Stopping Threshold Research Articles

Applying machine classifiers to update searches: Analysis from two case studies.

Manual screening of citation records could be reduced by using machine classifiers to remove records of very low relevance. This seems particularly feasible for update searches, where a machine classifier can be trained from past screening decisions. However, feasibility is unclear for broad topics. We evaluate the performance and implementation of machine classifiers for update searches of public health research using two case studies. The first study evaluates the impact of using different sets of training data on classifier performance, comparing recall and screening reduction with a manual screening ‘gold standard’. The second study uses screening decisions from a review to train a classifier that is applied to rank the update search results. A stopping threshold was applied in the absence of a gold standard. Time spent screening titles and abstracts of different relevancy‐ranked records was measured. Results: Study one: Classifier performance varies according to the training data used; all custom‐built classifiers had a recall above 93% at the same threshold, achieving screening reductions between 41% and 74%. Study two: applying a classifier provided a solution for tackling a large volume of search results from the update search, and screening volume was reduced by 61%. A tentative estimate indicates over 25 h screening time was saved. In conclusion, custom‐built machine classifiers are feasible for reducing screening workload from update searches across a range of public health interventions, with some limitation on recall. Key considerations include selecting a training dataset, agreeing stopping thresholds and processes to ensure smooth workflows.

Enabling automation and edge intelligence over resource constraint IoT devices for smart home

Smart home applications are pervasive and have gained popularity due to the overwhelming use of Internet of Things (IoT). The revolution in IoT technologies made homes more convenient, efficient and perhaps more secure. The need to advance smart home technology is necessary at this stage as IoT is abundantly used in automation industry. However, most of the proposed solutions are lacking in certain key areas of the system i.e., high interoperability, data independence, privacy, and optimization in general. The use of machine learning algorithms requires high-end hardware and are usually deployed on servers, where computation is convenient, but at the cost of bandwidth. However, more recently edge AI enabled systems are being proposed to shift the computation burden from the server side to the client side enabling smart devices. In this paper, we take advantage of the edge AI enabled technology to propose a fully featured cohesive system for smart home based on IoT and edge computing. The proposed system makes use of industry standards adopted for fog computing as well as providing robust responses from connected IoT sensors in a typical smart home. The proposed system employs edge devices as a computational platform in terms of reducing energy costs and provides security, while remotely controlling all appliances behind a secure gateway. A case study of human fall detection is evaluated by a custom lightweight deep neural network architecture implemented over the edge device of the proposed framework. The case study was validated using the Le2i dataset. During the training, the early stopping threshold was achieved with 98% accuracy for training set and 94% for validation set. The model size of the network was 6.4 MB which is significantly lower than other networks with similar performance.

Age and Energy Tradeoff for Multicast Networks With Short Packet Transmissions

Age of information (AoI) and energy efficiency (EE), as two important performance metrics in status update systems, usually cannot be simultaneously optimized. This paper investigates the age and energy tradeoff for multicast networks with retransmissions, where each sensed status update is encoded as a short blocklength packet and is broadcasted to multiple destinations via independent and identically distributed error-prone channels. By considering the stopping threshold, the average AoI and EE expressions for stopping at earliest- <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</i> , stopping at preselected- <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</i> , and wait-for-all schemes are derived as a function of the packet length. On this basis, the average AoI-EE ratio is introduced and is minimized to trade off age and energy by optimizing the packet length. Numerical results indicate that among the three schemes, earliest- <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</i> scheme attains the minimum AoI when the packet length is small, and preselected- <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</i> scheme attains the maximum EE when the packet length is large. An optimal packet length can always be found for each scheme to minimize age-energy ratio. Moreover, the selection of stopping threshold <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</i> will also impact the age and energy performance.

Patient preferences for gene therapy in haemophilia: Results from the PAVING threshold technique survey.

