Anal Methods. 2024 Nov 21. doi: 10.1039/d4ay01755b. Online ahead of print.

ABSTRACT

Light sources exhibit significant absorption and scattering effects during the transmission through biological tissues, posing challenges in identifying heterogeneities in multi-spectral images. This paper introduces a fusion of techniques encompassing the spatial pyramid matching model (SPM), modulation and demodulation (M_D), and frame accumulation (FA). These techniques not only elevate image quality but also augment the precision of heterogeneous classification in multi-spectral transmission images (MTI) within deep learning network models (DLNM). Initially, experiments are designed to capture MTI of phantoms. Subsequently, the images are preprocessed separately through a combination of different techniques such as SPM, M_D and FA. Ultimately, multi-spectral fusion pseudo-color images derived from U-Net semantic segmentation are fed into VGG16/19 and ResNet50/101 networks for heterogeneous classification. Among them, different combinations of SPM, M_D and FA significantly enhance the quality of images, facilitating the extraction of heterogeneous feature information from multi-spectral images. In comparison to the classification accuracy achieved in the original image VGG and ResNet network models, all images after preprocessing effectively improved the classification accuracy of heterogeneities. Following scatter correction, images processed with 3.5 Hz modulation-demodulation combined with frame accumulation (M_D-FA) attain the highest classification accuracy for heterogeneities in the VGG19 and ResNet101 models, achieving accuracies of 95.47% and 98.47%, respectively. In conclusion, this paper utilizes different combinations of SPM, M_D and FA techniques to not only enhance the quality of images but also further improve the accuracy of DLNM in heterogeneous classification, which will promote the clinical application of MTI technique in breast tumor screening.

PMID:39569814 | DOI:10.1039/d4ay01755b

Trends Hear. 2024 Jan-Dec;28:23312165241298606. doi: 10.1177/23312165241298606.

ABSTRACT

Cochlear implant (CI) users, even with substantial speech comprehension, generally have poor sensitivity to pitch information (or fundamental frequency, F0). This insensitivity is often attributed to limited spectral and temporal resolution in the CI signals. However, the pitch sensitivity markedly varies among individuals, and some users exhibit fairly good sensitivity. This indicates that the CI signal contains sufficient information about F0, and users' sensitivity is predominantly limited by other physiological conditions such as neuroplasticity or neural health. We estimated the upper limit of F0 information that a CI signal can convey by decoding F0 from simulated CI signals (multi-channel pulsatile signals) with a deep neural network model (referred to as the CI model). We varied the number of electrode channels and the pulse rate, which should respectively affect spectral and temporal resolutions of stimulus representations. The F0-estimation performance generally improved with increasing number of channels and pulse rate. For the sounds presented under quiet conditions, the model performance was at best comparable to that of a control waveform model, which received raw-waveform inputs. Under conditions in which background noise was imposed, the performance of the CI model generally degraded by a greater degree than that of the waveform model. The pulse rate had a particularly large effect on predicted performance. These observations indicate that the CI signal contains some information for predicting F0, which is particularly sufficient for targets under quiet conditions. The temporal resolution (represented as pulse rate) plays a critical role in pitch representation under noisy conditions.

PMID:39569552 | DOI:10.1177/23312165241298606

J Am Med Inform Assoc. 2024 Nov 21:ocae271. doi: 10.1093/jamia/ocae271. Online ahead of print.

ABSTRACT

OBJECTIVE: Machine learning applications for longitudinal electronic health records often forecast the risk of events at fixed time points, whereas survival analysis achieves dynamic risk prediction by estimating time-to-event distributions. Here, we propose a novel conditional variational autoencoder-based method, DySurv, which uses a combination of static and longitudinal measurements from electronic health records to estimate the individual risk of death dynamically.

MATERIALS AND METHODS: DySurv directly estimates the cumulative risk incidence function without making any parametric assumptions on the underlying stochastic process of the time-to-event. We evaluate DySurv on 6 time-to-event benchmark datasets in healthcare, as well as 2 real-world intensive care unit (ICU) electronic health records (EHR) datasets extracted from the eICU Collaborative Research (eICU) and the Medical Information Mart for Intensive Care database (MIMIC-IV).

