arXiv subjects

Han Yuan

Publications and source records attributed to Han Yuan.

36 records · Page 2Linked to original sources

arXiv2022

Balanced background and explanation data are needed in explaining deep learning models with SHAP: An empirical study on clinical decision making

Objective: Shapley additive explanations (SHAP) is a popular post-hoc technique for explaining black box models. While the impact of data imbalance on predictive models has been extensively studied, it remains largely unknown with respect to SHAP-based model explanations. This study sought to investigate the effects of data imbalance on SHAP explanations for deep learning models, and to propose a strategy to mitigate these effects. Materials and Methods: We propose to adjust class distributions in the background and explanation data in SHAP when explaining black box models. Our data balancing strategy is to compose background data and explanation data with an equal distribution of classes. To evaluate the effects of data adjustment on model explanation, we propose to use the beeswarm plot as a qualitative tool to identify "abnormal" explanation artifacts, and quantitatively test the consistency between variable importance and prediction power. We demonstrated our proposed approach in an empirical study that predicted inpatient mortality using the Medical Information Mart for Intensive Care (MIMIC-III) data and a multilayer perceptron. Results: Using the data balancing strategy would allow us to reduce the number of the artifacts in the beeswarm plot, thus mitigating the negative effects of data imbalance. Additionally, with the balancing strategy, the top-ranked variables from the corresponding importance ranking demonstrated improved discrimination power. Discussion and Conclusion: Our findings suggest that balanced background and explanation data could help reduce the noise in explanation results induced by skewed data distribution and improve the reliability of variable importance ranking. Furthermore, these balancing procedures improve the potential of SHAP in identifying patients with abnormal characteristics in clinical applications.

cs.LG↗

arXiv2021

Deep learning for temporal data representation in electronic health records: A systematic review of challenges and methodologies

Objective: Temporal electronic health records (EHRs) can be a wealth of information for secondary uses, such as clinical events prediction or chronic disease management. However, challenges exist for temporal data representation. We therefore sought to identify these challenges and evaluate novel methodologies for addressing them through a systematic examination of deep learning solutions. Methods: We searched five databases (PubMed, EMBASE, the Institute of Electrical and Electronics Engineers [IEEE] Xplore Digital Library, the Association for Computing Machinery [ACM] digital library, and Web of Science) complemented with hand-searching in several prestigious computer science conference proceedings. We sought articles that reported deep learning methodologies on temporal data representation in structured EHR data from January 1, 2010, to August 30, 2020. We summarized and analyzed the selected articles from three perspectives: nature of time series, methodology, and model implementation. Results: We included 98 articles related to temporal data representation using deep learning. Four major challenges were identified, including data irregularity, data heterogeneity, data sparsity, and model opacity. We then studied how deep learning techniques were applied to address these challenges. Finally, we discuss some open challenges arising from deep learning. Conclusion: Temporal EHR data present several major challenges for clinical prediction modeling and data utilization. To some extent, current deep learning solutions can address these challenges. Future studies can consider designing comprehensive and integrated solutions. Moreover, researchers should incorporate additional clinical domain knowledge into study designs and enhance the interpretability of the model to facilitate its implementation in clinical practice.

cs.LG↗

arXiv2021

AutoScore-Imbalance: An interpretable machine learning tool for development of clinical scores with rare events data

Background: Medical decision-making impacts both individual and public health. Clinical scores are commonly used among a wide variety of decision-making models for determining the degree of disease deterioration at the bedside. AutoScore was proposed as a useful clinical score generator based on machine learning and a generalized linear model. Its current framework, however, still leaves room for improvement when addressing unbalanced data of rare events. Methods: Using machine intelligence approaches, we developed AutoScore-Imbalance, which comprises three components: training dataset optimization, sample weight optimization, and adjusted AutoScore. All scoring models were evaluated on the basis of their area under the curve (AUC) in the receiver operating characteristic analysis and balanced accuracy (i.e., mean value of sensitivity and specificity). By utilizing a publicly accessible dataset from Beth Israel Deaconess Medical Center, we assessed the proposed model and baseline approaches in the prediction of inpatient mortality. Results: AutoScore-Imbalance outperformed baselines in terms of AUC and balanced accuracy. The nine-variable AutoScore-Imbalance sub-model achieved the highest AUC of 0.786 (0.732-0.839) while the eleven-variable original AutoScore obtained an AUC of 0.723 (0.663-0.783), and the logistic regression with 21 variables obtained an AUC of 0.743 (0.685-0.800). The AutoScore-Imbalance sub-model (using down-sampling algorithm) yielded an AUC of 0. 0.771 (0.718-0.823) with only five variables, demonstrating a good balance between performance and variable sparsity. Conclusions: The AutoScore-Imbalance tool has the potential to be applied to highly unbalanced datasets to gain further insight into rare medical events and to facilitate real-world clinical decision-making.

