Partitioning Model Research Articles

Use of a machine learning algorithm with a focus on spinopelvic parameters to predict development of symptomatic tethered cord after initial untethering surgery.

Among patients with a history of prior lipomyelomeningocele repair, an association between increased lumbosacral angle (LSA) and cord retethering has been described. The authors sought to build a predictive algorithm to determine which complex tethered cord patients will develop the symptoms of spinal cord retethering after initial surgical repair with a focus on spinopelvic parameters. An electronic medical record database was reviewed to identify patients with complex tethered cord (e.g., lipomyelomeningocele, lipomyeloschisis, myelocystocele) who underwent detethering before 12 months of age between January 1, 2008, and June 30, 2022. Descriptive statistics were used to characterize the patient population. The Caret package in R was used to develop a machine learning model that predicted symptom development by using spinopelvic parameters. A total of 72 patients were identified (28/72 [38.9%] were male). The most commonly observed dysraphism was lipomyelomeningocele (41/72 [56.9%]). The mean ± SD age at index MRI was 2.1 ± 2.2 months, at which time 87.5% of patients (63/72) were asymptomatic. The mean ± SD lumbar lordosis at the time of index MRI was 23.8° ± 11.1°, LSA was 36.5° ± 12.3°, sacral inclination was 30.4° ± 11.3°, and sacral slope was 23.0° ± 10.5°. Overall, 39.6% (25/63) of previously asymptomatic patients developed new symptoms during the mean ± SD follow-up period of 44.9 ± 47.2 months. In the recursive partitioning model, patients whose LSA increased at a rate ≥ 5.84°/year remained asymptomatic, whereas those with slower rates of LSA change experienced neurological decline (sensitivity 77.5%, specificity 84.9%, positive predictive value 88.9%, and negative predictive value 70.9%). This is the first study to build a machine learning algorithm to predict symptom development of spinal cord retethering after initial surgical repair. The authors found that, after initial surgery, patients who demonstrate a slower rate of LSA change per year may be at risk of developing neurological symptoms.

Semantic-aware automatic extraction method for bottom sediment annotations in raster nautical charts

ABSTRACT The automatic extraction of bottom sediment annotations in large-scale raster nautical charts has limitations, including an imprecise semantic information description and low efficiency. To overcome them, we propose a convolutional neural network (CNN)-based method for the automatic extraction of bottom sediment annotations in raster nautical charts, using image processing techniques to improve it. First, an adaptive chart partitioning model that considers element completeness is constructed. Second, a principle for the unique identification of elements based on spatial conflicts is designed. Finally, a model for accurately extracting semantic information for bottom sediment annotations is established. To evaluate the effectiveness of the proposed method, we implemented a model based on the PyTorch framework and used the PIL library to analyze the results. We also conducted comparative experiments on multiple CNN models to recommend the selection of such models in the proposed method by comparing their classification and recognition performance. The experimental results indicate that (1) the proposed model can achieve high-precision extraction of bottom sediment annotations in raster nautical charts. (2) Furthermore, the proposed model generally has high recognition accuracy and semantic completeness, with better recognition precision than traditional pattern recognition methods.

Numerical investigation on boiling crisis in a full-length 5 × 5 fuel assembly under typical pressurized water reactor conditions

Critical heat flux (CHF) presents a research priority in the Pressurized Water Reactor (PWR) safety, as it can lead to the severe accidents and release radioactive materials. In this study, a wall heat flux partition model, proposed by Demarly [1], was employed within the two-fluid model by Computational Fluid Dynamics (CFD) methodology to investigate the characteristics of CHF under typical PWR conditions. A multi-scale interface model was coupled to consider the interfacial transfers in bulk flow during flow regime transition. The models were validated by using experimental data on subcooled flow boiling, boiling crisis in vertical tubes, and rod bundles. The PWR sub-channel and bundle test (PSBT) benchmark, comprising a full-length 5 × 5 fuel assembly with three different types of spacers, was simulated in this study. The CHF value was evaluated, demonstrating an acceptable accuracy compared to the experimental data, with a deviation of less than 10 % in the selected conditions. The CHF location in the cross section was successfully predicted by CFD approach, while the axial location of CHF was underestimated than experimental measurement. Besides, the impacts of mixing vanes with different deflection angles on boiling heat transfer characteristics were analyzed, revealing the crucial role of mixing vanes in the distribution of vapor and dry-patch area fraction in both the circumferential and flow directions.