RESULTS: DySurv outperforms other existing statistical and deep learning approaches to time-to-event analysis across concordance and other metrics. It achieves time-dependent concordance of over 60% in the eICU case. It is also over 12% more accurate and 22% more sensitive than in-use ICU scores like Acute Physiology and Chronic Health Evaluation (APACHE) and Sequential Organ Failure Assessment (SOFA) scores. The predictive capacity of DySurv is consistent and the survival estimates remain disentangled across different datasets.

DISCUSSION: Our interdisciplinary framework successfully incorporates deep learning, survival analysis, and intensive care to create a novel method for time-to-event prediction from longitudinal health records. We test our method on several held-out test sets from a variety of healthcare datasets and compare it to existing in-use clinical risk scoring benchmarks.

CONCLUSION: While our method leverages non-parametric extensions to deep learning-guided estimations of the survival distribution, further deep learning paradigms could be explored.

PMID:39569428 | DOI:10.1093/jamia/ocae271

J Hepatocell Carcinoma. 2024 Nov 16;11:2223-2239. doi: 10.2147/JHC.S493478. eCollection 2024.

ABSTRACT

PURPOSE: To construct a 2.5-dimensional (2.5D) CT radiomics-based deep learning (DL) model to predict early postoperative recurrence of hepatocellular carcinoma (HCC).

PATIENTS AND METHODS: We retrospectively analyzed the data of patients who underwent HCC resection at 2 centers. The 232 patients from center 1 were randomly divided into the training (162 patients) and internal validation cohorts (70 patients); 91 patients from center 2 formed the external validation cohort. We developed a 2.5D DL model based on a central 2D image with the maximum tumor cross-section and adjacent slices. Multiple views (transverse, sagittal, and coronal) and phases (arterial, plain, and portal) were incorporated. Multi-instance learning techniques were applied to the extracted data; the resulting comprehensive feature set was modeled using Logistic Regression, RandomForest, ExtraTrees, XGBoost, and LightGBM, with 5-fold cross validation and hyperparameter optimization with Grid-search. Receiver operating characteristic curves, calibration curves, DeLong test, and decision curve analysis were used to evaluate model performance.

RESULTS: The 2.5D DL model performed well in the training (AUC: 0.920), internal validation (AUC: 0.825), and external validation cohorts (AUC: 0.795). The 3D DL model performed well in the training cohort and poorly in the internal and external validation cohorts (AUCs: 0.751, 0.666, and 0.567, respectively), indicating overfitting. The combined model (2.5D DL+clinical) performed well in all cohorts (AUCs: 0.921, 0.835, 0.804). The Hosmer-Lemeshow test, DeLong test, and decision curve analysis confirmed the superiority of the combined model over the other signatures.

CONCLUSION: The combined model integrating 2.5D DL and clinical features accurately predicts early postoperative HCC recurrence.

PMID:39569409 | PMC:PMC11577935 | DOI:10.2147/JHC.S493478

iScience. 2024 Oct 21;27(11):111209. doi: 10.1016/j.isci.2024.111209. eCollection 2024 Nov 15.

ABSTRACT

Noninvasive magnetic resonance imaging (MRI) of the relayed nuclear Overhauser effect (rNOE) constitutes a promising approach for gaining biological insights into various pathologies, including brain cancer, kidney injury, ischemic stroke, and liver disease. However, rNOE imaging is time-consuming and prone to biases stemming from the water T1 and the semisolid magnetization transfer (MT) contrasts. Here, we developed a rapid rNOE quantification approach, combining magnetic resonance fingerprinting (MRF) acquisition with deep-learning-based reconstruction. The method was systematically validated using tissue-mimicking phantoms, wild-type mice (n = 7), and healthy human volunteers (n = 5). In vitro rNOE parameter maps generated by MRF were highly correlated with ground truth (r > 0.98, p < 0.001). Simultaneous mapping of the rNOE and the semisolid MT exchange parameters in mice and humans were in agreement with previously reported literature values. Whole-brain 3D parameter mapping in humans took less than 5 min (282 s for acquisition and less than 2 s for reconstruction). With its demonstrated ability to rapidly extract quantitative molecular maps, deep rNOE-MRF can potentially serve as a valuable tool for the characterization and detection of molecular abnormalities in vivo.