ObjectivesThe aim of the Patient preferences to Assess Value IN Gene therapies (PAVING) study was to investigate trade‐offs that adult Belgian people with haemophilia (PWH) A and B are willing to make when choosing between prophylactic factor replacement therapy (PFRT) and gene therapy.MethodsThe threshold technique was used to quantify the minimum acceptable benefit (MAB) of a switch from PFRT to gene therapy in terms of ‘Annual bleeding rate’ (ABR), ‘Chance to stop prophylaxis’ (STOP), and ‘Quality of life’ (QOL). The design was supported by stakeholder involvement and included an educational tool on gene therapy. Threshold intervals were analysed using interval regression models in Stata 16.ResultsA total of 117 PWH completed the survey. Mean thresholds were identified for all benefits, but substantial preference heterogeneity was observed; especially for the STOP thresholds, where the distribution of preferences was bimodal. Time spent on the educational tool and residence were found to impact MAB thresholds. The most accepted (88% of PWH) gene therapy profile investigated in this study comprised of zero bleeds per year (vs. six for PFRT), 90% chance to stop prophylaxis, no impact on QoL, and 10 years of follow‐up on side effects (vs. 30 for PFRT).ConclusionsResults from this study proved the value of educating patients on novel treatments. Moreover, preference heterogeneity for novel treatments was confirmed in this study. In gene therapy decision‐making, preference heterogeneity and the impact of patient education on acceptance should be considered.

Modelling the primary damage in Fe and W: influence of the short-range interactions on the cascade properties: Part 2 – multivariate multiple linear regression analysis of displacement cascades

Neutrons and high energy ions create displacement cascades in materials, which have been simulated using Molecular Dynamics, for many decades now. The breakthrough of this work is to explore in a large statistics of more than 7000 cascades the relation between early cascade morphology and the final primary damage using a multivariate multiple regression analysis. For two energies in Fe and W, the total number of defects, the number of SIA and vacancy clusters and their full size distributions have been characterized using a multivariate multiple linear regression analysis based on 7 descriptors of the primary damage and 3 morphology descriptors. We find that the combination of the volume and the sphericity is significant. This analysis highlights several cascade properties, among them, that the large and spherical cascades create less defects and in particular, less mono defects than small and fragmented ones. 13 interatomic potentials differing either by their equilibrium part or the way they were hardened have been included in this study and the multivariate analysis shows that the choice of potential has a limited influence on the total number of defects but a large one on the number of mono vacancies. On average, soft potentials create cascades of larger volume, smaller sphericity and producing more defects than hard potentials. Finally, the formation of vacancy clusters is different in Fe than in W. In Fe, the fraction of vacancies in clusters is larger than that of SIAs and larger vacancy clusters are created than SIA clusters. In W, it is the opposite. The reasons are the differences of stopping power and threshold displacement energies, which result in different spatial distributions of open volumes that form during the expansion stage of the cascade.

ON PROFITABILITY OF NAKAMOTO DOUBLE SPEND

Nakamoto doublespend strategy, described in Bitcoin foundational article, leads to total ruin with positive probability. The simplest strategy that avoids this risk incorporates a stopping threshold when success is unlikely. We compute the exact profitability and the minimal double spend that is profitable for this strategy. For a given amount of the transaction, we determine the minimal number of confirmations to be requested by the recipient that makes the double-spend strategy non-profitable. This number of confirmations is only 1 or 2 for average transactions and for a small relative hashrate of the attacker. This is substantially lower than the original Nakamoto number, which is about six confirmations and is widely used. Nakamoto analysis is only based on the success probability of the attack instead of on a profitability analysis that we carry out.

Cognitive models of optimal sequential search with recall

Many everyday decisions require sequential search, according to which available choice options are observed one at a time, with each observation involving some cost to the decision maker. In these tasks, decision makers need to trade-off the chances of finding better options with the cost of search. Optimal strategies in such tasks involve threshold decision rules, which terminate the search as soon as an option exceeding a reward value is found. Threshold rules can be seen as special cases of well-known algorithmic decision processes, such as the satisficing heuristic. Prior work has found that decision makers do use threshold rules, however the stopping thresholds observed in data are typically smaller than the (expected value maximizing) optimal threshold. We put forward an array of cognitive models and use parametric model fits on participant-level search data to examine why decision makers adopt seemingly suboptimal thresholds. We find that people's behavior is consistent with optimal search if we allow participants to display risk aversion, psychological effort cost, and decision error. Thus, decision makers appear to be able to search in a resource-rational manner that maximizes stochastic risk averse utility. Our findings shed light on the psychological factors that guide sequential decision making, and show how threshold models can be used to describe both computational and algorithmic aspects of search behavior.