A Light Touch: Solar Photocatalysis Detoxifies Oil Sands Process-Affected Waters Prior to Significant Treatment of Naphthenic Acids.

Environmental reclamation of Canada's oil sands tailings ponds is among the single largest water treatment challenges globally. The toxicity of oil sands process-affected water (OSPW) has been associated with its dissolved organics, a complex mixture of naphthenic acid fraction components (NAFCs). Here, we evaluated solar treatment with buoyant photocatalysts (BPCs) as a passive advanced oxidation process (P-AOP) for OSPW remediation. Photocatalysis fully degraded naphthenic acids (NAs) and acid extractable organics (AEO) in 3 different OSPW samples. However, classical NAs and AEO, traditionally considered among the principal toxicants in OSPW, were not correlated with OSPW toxicity herein. Instead, nontarget petroleomic analysis revealed that low-polarity organosulfur compounds, composing <10% of the total AEO, apparently accounted for the majority of waters' toxicity to fish, as described by a model of tissue partitioning. These findings have implications for OSPW release, for which a less extensive but more selective treatment may be required than previously expected.

On the generality of the finite element modeling physical fields in biological systems by the multiscale smeared concept (Kojic transport model)

The biomechanical and biochemical processes in the biological systems of living organisms are extremely complex. Advances in understanding these processes are mainly achieved by laboratory and clinical investigations, but in recent decades they are supported by computational modeling. Besides enormous efforts and achievements in this modeling, there still is a need for new methods that can be used in everyday research and medical practice. In this report, we give a view of the generality of the finite element methodology introduced by the first author and supported by his collaborators. It is based on the multiscale smeared physical fields, termed as Kojic Transport Model (KTM), published in several journal papers and summarized in a recent book (Kojic et al., 2022) [1]. We review relevant literature to demonstrate the distinctions and advantages of our methodology and indicate possible further applications. We refer to our published results by a selection of a few examples which include modeling of partitioning, blood flow, molecular transport within the pancreas, multiscale-multiphysics model of coupling electrical field and ion concentration, and a model of convective-diffusive transport within the lung parenchyma. Two new examples include a model of convective-diffusive transport within a growing tumor, and drug release from nanofibers with fiber degradation.

Robust land surface temperature record for north China over the past 21,000 years.

Numerous proxy reconstructions have provided general insight into late Quaternary East Asian Monsoon variability. However, challenges persist in precisely assessing absolute temperature impacts on proxy variations. Here, we use two independent paleothermometers, based on bacterial membrane lipids and clumped isotopes of snail shells, in the same section of the western Chinese Loess Plateau to establish a robust land surface temperature record spanning the past approximately 21,000 years. Our independent temperature records consistently reveal (i) similar land surface temperatures between the Last Glacial Maximum and late Holocene and (ii) a gradual cooling Holocene, which contrasts with the climate model predictions. We propose that changes in soil moisture availability over the deglaciation modulates the land surface temperature recorded by the proxies. A land surface energy partitioning model confirms this mechanism, suggesting that effects of soil moisture availability should be properly considered when comparing proxy records with climate model outputs.