PMID:39569380 | PMC:PMC11576397 | DOI:10.1016/j.isci.2024.111209

Patterns (N Y). 2024 Sep 30;5(10):101063. doi: 10.1016/j.patter.2024.101063. eCollection 2024 Oct 11.

ABSTRACT

Organoids and 3D imaging techniques are crucial for studying human tissue structure and function, but traditional 3D reconstruction methods are expensive and time consuming, relying on complete z stack confocal microscopy data. This paper introduces VONet, a deep learning-based system for 3D organoid rendering that uses a fully convolutional neural network to reconstruct entire 3D structures from a minimal number of z stack images. VONet was trained on a library of over 39,000 virtual organoids (VOs) with diverse structural features and achieved an average intersection over union of 0.82 in performance validation. Remarkably, VONet can predict the structure of deeper focal plane regions, unseen by conventional confocal microscopy. This innovative approach and VO dataset offer significant advancements in 3D bioimaging technologies.

PMID:39569212 | PMC:PMC11573902 | DOI:10.1016/j.patter.2024.101063

RSC Adv. 2024 Nov 20;14(50):37035-37044. doi: 10.1039/d4ra05422a. eCollection 2024 Nov 19.

ABSTRACT

Semi-empirical quantum chemistry methods estimate the binding free energies of protein-ligand complexes. We present an integrated approach combining the GFN2-xTB method with de novo design for the generation and evaluation of potential inhibitors of acetylcholinesterase (AChE). We employed chemical language model-based molecule generation to explore the synthetically accessible chemical space around the natural product Huperzine A, a potent AChE inhibitor. Four distinct molecular libraries were created using structure- and ligand-based molecular de novo design with SMILES and SELFIES representations, respectively. These libraries were computationally evaluated for synthesizability, novelty, and predicted biological activity. The candidate molecules were subjected to molecular docking to identify hypothetical binding poses, which were further refined using Gibbs free energy calculations. The structurally novel top-ranked molecule was chemically synthesized and biologically tested, demonstrating moderate micromolar activity against AChE. Our findings highlight the potential and certain limitations of integrating deep learning-based molecular generation with semi-empirical quantum chemistry-based activity prediction for structure-based drug design.

PMID:39569121 | PMC:PMC11577348 | DOI:10.1039/d4ra05422a

Front Neurorobot. 2024 Nov 6;18:1484751. doi: 10.3389/fnbot.2024.1484751. eCollection 2024.

ABSTRACT

INTRODUCTION: Slip detection is crucial for achieving stable grasping and subsequent operational tasks. A grasp action is a continuous process that requires information from multiple sources. The success of a specific grasping maneuver is contingent upon the confluence of two factors: the spatial accuracy of the contact and the stability of the continuous process.

METHODS: In this paper, for the task of perceiving grasping results using visual-haptic information, we propose a new method for slip detection, which synergizes visual and haptic information from spatial-temporal dual dimensions. Specifically, the method takes as input a sequence of visual images from a first-person perspective and a sequence of haptic images from a gripper. Then, it extracts time-dependent features of the whole process and spatial features matching the importance of different parts with different attention mechanisms. Inspired by neurological studies, during the information fusion process, we adjusted temporal and spatial information from vision and haptic through a combination of two-step fusion and gate units.

RESULTS AND DISCUSSION: To validate the effectiveness of method, we compared it with traditional CNN net and models with attention. It is anticipated that our method achieves a classification accuracy of 93.59%, which is higher than that of previous works. Attention visualization is further presented to support the validity.

PMID:39569062 | PMC:PMC11576469 | DOI:10.3389/fnbot.2024.1484751

Front Surg. 2024 Nov 6;11:1458569. doi: 10.3389/fsurg.2024.1458569. eCollection 2024.

ABSTRACT

OBJECTIVE: To develop and validate an artificial intelligence diagnostic model for identifying calcified lumbar disc herniation based on lateral lumbar magnetic resonance imaging(MRI).