cs.LG↗

arXiv2021

AutoScore-Survival: Developing interpretable machine learning-based time-to-event scores with right-censored survival data

Scoring systems are highly interpretable and widely used to evaluate time-to-event outcomes in healthcare research. However, existing time-to-event scores are predominantly created ad-hoc using a few manually selected variables based on clinician's knowledge, suggesting an unmet need for a robust and efficient generic score-generating method. AutoScore was previously developed as an interpretable machine learning score generator, integrated both machine learning and point-based scores in the strong discriminability and accessibility. We have further extended it to time-to-event data and developed AutoScore-Survival, for automatically generating time-to-event scores with right-censored survival data. Random survival forest provides an efficient solution for selecting variables, and Cox regression was used for score weighting. We illustrated our method in a real-life study of 90-day mortality of patients in intensive care units and compared its performance with survival models (i.e., Cox) and the random survival forest. The AutoScore-Survival-derived scoring model was more parsimonious than survival models built using traditional variable selection methods (e.g., penalized likelihood approach and stepwise variable selection), and its performance was comparable to survival models using the same set of variables. Although AutoScore-Survival achieved a comparable integrated area under the curve of 0.782 (95% CI: 0.767-0.794), the integer-valued time-to-event scores generated are favorable in clinical applications because they are easier to compute and interpret. Our proposed AutoScore-Survival provides an automated, robust and easy-to-use machine learning-based clinical score generator to studies of time-to-event outcomes. It provides a systematic guideline to facilitate the future development of time-to-event scores for clinical applications.

cs.LG↗

arXiv2019

Elastic deep learning in multi-tenant GPU cluster

We study how to support elasticity, i.e., the ability to dynamically adjust the parallelism (number of GPUs), for deep neural network (DNN) training. Elasticity can benefit multi-tenant GPU cluster management in many ways, e.g., achieving various scheduling objectives (e.g., job throughput, job completion time, GPU efficiency) according to cluster load variations, maximizing the use of transient idle resources, performance profiling, job migration, and straggler mitigation. However, existing parallelism adjustment strategies incur high overheads, which hinder many applications from making effective use of elasticity. We propose EDL to enable low-overhead elastic deep learning with a simple API. We present techniques that are necessary to reduce the overhead of parallelism adjustments, such as stop-free scaling and dynamic data pipeline. We also demonstrate that EDL can indeed bring significant benefits to the above-listed applications in GPU cluster management.

cs.DC↗

arXiv2017

Testing the nature of neutrinos from four-body $τ$ decays

This paper is designed to discuss four-body lepton number violating tau decay. We study the processes $τ^+ \to e^+ e^+ π^- ν_τ$ and $τ^+ \to e^+ e^+ π^- ν_e$ to determine the nature of neutrino. The first process violates lepton number by two units which can only happen through an internal Majorana. The second one conserves lepton number but violates lepton flavor which can take place with both Majorana neutrino and Dirac neutrino. We calculate their branching ratio $Br$ and differential branching ratio $dBr/dE_π$ to distinguish Majorana neutrino vs. Dirac neutrino. We also propose the possibility of experiment to perform this detection.