Numerical Simulation of Macro-Segregation Phenomena in Transition Blooms with Various Carbon Contents

This paper presents a numerical simulation of the steel grade transition from the ladle nozzle to the solidification end of the bloom. The simulation is based on models encompassing fluid flow, solidification, heat transfer, an electromagnetic field, and solute transport. To validate the accuracy of the steel grade transition model, transition blooms of high-carbon steel are sampled. Subsequently, the model is applied to investigating the steel grade transition between medium-carbon steel and low-carbon steel. The findings indicate that the regions exhibiting significant differences between their molten steel flow velocity and bloom casting speed in the strand model are primarily concentrated within 1 m below the meniscus. Additionally, the mushy zone in the strand model possesses a substantial volume. Solute elements continuously permeate the liquid phase from the solid phase through the mushy zone. Consequently, the distribution of solute elements in the transition bloom is primarily influenced by the molten steel flow in the tundish and macro-segregation in the casting process. The segregation degree of each solute element varies among grades with different carbon contents. In the austenite phase, the segregation degree of each element follows the order C &gt; Si &gt; Mo &gt; Mn &gt; Cr &gt; Ni, while in the ferrite phase, the segregation degree is ordered as C &gt; Si = Mn. Considering macro-segregation, the transition bloom partition model proves to be more stringent than the original partition method. This results in longer transition blooms when a significant difference exists between the new and old grades. For example, in Scheme 1, the original plan transition bloom length is 8.88 m, whereas the new plan transition bloom length is 10.88 m. Similarly, in Scheme 2, the original plan transition bloom length is 34.64 m, and the new plan transition bloom length is 35.16 m. Conversely, shorter partition intervals occur when there is an overlap in the composition of the new and old grades. In Scheme 3, the original plan partition interval for the new and old grades is 4.08 m, while the new plan partition interval is reduced to 0.94 m.

Optimal Model Partitioning with Low-Overhead Profiling on the PIM-based Platform for Deep Learning Inference

Recently Processing-in-Memory (PIM) has become a promising solution to achieve energy-efficient computation in data-intensive applications by placing computation near or inside the memory. In most Deep Learning (DL) frameworks, a user manually partitions a model’s computational graph (CG) onto the computing devices by considering the devices’ capability and the data transfer. The Deep Neural Network (DNN) models become increasingly complex for improving accuracy; thus, it is exceptionally challenging to partition the execution to achieve the best performance, especially on a PIM-based platform requiring frequent offloading of large amounts of data. This article proposes two novel algorithms for DL inference to resolve the challenge: low-overhead profiling and optimal model partitioning. First, we reconstruct CG by considering the devices’ capability to represent all the possible scheduling paths. Second, we develop a profiling algorithm to find the required minimum profiling paths to measure all the node and edge costs of the reconstructed CG. Finally, we devise the model partitioning algorithm to get the optimal minimum execution time using the dynamic programming technique with the profiled data. We evaluated our work by executing the BERT, RoBERTa, and GPT-2 models on the ARM multicores with the PIM-modeled FPGA platform with various sequence lengths. For three computing devices in the platform, i.e., CPU serial/parallel and PIM executions, we could find all the costs only in four profile runs, three for node costs and one for edge costs. Also, our model partitioning algorithm achieved the highest performance in all the experiments over the execution with manually assigned device priority and the state-of-the-art greedy approach.

Factors Impacting Survival After Transarterial Radioembolization in Patients with Unresectable Intrahepatic Cholangiocarcinoma: A Combined Analysis of the Prospective CIRT Studies

PurposeTransarterial radioembolization (TARE) with Yttrium-90 resin microspheres is a treatment option for patients with intrahepatic cholangiocarcinoma (ICC). However, optimising the timing of TARE in relation to systemic therapies and patient selection remains challenging. We report here on the effectiveness, safety, and prognostic factors associated with TARE for ICC in a combined analysis of the prospective observational CIRT studies (NCT02305459 and NCT03256994).MethodsA combined analysis of 174 unresectable ICC patients enrolled between 2015 and 2020 was performed. Patient characteristics and treatment-related data were collected at baseline; adverse events and time-to-event data (overall survival [OS], progression-free survival [PFS] and hepatic PFS) were collected at every follow-up visit. Log-rank tests and a multivariable Cox proportional hazard model were used to identify prognostic factors.ResultsPatients receiving a first-line strategy of TARE in addition to any systemic treatment had a median OS and PFS of 32.5 months and 11.3 months. Patients selected for first-line TARE alone showed a median OS and PFS of 16.2 months and 7.4 months, whereas TARE as 2nd or further treatment-line resulted in a median OS and PFS of 12 and 9.3 months (p = 0.0028), and 5.1 and 3.5 months (p = 0.0012), respectively. Partition model dosimetry was an independent predictor for better OS (HR 0.59 [95% CI 0.37–0.94], p = 0.0259). No extrahepatic disease, no ascites, and < 6.1 months from diagnosis to treatment were independent predictors for longer PFS.ConclusionThis combined analysis indicates that in unresectable ICC, TARE in combination with any systemic treatment is a promising treatment option.Level of evidence: level 3, Prospective observationalGraphic