METHODS: During the period from January 2019 to March 2024, patients meeting the inclusion criteria were collected. All patients had undergone both lumbar spine MRI and computed tomography(CT) examinations, with regions of interest (ROI) clearly marked on the lumbar sagittal MRI images. The participants were then divided into separate sets for training, testing, and external validation. Ultimately, we developed a deep learning model using the ResNet-34 algorithm model and evaluated its diagnostic efficacy.

RESULTS: A total of 1,224 eligible patients were included in this study, consisting of 610 males and 614 females, with an average age of 53.34 ± 10.61 years. Notably, the test datasets displayed an impressive classification accuracy rate of 91.67%, whereas the external validation datasets achieved a classification accuracy rate of 88.76%. Among the test datasets, the ResNet34 model outperformed other models, yielding the highest area under the curve (AUC) of 0.96 (95% CI: 0.93, 0.99). Additionally, the ResNet34 model also exhibited superior performance in the external validation datasets, exhibiting an AUC of 0.88 (95% CI: 0.80, 0.93).

CONCLUSION: In this study, we established a deep learning model with excellent performance in identifying calcified intervertebral discs, thereby offering a valuable and efficient diagnostic tool for clinical surgeons.

PMID:39569028 | PMC:PMC11576459 | DOI:10.3389/fsurg.2024.1458569

Front Microbiol. 2024 Nov 6;15:1495709. doi: 10.3389/fmicb.2024.1495709. eCollection 2024.

ABSTRACT

BACKGROUND: Acute hematogenous osteomyelitis is the most common form of osteomyelitis in children. In recent years, the incidence of osteomyelitis has been steadily increasing. For pediatric patients, clearly describing their symptoms can be quite challenging, which often necessitates the use of complex diagnostic methods, such as radiology. For those who have been diagnosed, the ability to culture the pathogenic bacteria significantly affects their treatment plan.

METHOD: A total of 634 patients under the age of 18 were included, and the correlation between laboratory indicators and osteomyelitis, as well as several diagnoses often confused with osteomyelitis, was analyzed. Based on this, a Transformer-based deep learning model was developed to identify osteomyelitis patients. Subsequently, the correlation between laboratory indicators and the length of hospital stay for osteomyelitis patients was examined. Finally, the correlation between the successful cultivation of pathogenic bacteria and laboratory indicators in osteomyelitis patients was analyzed, and a deep learning model was established for prediction.

RESULT: The laboratory indicators of patients are correlated with the presence of acute hematogenous osteomyelitis, and the deep learning model developed based on this correlation can effectively identify patients with acute hematogenous osteomyelitis. The laboratory indicators of patients with acute hematogenous osteomyelitis can partially reflect their length of hospital stay. Although most laboratory indicators lack a direct correlation with the ability to culture pathogenic bacteria in patients with acute hematogenous osteomyelitis, our model can still predict whether the bacteria can be successfully cultured.

CONCLUSION: Laboratory indicators, as easily accessible medical information, can identify osteomyelitis in pediatric patients. They can also predict whether pathogenic bacteria can be successfully cultured, regardless of whether the patient has received antibiotics beforehand. This not only simplifies the diagnostic process for pediatricians but also provides a basis for deciding whether to use empirical antibiotic therapy or discontinue treatment for blood cultures.

PMID:39568996 | PMC:PMC11578118 | DOI:10.3389/fmicb.2024.1495709

Chem Sci. 2024 Oct 30. doi: 10.1039/d4sc02233e. Online ahead of print.