hep-ph↗

arXiv2017

Four-body decays of $B$ meson with lepton number violation

This paper is designed to calculate the branching ratio of four-body decays of $B$ meson with lepton number changed by 2. With the new experimental data limit to lepton-number violation processes, we update the upper limits of mixing parameters between heavy Majorana neutrino and charged leptons. Afterwards, we calculate the branching ratio of $B^0(P)\rightarrow D^{*-}(P_1)\ell^+_1(P_2)\ell^+_2(P_3)M_2^-(P_4)$ using the updated parameters. It is found that the most hopeful decay channel is $B^0(P)\rightarrow D^{*-}(P_1)e^+_1(P_2)e^+_2(P_3)ρ^-(P_4)$ or $B^0(P)\rightarrow D^{*-}(P_1)e(μ)^+_1(P_2)μ(e)^+_2(P_3)ρ^-(P_4)$, whose branching ratio can reach about $10^{-4}$ with heavy Majorana neutrino mass range around $2~\mathrm{GeV}$.

hep-ph↗

arXiv2017

Study of the excited $1^-$ charm and charm-strange mesons

We give a systematical study on the recently reported excited charm and charm-strange mesons with potential $1^-$ spin-parity, including the $D^*_{s1}(2700)^+$, $D^*_{s1}(2860)^+$, $D^*(2600)^0$, $D^*(2650)^0$, $D^*_1(2680)^0$ and $D^*_1(2760)^0$. The main strong decay properties are obtained by the framework of Bethe-Salpeter (BS) methods. Our results reveal that the two $1^-$ charm-strange mesons can be well described by the further $2^3\!S_1$-$1^3\!D_1$ mixing scheme with a mixing angle of $8.7^{+3.9}_{-3.2}$ degrees. The predicted decay ratio $\frac{\mathcal{B}(D^*K)}{\mathcal{B}(D~K)}$ for $D^*_{s1}(2860)$ is $0.62^{+0.22}_{-0.12}$.~$D^*(2600)^0$ can also be explained as the $2^3\!S_1$ predominant state with a mixing angle of $-(7.5^{+4.0}_{-3.3})$ degrees. Considering the mass range, $D^*(2650)^0$ and $D^*_1(2680)^0$ are more likely to be the $2^3\!S_1$ predominant states, although the total widths under both the $2^3\!S_1$ and $1^3\!D_1$ assignments have no great conflict with the current experimental data. The calculated width for LHCb $D^*_1(2760)^0$ seems about 100 \si{MeV} larger than experimental measurement if taking it as $1^3\!D_1$ or $1^3\!D_1$ dominant state $c\bar u$. The comparisons with other calculations and several important decay ratios are also present. For the identification of these $1^-$ charm mesons, further experimental information, such as $\frac{\mathcal{B}(Dπ)}{\mathcal{B}(D^*π)}$ are necessary.

hep-ph↗

arXiv2016

Decays of $B$, $B_s$ and $B_c$ to $D$-wave heavy-light mesons

We study the weak decays of $\bar{B}_{(s)}$ and $B_c$ into $D$-wave heavy-light mesons, including $J^P=2^-$ ($D_{(s)2},D'_{(s)2},B_{(s)2}, B'_{(s)2}$) and $3^-$~($D^*_{(s)3}, B^*_{(s)3}$) states. The weak decay hadronic matrix elements are achieved based on the instantaneous Bethe-Salpeter method. The branching ratios for $\bar{B}$ decays are $\mathcal{B}[\bar{B}\to D_{2}e\barν_e] = 1.1^{-0.3}_{+0.3} \times 10^{-3}$, $\mathcal{B}[\bar{B}\to D'_2e\barν_e]=4.1^{-0.8}_{+0.9} \times 10^{-4}$, and $\mathcal{B}[\bar{B}\to D^*_3e\barν_e]=1.0^{-0.2}_{+0.2} \times 10^{-3}$, respectively. For semi-electronic decays of $\bar B_s$ to $D_{s2}$, $D'_{s2}$, and $D^*_{s3}$, the corresponding branching ratios are $1.7^{-0.5}_{+0.5}\times 10^{-3}$, $5.2^{-1.5}_{+1.6}\times 10^{-4}$, and $1.5^{-0.4}_{+0.4}\times 10^{-3}$, respectively. The branching ratios of semi-electronic decays of $B_c$ to $D$-wave $D$ mesons are in the order of $10^{-5}$. We also achieved the forward-backward asymmetry, angular spectra, and lepton momentum spectra. In particular the distribution of decay widths for $2^-$ states $D_2$ and $D'_2$ varying along with mixing angle are presented.