Comparisons of rainfall partitioning and canopy interception modelling in pure and mixed forests on the Chinese Loess Plateau

AbstractRainfall partitioning is a key process in forest ecosystems that affects the forest hydrological cycle and ecological services. However, little is known about the rainfall partitioning in Pinus tabuliformis (PT), Robinia pseudoacacia (RP), and Pinus tabuliformis×Robinia pseudoacacia (PT × RP) forests, which are widely distributed on the Chinese Loess Plateau (CLP) for controlling soil and water losses. The objectives of this study were to compare the rainfall partitioning and conduct interception modelling in PT, RP, and PT × RP forests, using data from the Ansai and Changwu sites on the CLP. The sites feature similar weather conductions and both pure and mixed forests of the same age. The tree density of PT × RP was moderate at the Ansai site but higher at the Changwu site compared to PT and RP. We quantitatively measured the throughfall (Tf), stemflow (Sf), and interception loss (Ic) for the forests at the two sites during the growing season (May–October, 2022), and simulated Ic using the Revised Gash Analytical Model (RGAM) and Liu model. The results showed that measured Ic for PT × RP (26.0% of the total gross rainfall, Pg) was larger than for RP (22.0% of Pg) but smaller than for PT (27.8% of Pg) at the Ansai site, and vice versa for the Tf and Sf (RP &gt; PT × RP &gt; PT), while the measured Ic for PT × RP (33.7% of Pg) at the Changwu site was larger than both for PT (31.1% of Pg) and RP (19.5% of Pg). The Tf and Sf for PT × RP was smaller than for PT and RP. Given the similar weather conditions, differences in Ic for the pure and mixed forests at each site were attributed to canopy storage capacity, wet canopy evaporation rate, and tree density. Additionally, the RGAM and the Liu model effectively simulated and captured the differences in Ic for PT, RP, and PT × RP. However, the RGAM performed better with a relative error less than −7.3%, Nash‐Sutcliffe model efficiency ranging from 0.82 to 0.95, and root mean square error less than 0.5 mm. According to the RGAM, evaporation from the wet canopy during rainfall events and after rainfall cessation were the two most influential components. The results of this study suggest that mixed forest can optimize rainfall partitioning compared to pure forests, but a reasonable density should be adopted to increase rainwater replenishment to the forest floor.

Predicting brain age using partition modeling strategy and atlas-based attentional enhancement in the Chinese population.

As a biomarker of human brain health during development, brain age is estimated based on subtle differences in brain structure from those under typical developmental. Magnetic resonance imaging (MRI) is a routine diagnostic method in neuroimaging. Brain age prediction based on MRI has been widely studied. However, few studies based on Chinese population have been reported. This study aimed to construct a brain age predictive model for the Chinese population across its lifespan. We developed a partition prediction method based on transfer learning and atlas attention enhancement. The participants were separated into four age groups, and a deep learning model was trained for each group to identify the brain regions most critical for brain age prediction. The Atlas attention-enhancement method was also used to help the models focus only on critical brain regions. The proposed method was validated using 354 participants from domestic datasets. For prediction performance in the testing sets, the mean absolute error was 2.218 ± 1.801years, and the Pearson correlation coefficient (r) was 0.969, exceeding previous results for wide-range brain age prediction. In conclusion, the proposed method could provide brain age estimation to assist in assessing the status of brain health.