ABSTRACT

Protein-protein interactions are pivotal in numerous biological processes. The computational design of these interactions facilitates the creation of novel binding proteins, crucial for advancing biopharmaceutical products. With the evolution of artificial intelligence (AI), protein design tools have swiftly transitioned from scoring-function-based to AI-based models. However, many AI models for protein design are constrained by assuming complete unfamiliarity with the amino acid sequence of the input protein, a feature most suited for de novo design but posing challenges in designing protein-protein interactions when the receptor sequence is known. To bridge this gap in computational protein design, we introduce ProBID-Net. Trained using natural protein-protein complex structures and protein domain-domain interface structures, ProBID-Net can discern features from known target protein structures to design specific binding proteins based on their binding sites. In independent tests, ProBID-Net achieved interface sequence recovery rates of 52.7%, 43.9%, and 37.6%, surpassing or being on par with ProteinMPNN in binding protein design. Validated using AlphaFold-Multimer, the sequences designed by ProBID-Net demonstrated a close correspondence between the design target and the predicted structure. Moreover, the model's output can predict changes in binding affinity upon mutations in protein complexes, even in scenarios where no data on such mutations were provided during training (zero-shot prediction). In summary, the ProBID-Net model is poised to significantly advance the design of protein-protein interactions.

PMID:39568891 | PMC:PMC11575592 | DOI:10.1039/d4sc02233e

Int J Biometeorol. 2024 Nov 21. doi: 10.1007/s00484-024-02818-y. Online ahead of print.

ABSTRACT

Ischemic stroke (IS) is one of the top risk factors for death and disability. Meteorological conditions have an effect on IS attack. In this study, we try to develop models of medical meteorological forecast for IS attack based on machine learning and deep learning algorithms. The medical meteorological forecast would be beneficial to public health in IS events prevention and treatment. We collected data on IS attacks and climatology in each day from 18th September 2016 to 31th December 2020 in Haikou. Data on IS attacks were from the number of hospital admissions due to IS attack among general population. The random forest (RF) regression and long short-term memory (LSTM) algorithms were respectively used to develop the predictive model based on meteorological data. Performance of the model was assessed by mean squared error (MSE) and root mean squared error (RMSE). A total of 42849 IS attacks was included in this study. IS attacks were significantly decreased in winter. The pattern of climatological data was observed the regularity in seasons. For the performance of RF regression model, the MSE is 243, and the RMSE is 15.6. For LSTM model, the MSE is 36, and the RMSE is 6. In conclusion, LSTM model is more accurate than RF regression model to predict IS attacks in general population based on meteorological data. LSTM model showed acceptable accuracy for the prediction and could be used as medical meteorological forecast to predict IS attack among population according to local climate.

PMID:39567379 | DOI:10.1007/s00484-024-02818-y

J Am Coll Radiol. 2024 Nov 18:S1546-1440(24)00919-0. doi: 10.1016/j.jacr.2024.11.009. Online ahead of print.

ABSTRACT

OBJECTIVE: To assess whether the implementation of deep learning (DL) computer-aided detection (CAD) that screens for suspected pneumothorax (PTX) on chest radiography (CXR) combined with an electronic notification system (ENS) that simultaneously alerts both the radiologist and the referring clinician would affect time to treatment (TTT) in a real-world clinical practice.

METHODS: In May 2022, a commercial deep learning-based CAD and ENS (DL-CAD-ENS) was introduced for all CXRs at an 818-bed general hospital, with 33 attending doctors and their residents using ENS, while 155 others used only CAD. We used difference-in-differences estimates to compare TTT between the CAD and ENS group and the CAD-only group for the period from January 2018 to April 2022 and from May 2022 to April 2023.

RESULTS: A total of 603,028 CXRs from 140,841 unique patients were included, with a PTX prevalence of 2.0%. There was a significant reduction in TTT for supplemental oxygen therapy for the CAD and ENS group compared with the CAD-only group in the post-implementation period (-143.8 min; 95% confidence interval [CI], -277.8 to -9.9; P = .035). However, there was no significant difference in TTT for other treatments, including aspiration or tube-thoracostomy (14.4 min; 95% CI, -35.0 to 63.9) and consultation with the thoracic and cardiovascular surgery department (86.3 min; 95% CI, -175.1 to 347.6).

CONCLUSION: The introduction of a DL-CAD-ENS reduced the time to initiate oxygen supplementation for patients with PTX.

PMID:39566875 | DOI:10.1016/j.jacr.2024.11.009

Magn Reson Imaging. 2024 Nov 18:110277. doi: 10.1016/j.mri.2024.110277. Online ahead of print.

