount significance. This study aims to achieve this by primarily collecting and processing relevant vehicle trajectory data alongside driver-vehicle-road-environment data. A comprehensive risk assessment matrix is constructed to assess driving risks, incorporating multiple conflict and traffic flow indicators with statistically temporal stability. The Entropy weight-TOPSIS method and the K-means algorithm are employed to determine the risk scores and levels of the target arterial and collector roads. Using risk levels as the outcome variables and multi-scale features as the explanatory variables, random parameters models with heterogeneity in means and variances are developed to identify the determinants of driving risks at different levels. Likelihood ratio tests and comparisons of out-of-sample and within-sample prediction are conducted. Results reveal significant statistical differences in the risk profiles between arterial and collector roads. The marginal effects of significant parameters are then calculated separately for arterial and collector roads, indicating that several factors have different impacts on the probability of risk levels for arterial and collector roads, such as the number of movable elements in road landscape pictures, the standard deviation of the vehicle's lateral acceleration, the average standard deviation of speed for all vehicles on the road segment, and the number of one-way lanes on the road segment. Some practical implications are provided based on the findings. Future research can be implemented by expanding the collected data to different regions and cities over longer periods.

PMID:39002352 | DOI:10.1016/j.aap.2024.107712

Comput Biol Med. 2024 Jul 12;179:108845. doi: 10.1016/j.compbiomed.2024.108845. Online ahead of print.

ABSTRACT

BACKGROUND: Brain extraction in magnetic resonance imaging (MRI) data is an important segmentation step in many neuroimaging preprocessing pipelines. Image segmentation is one of the research fields in which deep learning had the biggest impact in recent years. Consequently, traditional brain extraction methods are now being replaced by deep learning-based methods.

METHOD: Here, we used a unique dataset compilation comprising 7837 T1-weighted (T1w) MR images from 191 different OpenNeuro datasets in combination with advanced deep learning methods to build a fast, high-precision brain extraction tool called deepbet.

RESULTS: deepbet sets a novel state-of-the-art performance during cross-dataset validation with a median Dice score (DSC) of 99.0 on unseen datasets, outperforming the current best performing deep learning (DSC=97.9) and classic (DSC=96.5) methods. While current methods are more sensitive to outliers, deepbet achieves a Dice score of >97.4 across all 7837 images from 191 different datasets. This robustness was additionally tested in 5 external datasets, which included challenging clinical MR images. During visual exploration of each method's output which resulted in the lowest Dice score, major errors could be found for all of the tested tools except deepbet. Finally, deepbet uses a compute efficient variant of the UNet architecture, which accelerates brain extraction by a factor of ≈10 compared to current methods, enabling the processing of one image in ≈2 s on low level hardware.

CONCLUSIONS: In conclusion, deepbet demonstrates superior performance and reliability in brain extraction across a wide range of T1w MR images of adults, outperforming existing top tools. Its high minimal Dice score and minimal objective errors, even in challenging conditions, validate deepbet as a highly dependable tool for accurate brain extraction. deepbet can be conveniently installed via "pip install deepbet" and is publicly accessible at https://github.com/wwu-mmll/deepbet.

PMID:39002314 | DOI:10.1016/j.compbiomed.2024.108845

Neuroimage Clin. 2024 Jul 2;43:103638. doi: 10.1016/j.nicl.2024.103638. Online ahead of print.

ABSTRACT

Machine learning offers great potential for automated prediction of post-stroke symptoms and their response to rehabilitation. Major challenges for this endeavour include the very high dimensionality of neuroimaging data, the relatively small size of the datasets available for learning and interpreting the predictive features, as well as, how to effectively combine neuroimaging and tabular data (e.g. demographic information and clinical characteristics). This paper evaluates several solutions based on two strategies. The first is to use 2D images that summarise MRI scans. The second is to select key features that improve classification accuracy. Additionally, we introduce the novel approach of training a convolutional neural network (CNN) on images that combine regions-of-interests (ROIs) extracted from MRIs, with symbolic representations of tabular data. We evaluate a series of CNN architectures (both 2D and a 3D) that are trained on different representations of MRI and tabular data, to predict whether a composite measure of post-stroke spoken picture description ability is in the aphasic or non-aphasic range. MRI and tabular data were acquired from 758 English speaking stroke survivors who participated in the PLORAS study. Each participant was assigned to one of five different groups that were matched for initial severity of symptoms, recovery time, left lesion size and the months or years post-stroke that spoken description scores were collected. Training and validation were carried out on the first four groups. The fifth (lock-box/test set) group was used to test how well model accuracy generalises to new (unseen) data. The classification accuracy for a baseline logistic regression was 0.678 based on lesion size alone, rising to 0.757 and 0.813 when initial symptom severity and recovery time were successively added. The highest classification accuracy (0.854), area under the curve (0.899) and F1 score (0.901) were observed when 8 regions of interest were extracted from each MRI scan and combined with lesion size, initial severity and recovery time in a 2D Residual Neural Network (ResNet). This was also the best model when data were limited to the 286 participants with moderate or severe initial aphasia (with area under curve = 0.865), a group that would be considered more difficult to classify. Our findings demonstrate how imaging and tabular data can be combined to achieve high post-stroke classification accuracy, even when the dataset is small in machine learning terms. We conclude by proposing how the current models could be improved to achieve even higher levels of accuracy using images from hospital scanners.

PMID:39002223 | DOI:10.1016/j.nicl.2024.103638