Reduction of a Real Power System: Turkish Transmission System Case

This article introduces an enhanced Radial, Equivalent, Independent (REI) model for power system reduction, aiming to enhance computational efficiency while preserving accuracy. Unlike the conventional REI method, which divides the system into two regions, the proposed model partitions the grid into more suitable regions, maintaining regional control centers for a more effective reduction process. The model significantly decreases the number of busbars, transmission lines, generators, and load models while retaining the overall geographical structure of the grid. An 8-step application process is outlined for implementing this approach. Validation on a 48-bus power system and the IEEE 39-bus system demonstrates its efficiency. Application on the Turkish transmission system results in a 57% reduction in busbars (from 162 to 70) and a 49% reduction in transmission lines, simplifying the model while achieving an impressive average active load flow error of only 3.51%. The study concludes that the extended REI model offers improved computational efficiency without compromising accuracy, making it promising for applications like network planning, market simulations, and capacity management.

A Bayesian survival treed hazards model using latent Gaussian processes.

Survival models are used to analyze time-to-event data in a variety of disciplines. Proportional hazard models provide interpretable parameter estimates, but proportional hazard assumptions are not always appropriate. Non-parametric models are more flexible but often lack a clear inferential framework. We propose a Bayesian treed hazards partition model that is both flexible and inferential. Inference is obtained through the posterior tree structure and flexibility is preserved by modeling the log-hazard function in each partition using a latent Gaussian process. An efficient reversible jump Markov chain Monte Carlo algorithm is accomplished by marginalizing the parameters in each partition element via a Laplace approximation. Consistency properties for the estimator are established. The method can be used to help determine subgroups as well as prognostic and/or predictive biomarkers in time-to-event data. The method is compared with some existing methods on simulated data and a liver cirrhosis dataset.

Personalized treatment selection via product partition models with covariates.

Precision medicine is an approach for disease treatment that defines treatment strategies based on the individual characteristics of the patients. Motivated by an open problem in cancer genomics, we develop a novel model that flexibly clusters patients with similar predictive characteristics and similar treatment responses; this approach identifies, via predictive inference, which one among a set of treatments is better suited for a new patient. The proposed method is fully model based, avoiding uncertainty underestimation attained when treatment assignment is performed by adopting heuristic clustering procedures, and belongs to the class of product partition models with covariates, here extended to include the cohesion induced by the normalized generalized gamma process. The method performs particularly well in scenarios characterized by considerable heterogeneity of the predictive covariates in simulation studies. A cancer genomics case study illustrates the potential benefits in terms of treatment response yielded by the proposed approach. Finally, being model based, the approach allows estimating clusters' specific response probabilities and then identifying patients more likely to benefit from personalized treatment.

Clustering blood donors via mixtures of product partition models with covariates.

Motivated by the problem of accurately predicting gap times between successive blood donations, we present here a general class of Bayesian nonparametric models for clustering. These models allow for the prediction of new recurrences, accommodating covariate information that describes the personal characteristics of the sample individuals. We introduce a prior for the random partition of the sample individuals, which encourages two individuals to be co-clustered if they have similar covariate values. Our prior generalizes product partition models with covariates (PPMx) models in the literature, which are defined in terms of cohesion and similarity functions. We assume cohesion functions that yield mixtures of PPMx models, while our similarity functions represent the denseness of a cluster. We show that including covariate information in the prior specification improves the posterior predictive performance and helps interpret the estimated clusters in terms of covariates in the blood donation application.