ABSTRACT

PURPOSE: BUDA-cEPI has been shown to achieve high-quality, high-resolution diffusion magnetic resonance imaging (dMRI) with fast acquisition time, particularly when used in conjunction with S-LORAKS reconstruction. However, this comes at a cost of more complex reconstruction that is computationally prohibitive. In this work we develop rapid reconstruction pipeline for BUDA-cEPI to pave the way for its deployment in routine clinical and neuroscientific applications. The proposed reconstruction includes the development of ML-based unrolled reconstruction as well as rapid ML-based B0 and eddy current estimations that are needed. The architecture of the unroll network was designed so that it can mimic S-LORAKS regularization well, with the addition of virtual coil channels.

METHODS: BUDA-cEPI RUN-UP - a model-based framework that incorporates off-resonance and eddy current effects was unrolled through an artificial neural network with only six gradient updates. The unrolled network alternates between data consistency (i.e., forward BUDA-cEPI and its adjoint) and regularization steps where U-Net plays a role as the regularizer. To handle the partial Fourier effect, the virtual coil concept was also introduced into the reconstruction to effectively take advantage of the smooth phase prior and trained to predict the ground-truth images obtained by BUDA-cEPI with S-LORAKS.

RESULTS: The introduction of the Virtual Coil concept into the unrolled network was shown to be key to achieving high-quality reconstruction for BUDA-cEPI. With the inclusion of an additional non-diffusion image (b-value = 0 s/mm2), a slight improvement was observed, with the normalized root mean square error further reduced by approximately 5 %. The reconstruction times for S-LORAKS and the proposed unrolled networks were approximately 225 and 3 s per slice, respectively.

CONCLUSION: BUDA-cEPI RUN-UP was shown to reduce the reconstruction time by ~88× when compared to the state-of-the-art technique, while preserving imaging details as demonstrated through DTI application.

PMID:39566835 | DOI:10.1016/j.mri.2024.110277

Surv Ophthalmol. 2024 Nov 18:S0039-6257(24)00139-5. doi: 10.1016/j.survophthal.2024.11.007. Online ahead of print.

ABSTRACT

Artificial intelligence (AI)-based methods have been extensively used for the detection and management of various common retinal conditions, but their targeted development for inherited retinal diseases (IRD) is still nascent. In the context of limited availability of retinal subspecialists, genetic testing and genetic counseling, there is a high need for accurate and accessible diagnostic methods. The currently available AI studies, aiming for detection, classification, and prediction of IRD, remain mainly retrospective and include relatively limited numbers of patients due to their scarcity. We summarize the latest findings and clinical implications of machine-learning algorithms in IRD, highlighting the achievements and challenges of AI to assist ophthalmologists in their clinical practice.

PMID:39566565 | DOI:10.1016/j.survophthal.2024.11.007

Clin Imaging. 2024 Nov 13;117:110356. doi: 10.1016/j.clinimag.2024.110356. Online ahead of print.

ABSTRACT

Accurate image interpretation is essential in the field of radiology to the healthcare team in order to provide optimal patient care. This article discusses the use of artificial intelligence (AI) confidence levels to enhance the accuracy and dependability of its radiological diagnoses. The current advances in AI technologies have changed how radiologists and clinicians make the diagnoses of pathological conditions such as aneurysms, hemorrhages, pneumothorax, pneumoperitoneum, and particularly fractures. To enhance the utility of these AI models, radiologists need a more comprehensive understanding of the model's levels of confidence and certainty behind the results they produce. This allows radiologists to make more informed decisions that have the potential to drastically change a patient's clinical management. Several AI models, especially those utilizing deep learning models (DL) with convolutional neural networks (CNNs), have demonstrated significant potential in identifying subtle findings in medical imaging that are often missed by radiologists. It is necessary to create standardized levels of confidence metrics in order for AI systems to be relevant and reliable in the clinical setting. Incorporating AI into clinical practice does have certain obstacles like the need for clinical validation, concerns regarding the interpretability of AI system results, and addressing confusion and misunderstandings within the medical community. This study emphasizes the importance of AI systems to clearly convey their level of confidence in radiological diagnosis. This paper highlights the importance of conducting research to establish AI confidence level metrics that are limited to a specific anatomical region or lesion type. KEY POINT OF THE VIEW: Accurate fracture diagnosis relies on radiologic certainty, where Artificial intelligence (AI), especially convolutional neural networks (CNNs) and deep learning (DL), shows promise in enhancing X-ray interpretation amidst a shortage of radiologists. Overcoming integration challenges through improved AI interpretability and education is crucial for widespread acceptance and better patient outcomes.