PBI55 Patient Preferences for GENE Therapy in Hemophilia: Results from the Paving Threshold Technique Survey

The aim of the Patient preferences to Assess Value IN Gene therapies (PAVING) study was to investigate trade-offs that adult Belgian hemophilia A and B patients are willing to make when choosing between prophylactic factor replacement therapy (PFRT) and gene therapy. The threshold technique was used to quantify the minimum acceptable benefit (MAB) of a switch from PFRT to gene therapy for ‘Annual bleeding rate’ (ABR), ‘Chance to stop prophylaxis’ (STOP), and ‘Quality of life’ (QOL). ‘Time that side effects have been studied’ was also included as an attribute. An educational tool provided patients with information on gene therapy. The design was supported by stakeholder (including HTA, payers and patients) involvement. Threshold intervals were analyzed using interval regression models in Stata 16. A total of 117 patients completed the survey. Mean thresholds were identified for all benefits (ABR, STOP and QOL), but substantial preference heterogeneity was observed. Heterogeneity in STOP thresholds was found to be most extreme and the majority of patients could be classified in two opposite groups. Time spent on the educational tool and residence were found to impact MAB thresholds. Most patients (88%) would accept gene therapy if it would result in 0 bleeds per year (vs. 6 for PFRT), 90% chance to stop prophylaxis, had no impact on QOL, and if there was 10 years of follow-up on side effects (vs. 30 for PFRT). Results from this study confirmed the value of educating patients on novel treatments (e.g. through educational tools) prior to measuring their preferences. Moreover, preference heterogeneity for novel treatments can be expected and was confirmed in this study. In making decisions on gene therapy, regulators, HTA and payers should consider preference heterogeneity and the impact of patient education on acceptance.

Defining a prevalence level to describe the elimination of Lymphatic Filariasis (LF) transmission and designing monitoring & evaluating (M&E) programmes post the cessation of mass drug administration (MDA).

The global decline in prevalence of lymphatic filariasis has been one of the major successes of the WHO’s NTD programme. The recommended strategy of intensive, community-wide mass drug administration, aims to break localised transmission by either reducing the prevalence of microfilaria positive infections to below 1%, or antigen positive infections to below 2%. After the threshold is reached, and mass drug administration is stopped, geographically defined evaluation units must pass Transmission Assessment Surveys to demonstrate that transmission has been interrupted. In this study, we use an empirically parameterised stochastic transmission model to investigate the appropriateness of 1% microfilaria-positive prevalence as a stopping threshold, and statistically evaluate how well various monitoring prevalence-thresholds predict elimination or disease resurgence in the future by calculating their predictive value. Our results support the 1% filaremia prevalence target as appropriate stopping criteria. However, because at low prevalence-levels random events dominate the transmission dynamics, we find single prevalence measurements have poor predictive power for predicting resurgence, which suggests alternative criteria for restarting MDA may be beneficial.

Earthquake response of the train–slab ballastless track–subgrade system: A shaking table test study

Earthquake is one of the major factors that provoke train derailment. To investigate dynamic responses of the train–slab ballastless track–subgrade system and derailment features of the train subjected to earthquake, a first large-scale shaking table test on this system was carried out in China. The loaded seismic wave is an earthquake wave of Zhengzhou Yellow River Bridge Site provided by the China Earthquake Administration. It was observed that the elevation of subgrade and embankment could magnify acceleration responses, and the transverse amplification effect of the elevation in soil was more pronounced than the vertical amplification effect. The acceleration response of a track slab was weaker than that of the top subgrade because of integrality and rigidity of the track slab. The bogie shows a shock-absorbing effect in the transverse direction. When the input amplitude was not larger than 0.14 g, the transverse and the vertical acceleration of the rail increased with rising earthquake amplitudes and fluctuated later. Meanwhile, the current stopping threshold for trains is conservative because the derailment factor was less than 0.8 and the peak lateral displacement of wheels was much smaller than the cross-sectional size of the rail model under a 0.12 g earthquake excitation. The current stopping threshold in codes could be reasonable in view of limited quantity and limited types of loaded earthquake in the test. The critical derailment condition of trains was 0.23 g when the Zhengzhou Yellow River Bridge Site earthquake was loaded, and the displacement analysis showed that it would be a great danger for train safety.