Axial Compression Behavior of Elliptical Concrete-Filled Steel Tube Composite Short Columns with Encased Steel Considering Spherical-Cap Gap

This study explored the axial compression behavior of elliptical concrete-filled steel tubes with encased steel considering spherical-cap gap (GSECFST) composite short columns. We designed 25 composite column specimens by varying the steel tube yield strength (fty), steel skeleton yield strength (fsy), concrete cubic compression strength (fcu), steel tube thickness (t), steel skeleton sectional area (As), the long and short half-axis ratio (a/b), the gap ratio (Xsg), and the slenderness ratio (λ). Based on the nonlinear constitutive models of the materials and the nonlinear contact effect among materials, the ABAQUS 6.20 finite element software established the refined finite element models of these composite short columns. Also, the rationality of the finite element modeling with a spherical-cap gap was verified by comparing it with the existing experimental results. The influence regularity of various parameters on the load (N)-displacement (Δ) curves, bearing capacity, initial stiffness, and ductility of the composite short columns was obtained. In addition, the failure modes, N-Δ process, sectional strain distribution, and gap feature index of the constraint partition model for GSECFST axial compression short columns were revealed. The results showed a weakened interaction between the elliptical steel tube and concrete. Also, the axial compression bearing capacity, initial stiffness, and core concrete ductility were reduced because of the spherical-cap gap. As fty, fsy, fc, and Asy increased, the axial bearing capacity, initial stiffness, and ductility of GSECFST composite short columns improved significantly but decreased with increasing of a/b, Xsg, and λ. When the gap ratio of the spherical crown was less than 4%, the outer steel tubes in the mid-span area of the GSECFST composite short columns buckled in the direction of the elliptical short axis under axial compression, and the concrete expanded outward and crushed. The failures were similar to those of the specimens without the spherical-cap gap. Based on the sectional constraint partition model, we propose the calculation formula of axial compression bearing capacity for GSECFST composite short columns. Consequently, this study is a reference for the elastic-plastic analysis of frame systems with similar composite columns.

Exact Vertex Migration Model of Graph Partitioning Based on Mixed 0–1 Linear Programming and Iteration Algorithm

Exact Vertex Migration Model of Graph Partitioning Based on Mixed 0–1 Linear Programming and Iteration Algorithm

P121 Characterisation of the miRNAome in Crohn’s Disease patients reveals transcriptomic changes associated with tissue composition and drug treatment

Abstract Background Gene regulation plays a major role in Crohn's Disease (CD) susceptibility and drug response. MicroRNA (miRNA)-mediated mechanisms tune gene expression at the post-transcriptional level. MiRNAs are 20-to-25-nt long non-coding RNAs that repress protein expression by direct inhibition or cleavage of the targets. Evaluating the miRNAome is emerging as a promising tool to clarify pathogenesis and identify promising biomarkers. We characterised the regulatory miRNAome across disease subtypes, gut tissue locations and infliximab treatment in CD. Methods We generated 120 transcriptomic profiles from terminal ileum and left colon biopsies from Hospital Universitario de la Princesa (Madrid). This cohort is composed of 10 patients with active endoscopy, 10 quiescent CD cases, and 10 controls (30 in total). In a balanced setting, half of the samples were incubated with infliximab for 18 hours at 37ºC and the rest with basal medium (Table 1). We carried out an exploratory analysis to gain insights into the determinants of miRNAome variability, including variance partition analyses and linear mixed modelling to detect significantly altered miRNAs. We also harnessed results from several publications including GSE16879 dataset to perform a comparative study with infliximab response signatures. Results Tissue location is the primary determinant of miRNA expression variability. The Variance Partition model identified 30 top miRNAs whose variability is maximised across regions. However, patient-specific effects also play a relevant role. We also identified 96 differentially expressed miRNAs between terminal ileum and left colon, with an overrepresentation of molecules involved in stress and developmental pathways (Fig.1). Among these, we find miR200s and miR429, that are associated with suppression of epithelial-mesenchymal transition and are up-regulated in colon more than in ileum. Down-regulation of miR196 is also shown in terminal ileum. Regarding the effects of infliximab treatment, we detected nine deregulated miRNAs which are associated with cellular response pathways. These include miR200s and miR192, which targets NOD2 and down-regulates NF-κB. Our results show an up-regulation of miR192 upon infliximab incubation (log2FC=2.37), which is associated with a reduction in TNFα expression and, consequently, with induced overexpression of miR192. We conclude that this miRNA is a potential target for therapy development. Conclusion In short, tissue location, and to a lesser extent disease subtype, are the main determinants of miRNAome in CD. A few treatment-responsive miRNAs like miR192 and miR200s are identified, in spite of detection challenges, emphasizing the need to improve our comprehension of the miRNA dynamics associated with drug response.