PMID:39566394 | DOI:10.1016/j.clinimag.2024.110356

Water Res. 2024 Nov 8;268(Pt B):122766. doi: 10.1016/j.watres.2024.122766. Online ahead of print.

ABSTRACT

Chlorine dosing control is a critical task for health security, complying with drinking water quality standards while achieving customer satisfaction. In Japan, this became more difficult due to decreasing number of technical personnel as a result of declining population and huge retirement of veterans. As experienced-based dosing control still exists and is considered inefficient due to risks of inaccurate dosing and need of experienced manpower, streamlining of operations is needed. This study aims to address these concerns through the development and evaluation of a deep learning model for residual chlorine concentrations capable of forecasting long durations. In this paper, the model is investigated in seasonal timeframe analyzing trends and variations. Two water distributions systems of varying network-simple and complex, are also compared to further analyze model performance and versatility. The model utilizes the past 24 h of flow rate, chlorine level at treatment plant, water temperature, and residual chlorine at private homes as input to predict the 12 h ahead of residual chlorine at private homes evaluated at hourly training lengths of 0.5, 1, 1.5, 2 years. Results revealed that the model achieved high accuracy in predicting hourly residual chlorine with general increasing model error from winter, spring, autumn, and summer due to the progressing instabilities in chlorine concentrations from low to high temperatures. Moreover, smaller system tends to be more unstable incurring lower model performance relative to larger system. In terms of optimal training length, ≥1Y training models are found to have lesser chance of data drift occurrence prospectively reducing model retraining frequency in the future.

PMID:39566282 | DOI:10.1016/j.watres.2024.122766

J Orthop Surg Res. 2024 Nov 20;19(1):777. doi: 10.1186/s13018-024-05265-y.

ABSTRACT

BACKGROUND: Accurate measurement of the hip-knee-ankle (HKA) angle is essential for informed clinical decision-making in the management of knee osteoarthritis (OA). Knee OA is commonly associated with varus deformity, where the alignment of the knee shifts medially, leading to increased stress and deterioration of the medial compartment. The HKA angle, which quantifies this alignment, is a critical indicator of the severity of varus deformity and helps guide treatment strategies, including corrective surgeries. Current manual methods are labor-intensive, time-consuming, and prone to inter-observer variability. Developing an automated model for HKA angle measurement is challenging due to the elaborate process of generating handcrafted anatomical landmarks, which is more labor-intensive than the actual measurement. This study aims to develop a ResNet-based deep learning model that predicts the HKA angle without requiring explicit anatomical landmark annotations and to assess its accuracy and efficiency compared to conventional manual methods.

METHODS: We developed a deep learning model based on the variants of the ResNet architecture to process lower limb radiographs and predict HKA angles without explicit landmark annotations. The classification performance for the four stages of varus deformity (stage I: 0°-10°, stage II: 10°-20°, stage III: > 20°, others: genu valgum or normal alignment) was also evaluated. The model was trained and validated using a retrospective cohort of 300 knee OA patients (Kellgren-Lawrence grade 3 or higher), with horizontal flip augmentation applied to double the dataset to 600 samples, followed by fivefold cross-validation. An extended temporal validation was conducted on a separate cohort of 50 knee OA patients. The model's accuracy was assessed by calculating the mean absolute error between predicted and actual HKA angles. Additionally, the classification of varus deformity stages was conducted to evaluate the model's ability to provide clinically relevant categorizations. Time efficiency was compared between the automated model and manual measurements performed by an experienced orthopedic surgeon.