Derailment Behaviors of the Train-Ballasted Track-Subgrade System Subjected to Earthquake Using Shaking Table

Ballasted track is the most common track model in railway system and earthquake would be a great threaten for train running safety. This study investigates derailment features and lateral displacement response of the train-ballasted track-subgrade system subjected to earthquake using 1-g shaking table. The model was loaded using scaled Zhengzhou Huanghe Qiaozhi (ZHQ) earthquake wave, ALS earthquake and CHY004 earthquake which are with variable-characters. It is observed that current stopping threshold for train is conservative because derailment factor is less than 0.8. The derailment factor of train increase with increasing acceleration amplitude and the critical derailment condition of train under ZHQ earthquake, ALS earthquake and CHY004 earthquake is 0.18 g, 0.15 g and 0.15 g, respectively. The lateral displacement of rail is negligible compared with that of wheel. Lower-frequency earthquake would cause larger derailment factor with derailment analysis and displacement analysis. The maximum peak lateral displacement of wheel under critical derailment condition is significant, which means train safety is under a great danger.

Effect of Pharmacologic Prophylaxis on Venous Thromboembolism After Radical Prostatectomy: The PREVENTER Randomized Clinical Trial

BackgroundDirect high-quality evidence is lacking evaluating perioperative pharmacologic prophylaxis (PP) after radical prostatectomy (RP) to prevent venous thromboembolism (VTE) leading to significant practice variation. ObjectiveTo study the impact of in-hospital PP on symptomatic VTE incidence and adverse events after RP at 30 d, with the secondary objective of evaluating overall VTE in a screening subcohort. Design, setting, and participantsA prospective, phase 4, single-center, randomized trial of men with prostate cancer undergoing open or robotic-assisted laparoscopic RP was conducted (July 2017–November 2018). InterventionPP (subcutaneous heparin) plus routine care versus routine care alone. The screening subcohort was offered lower extremity duplex ultrasound at 30 d. Outcomes measurements and statistical analysisThe primary efficacy outcome was symptomatic VTE incidence (pulmonary embolism [PE] or deep venous thrombosis [DVT]). Primary safety outcomes included the incidence of symptomatic lymphocele, hematoma, or bleeding after surgery. Secondary outcomes were overall VTE, estimated blood loss, total surgical drain output, complications, and surveillance imaging bias. Fisher’s exact test and modified Poisson regression were performed. Results and limitationsA total of 501 patients (75% robotic) were randomized and >99% (500/501) completed follow-up. At second interim analysis (N = 445), the symptomatic VTE rate was 2.3% (four PE + DVT and one DVT) for routine care versus 0.9% (one PE + DVT and one DVT) for PP (relative risk 0.40 [95% confidence interval 0.08–2.03], p = 0.3) meeting a futility threshold for early stopping. In the screening subcohort, the overall VTE rate was 3.3% versus 2.4% (p = 0.7). Results were similar at the final analysis (symptomatic VTE: 2.0% vs 0.8%, p = 0.3; overall VTE: 2.9% vs 2.8%, p = 1). No differences were observed in safety or secondary outcomes. All VTE events (seven symptomatic and three asymptomatic) occurred in patients undergoing pelvic lymph node dissection. ConclusionsThis study was not able to demonstrate a statistically significant reduction in symptomatic VTE associated with PP. There was no increase in the development of symptomatic lymphoceles, bleeding, or other adverse events. Given that the event rate was lower than powered for, further research is needed among high-risk patients (Caprini score ≥8) or patients receiving pelvic lymph node dissection. Patient summaryIn this report, we randomized patients undergoing radical prostatectomy to perioperative pharmacologic prophylaxis or routine care alone. We found that pharmacologic prophylaxis did not reduce postoperative symptomatic venous thromboembolism significantly for men at routine risk. Importantly, pharmacologic prophylaxis did not increase adverse events, such as formation of lymphoceles or bleeding, and can safely be implemented when indicated for patients with risk factors undergoing radical prostatectomy.

Calculations on stopping time and return period

Establishing protective projects such as breakwaters and flood barriers is the key to preventing marine disasters caused by extreme sea conditions. One of the core technical problems in implementing these large-scale projects is how to determine the fortification criteria reasonably. In the past, research based on random variables studied the determined state of time for the stochastic process. For the first time, this paper studies the statistical characteristics of ocean environment elements from both the time and space with the real perspective of the stochastic process and introduces the concept of stopping time in stochastic processes into the analysis of storm surge. The relationship between the stopping time and threshold selection for the measured data is discussed. The relationship between the wave front displacement and the time exceeding the threshold $$\lambda$$ is given as $$\tau = \inf \left\{ {t \ge 0:\xi \left( t \right) > \lambda } \right\}$$. When the distribution is symmetrical and has Markov characteristics, there is $$P\left( {\tau \le t} \right) = 2P\left( {\xi \left( t \right) > \lambda } \right)$$. Additionally, the relationship between the calculation and the return period of the wave height is given. It is proved that the stopping time $$N = \inf \left\{ {n \ge 1:X_{n} > x_{0} } \right\}$$ obeys the geometric distribution, and the expectation of the stopping time is the return period in the ocean engineering. Through the analysis of the stopping time parameters, the rationality of applying the Gumbel distribution in extreme sea conditions for calculating return period is given. With the relationship between the service life of marine engineering and the commonly used parameters, the average life expectancy of offshore engineering is: $${\text{ET}} = 5$$.