Predictive Dosimetry and Outcomes of Hepatocellular Carcinoma Treated by Yttrium-90 Resin Microsphere Radioembolization: A Retrospective Analysis Using Technetium-99m Macroaggregated Albumin SPECT/CT and Planning Software

PurposeTo characterize estimated mean absorbed tumor dose (ADT), objective response (OR), and estimated target dose of hepatocellular carcinoma (HCC) after resin microsphere yttrium-90 (90Y) radioembolization using partition dosimetry. Materials and MethodsIn this retrospective, single-center study, multicompartment dosimetry of index tumors receiving 90Y radioembolization between October 2015 and June 2022 was performed using a commercial software package and pretreatment technetium-99m macroaggregated albumin single photon emission computed tomography (SPECT)/computed tomography (CT). In total, 101 patients with HCC underwent 102 treatments of 127 index tumors. Patients underwent imaging every 2–3 months after treatment to determine best response per modified Response Evaluation Criteria in Solid Tumors (mRECIST). Best response was defined as the greatest response category per mRECIST and categorized as OR or nonresponse (NR). A Cox proportional hazards model evaluated the probability of tumor OR and progression-free survival using ADT. ResultsThe median follow-up period was 148 days (interquartile range [IQR], 92–273 days). The median ADT of OR was 141.9 Gy (IQR, 89.4–215.8 Gy) compared with the median ADT of NR treatments of 70.8 Gy (IQR, 42.0–135.3 Gy; P < .001). Only ADT was predictive of response (hazard ratio = 2.79 [95% confidence interval {CI}: 1.44–5.40]; P = .003). At 6 months, an ADT of 157 Gy predicted 90.0% (95% CI: 41.3%–98.3%) probability of OR. At 1 year, an ADT of 157 Gy predicted 91.6% (95% CI: 78.3%–100%) probability of progression-free survival. Partition modeling and delivered activity were predictive of progression (P = .021 and P = .003, respectively). ConclusionsFor HCC treated with resin microspheres, tumors receiving higher ADT exhibited higher rates of OR. An ADT of 157 Gy predicted 90.0% OR at 6 months.

Multi-Compression Scale DNN Inference Acceleration based on Cloud-Edge-End Collaboration

Edge intelligence has emerged as a promising paradigm to accelerate DNN inference by model partitioning, which is particularly useful for intelligent scenarios that demand high accuracy and low latency. However, the dynamic nature of the edge environment and the diversity of end devices pose a significant challenge for DNN model partitioning strategies. Meanwhile, limited resources of the edge server make it difficult to manage resource allocation efficiently among multiple devices. In addition, most of the existing studies disregard the different service requirements of the DNN inference tasks, such as its high accuracy-sensitive or high latency-sensitive. To address these challenges, we propose a Multi-Compression Scale DNN Inference Acceleration (MCIA) based on cloud-edge-end collaboration. We model this problem as a mixed-integer multi-dimensional optimization problem, jointly optimizing the DNN model version choice, the partitioning choice, and the allocation of computational and bandwidth resources to maximize the tradeoff between inference accuracy and latency depending on the property of the tasks. Initially, we train multiple versions of DNN inference models with different compression scales in the cloud, and deploy them to end devices and edge server. Next, a deep reinforcement learning-based algorithm is developed for joint decision making of adaptive collaborative inference and resource allocation based on the current multi-compression scale models and the task property. Experimental results show that MCIA can adapt to heterogeneous devices and dynamic networks, and has superior performance compared with other methods.