RESULTS: The ResNet-50 model achieved a bias of - 0.025° with a standard deviation of 1.422° in the retrospective cohort and a bias of - 0.008° with a standard deviation of 1.677° in the temporal validation cohort. Using the ResNet-152 model, it accurately classified the four stages of varus deformity with weighted F1-score of 0.878 and 0.859 in the retrospective and temporal validation cohorts, respectively. The automated model was 126.7 times faster than manual measurements, reducing the total time from 49.8 min to 23.6 sec for the temporal validation cohort.

CONCLUSIONS: The proposed ResNet-based model provides an efficient and accurate method for measuring HKA angles and classifying varus deformity stages without the need for extensive landmark annotations. Its high accuracy and significant improvement in time efficiency make it a valuable tool for clinical practice, potentially enhancing decision-making and workflow efficiency in the management of knee OA.

PMID:39568048 | DOI:10.1186/s13018-024-05265-y

Sci Rep. 2024 Nov 20;14(1):28786. doi: 10.1038/s41598-024-79798-x.

ABSTRACT

Decentralized Federated Learning improves data privacy and eliminates single points of failure by removing reliance on centralized storage and model aggregation in distributed computing systems. Ensuring the integrity of computations during local model training is a significant challenge, especially before sharing gradient updates from each local client. Current methods for ensuring computation integrity often involve patching local models to implement cryptographic techniques, such as Zero-Knowledge Proofs. However, this approach becomes highly complex and sometimes impractical for large-scale models that use techniques such as random dropouts to improve training convergence. These random dropouts create non-deterministic behavior, making it challenging to verify model updates under deterministic protocols. We propose ProxyZKP, a novel framework combining Zero-Knowledge Proofs with polynomial proxy models to provide computation integrity in local training to address this issue. Each local node combines a private model for online deep learning applications and a proxy model that mediates decentralized model training by exchanging gradient updates. The multivariate polynomial nature of proxy models facilitates the application of Zero-Knowledge Proofs. These proofs verify the computation integrity of updates from each node without disclosing private data. Experimental results indicate that ProxyZKP significantly reduces computational load. Specifically, ProxyZKP achieves proof generation times that are 30-50% faster compared to established methods like zk-SNARKs and Bulletproofs. This improvement is largely due to the high parallelization potential of the univariate polynomial decomposition approach. Additionally, integrating Differential Privacy into the ProxyZKP framework reduces the risk of Gradient Inversion attacks by adding calibrated noise to the gradients, while maintaining competitive model accuracy. The results demonstrate that ProxyZKP is a scalable and efficient solution for ensuring training integrity in decentralized federated learning environments, particularly in scenarios with frequent model updates and the need for strong model scalability.

PMID:39567642 | DOI:10.1038/s41598-024-79798-x

Sci Rep. 2024 Nov 20;14(1):28804. doi: 10.1038/s41598-024-80253-0.

ABSTRACT

The aim of this study was to determine the effects of different shank types, tine types, and tractor forward speeds on the power and fuel consumption of chisel plows, stubble cover, soil surface roughness, and penetration resistance. In addition, using the obtained results, the draft power, fuel consumption, soil surface roughness, and soil stubble cover rate were modeled using Deep Learning Neural Network (DLNN) architectures and sensitivity analysis of these models were performed. Four different shank types (rigid, semi-spring, spring, and vibrating) and two different tine types (conventional and winged) were used at three different tractor forward speeds (4.5, 5.4 and 6.3 km.h- 1) were tested to this end. The obtained results indicated that the maximum draft power was achieved with the rigid type shank. The highest soil surface roughness values were observed for the vibrating shank type and winged tine type. Sixteen different network architectures were conducted using deep learning neural network methods and draft power, fuel consumption, soil surface roughness, and percent stubble cover were modeled. Sensitivity analyses were performed to indicate which modeled parameters were more sensitive to the factors using the obtained models. Draft power was modeled with 97.7% accuracy using the DLNN9 network architecture. Additionally, fuel consumption and soil roughness were best modeled with the DLNN13 network architecture, R2 values for those targets were 0.929 and 0.930 respectively. According to the sensitivity analysis, draft power, fuel consumption, soil roughness, and stubble cover rate were significantly affected by changes in the physical properties of the soil.

PMID:39567614 | DOI:10.1038/s41598-024-80253-0