A real-world assessment of procalcitonin combined with antimicrobial stewardship in a community ICU.

We evaluated the feasibility and impact of PCT-guided antibiotic duration combined with an established antibiotic stewardship program (ASP) in a community hospital intensive care unit (ICU). We implemented daily PCT levels for ICU patients receiving antibiotics. Our protocol recommended stopping antibiotic therapy if PCT met an absolute or relative stopping threshold. We evaluated the adherence to stopping criteria within 48h, antibiotic use [days of therapy (DOT) per 1000 patient-days (PD)], length of stay and ICU-mortality. We performed interrupted time series analysis to compare 24months before and 12months after implementation. A total of 297 antibiotic courses were monitored with PCT in 217 patients. Protocol adherence was 34% (absolute threshold: 39%, relative threshold: 12%). Antibiotic use pre-PCT was 935 DOTs/1000 PDs and post-PCT was 817 DOTs/1000 PDs (RRadj 0.73, 95% CI: 0.62 to 0.86). No statistically significant changes in clinical outcomes were noted. In the context of an established ASP in a community hospital ICU, PCT monitoring was feasible and associated with an adjusted overall decrease of 27% in antibiotic use with no adverse impact on clinical outcomes. Incorporating PCT testing to guide antibiotic duration can be successful if integrated into workflow and paired with ASP guidance.

Frequency-Domain Entropy-Based Blind Support Recovery from Multiple Measurement Vectors

Support recovery from multiple measurement vectors has been regarded as a critical aspect of compressive sensing. Most existing recovery algorithms require the prior knowledge of the sparsity or the noise power, which are unknown or even time-varying in actual applications, to determine the termination condition of the iterative recovery process. Motivated by the entropy concept from information theory, a frequency-domain entropy (FDE)-based blind support recovery algorithm is proposed, where the FDE is employed as the test statistic to determine whether there is a sparse signal remains in the residual signal. Specifically, the statistical distribution of the residual signal is proved to be Gaussian when the recovery iterations are adequately satisfied. Then, the cumulative distribution function of the test statistic and the closed-form expressions of the stop threshold are derived. Our proposed algorithm exploits the merit that the FDE is uncorrelated with the noise power, thus the estimation of the noise power is not required. Simulation results illustrate that the proposed algorithm is robust to different signal-to-noise ratios.

To quit or not to quit in dynamic search.

Searching for targets among similar distractors requires more time as the number of items increases, with search efficiency measured by the slope of the reaction-time (RT)/set-size function. Horowitz and Wolfe (Nature, 394(6693), 575-577, 1998) found that the target-present RT slopes were as similar for "dynamic" as for standard static search, even though the items were randomly reshuffled every 110 ms in dynamic search. Somewhat surprisingly, attempts to understand dynamic search have ignored that the target-absent RT slope was as low (or "flat") as the target-present slope-so that the mechanisms driving search performance under dynamic conditions remain unclear. Here, we report three experiments that further explored search in dynamic versus static displays. Experiment 1 confirmed that the target-absent:target-present slope ratio was close to or smaller than 1 in dynamic search, as compared with being close to or above 2 in static search. This pattern did not change when reward was assigned to either correct target-absent or correct target-present responses (Experiment 2), or when the search difficulty was increased (Experiment 3). Combining analysis of search sensitivity and response criteria, we developed a multiple-decisions model that successfully accounts for the differential slope patterns in dynamic versus static search. Two factors in the model turned out to be critical for generating the 1:1 slope ratio in dynamic search: the "quit-the-search" decision variable accumulated based upon the likelihood of "target absence" within each individual sample in the multiple-decisions process, whilst the stopping threshold was a linear function of the set size and reward manipulation.

Investigation on Experimental Method for Validating Performance of Passive Autocatalytic Hydrogen Recombiner

The startup and stop threshold of hydrogen recombination and the hydrogen removing rate are key performance indexes for the passive autocatalytic hydrogen recombiner (PAR). A straightforward and convenient experimental method for validating the performance of PAR is proposed. Put PAR in an airtight vessel, and hydrogen is feed to the vessel. If the PAR start hydrogen recombination and whole inlet hydrogen concentration variation curve versus time is under given line A, the startup threshold is confirmed to be not greater than A. If the final horizontal segment of the inlet hydrogen concentration variation curve versus time is under given line B, the stop threshold is confirmed to be not greater than B. If PAR realized steady operation status, the average flow rate of hydrogen fed into the vessel is the hydrogen removing rate. Test rig is designed and setup, and the experiment for validating the performance of PARQX-15 PAR is carried out. Successfully, the startup threshold is confirmed to be not greater than 2(v)%, and stop threshold is confirmed to be not greater than 0.5(v)%, and the hydrogen removing rate is greater than 536 g/h. Conclusion can be made that this method is practical and effective.

Early Stopping Criteria for Levenberg-Marquardt Based Neural Network Training Optimization

In this research we train a direct distributed neural network using Levenberg-Marquardt algorithm. In order to prevent overtraining, we proposed correctly recognized image percentage based on early stop condition and conduct the experiments with different stop thresholds for image classification problem. Experiment results show that the best early stop condition is 93% and other increase in stop threshold can lead to decrease in the quality of the neural network. The correct choice of early stop condition can prevent overtraining which led to the training of a neural network with considerable number of hidden neurons.

Quickest Detection for Changes in Maximal kNN Coherence of Random Matrices

This paper addresses the problem of quickest detection of a change in the maximal coherence between columns of an $n\times p$ random matrix based on a sequence of matrix observations having a single unknown change point. The random matrix is assumed to have identically distributed rows and the maximal coherence is defined as the largest of the $p \atopwithdelims ()2$ correlation coefficients associated with any row. Likewise the $k$ nearest neighbor (kNN) coherence is defined as the $k$ -th largest of these correlation coefficients. The forms of the pre- and postchange distributions of the observed matrices are assumed to belong to the family of elliptically contoured densities with sparse dispersion matrices but are otherwise unknown. A nonparametric stopping rule is proposed that is based on the maximal kNN sample coherence between columns of each observed random matrix. This is a summary statistic that is related to a test of existence of a hub vertex in a sample correlation graph having degree at least $k$ . Performance bounds on the delay and false alarm performance of the proposed stopping rule are obtained in the purely high-dimensional regime where $p\rightarrow \infty$ and $n$ is fixed. When the prechange dispersion matrix is diagonal it is shown that, among all functions of the proposed summary statistic, the proposed stopping rule is asymptotically optimal under a minimax quickest change detection model as the stopping threshold approaches infinity. The theory developed also applies to sequential hypothesis testing and fixed sample size tests.

EEG Classification with a Sequential Decision-Making Method in Motor Imagery BCI.

To develop subject-specific classifier to recognize mental states fast and reliably is an important issue in brain-computer interfaces (BCI), particularly in practical real-time applications such as wheelchair or neuroprosthetic control. In this paper, a sequential decision-making strategy is explored in conjunction with an optimal wavelet analysis for EEG classification. The subject-specific wavelet parameters based on a grid-search method were first developed to determine evidence accumulative curve for the sequential classifier. Then we proposed a new method to set the two constrained thresholds in the sequential probability ratio test (SPRT) based on the cumulative curve and a desired expected stopping time. As a result, it balanced the decision time of each class, and we term it balanced threshold SPRT (BTSPRT). The properties of the method were illustrated on 14 subjects' recordings from offline and online tests. Results showed the average maximum accuracy of the proposed method to be 83.4% and the average decision time of 2.77[Formula: see text]s, when compared with 79.2% accuracy and a decision time of 3.01[Formula: see text]s for the sequential Bayesian (SB) method. The BTSPRT method not only improves the classification accuracy and decision speed comparing with the other nonsequential or SB methods, but also provides an explicit relationship between stopping time, thresholds and error, which is important for balancing the speed-accuracy tradeoff. These results suggest that BTSPRT would be useful in explicitly adjusting the tradeoff between rapid decision-making and error-free device control.