hep-ph↗

arXiv2016

Weak Decays of $J/ψ$ and $Υ(1S)$

In this paper we study the weak decays of $J/ψ$ and $Υ(1S)$. Using the Bethe-Salpeter method, we calculate the hadronic transition amplitude and give the form factors. We find that two new form factors $h_1$ and $h_2$, which do not appear in existing literature, have contributions in $1^-\to 1^-$ decays. They affect the branching ratios of semi-leptonic and non-leptonic decays by the rate of $3\%\sim6\%$ and $2\%\sim14\%$, respectively, so their contributions can not be ignored and should be considered. Our results show that, for the semi-leptonic decay modes, the largest branching ratios are of the order of $10^{-10}$ both for $J/ψ$ and $Υ(1S)$ decays, and the largest branching ratios of non-leptonic decays are of the order of $10^{-9}$ for $J/ψ$ and $10^{-10}$ for $Υ(1S)$.

hep-ph↗

arXiv2016

$D$-Wave Charmonia $η_{c2}$($1{^1D_2}$), $ψ_2$($1{^3D_2}$), and $ψ_3$($1{^3D_3}$) in $B_c$ Decays

We study the semi-leptonic and non-leptonic decays of $B_c$ meson to $D$-wave charmonia, namely, $η_{c2}(1^1\!D_2)$, $ψ_2(1^3\!D_2)$, and $ψ_3(1^3\!D_3)$. In our calculations, the instantaneous Bethe-Salpeter method is applied to achieve the hadronic matrix elements. This method includes relativistic corrections which are important especially for the higher orbital excited states. For the semi-leptonic decay channels with electron as the final lepton, we get the branching ratios $\mathcal{B}[B_c \rightarrow η_{c2}e\barν_e] = 5.9^{-0.8}_{+1.0}\times 10^{-4}$, $\mathcal{B}[B_c \rightarrow ψ_2e\barν_e]=1.5^{-0.2}_{+0.3}\times 10^{-4}$, and $\mathcal{B}[B_c \rightarrow ψ_3e\barν_e]=3.5^{-0.6}_{+0.8}\times 10^{-4}$. The transition form factors, forward-backward asymmetries, and lepton spectra in these processes are also presented. For the non-leptonic decay channels, those with $ρ$ as the lighter meson have the largest branching ratios, $\mathcal{B}[B_c \rightarrow η_{c2}ρ] = 8.1^{-1.0}_{+1.0}\times 10^{-4}$, $\mathcal{B}[B_c \rightarrow ψ_2ρ]=9.6^{-1.0}_{+1.0}\times 10^{-5}$, and $\mathcal{B}[B_c \rightarrow ψ_3ρ]=4.1^{-0.7}_{+0.8}\times 10^{-4}$.

hep-ph↗

arXiv2016

Correlation between amygdala BOLD activity and frontal EEG asymmetry during real-time fMRI neurofeedback training in patients with depression

Real-time fMRI neurofeedback (rtfMRI-nf) is an emerging approach for studies and novel treatments of major depressive disorder (MDD). EEG performed simultaneously with an rtfMRI-nf procedure allows an independent evaluation of rtfMRI-nf brain modulation effects. Frontal EEG asymmetry in the alpha band is a widely used measure of emotion and motivation that shows profound changes in depression. However, it has never been directly related to simultaneously acquired fMRI data. We report the first study investigating electrophysiological correlates of the rtfMRI-nf procedure, by combining rtfMRI-nf with simultaneous and passive EEG recordings. In this pilot study, MDD patients in the experimental group (n=13) learned to upregulate BOLD activity of the left amygdala using an rtfMRI-nf during a happy emotion induction task. MDD patients in the control group (n=11) were provided with a sham rtfMRI-nf. Correlations between frontal EEG asymmetry in the upper alpha band and BOLD activity across the brain were examined. Average individual changes in frontal EEG asymmetry during the rtfMRI-nf task for the experimental group showed a significant positive correlation with the MDD patients' depression severity ratings, consistent with an inverse correlation between the depression severity and frontal EEG asymmetry at rest. Temporal correlations between frontal EEG asymmetry and BOLD activity were significantly enhanced, during the rtfMRI-nf task, for the amygdala and many regions associated with emotion regulation. Our findings demonstrate an important link between amygdala BOLD activity and frontal EEG asymmetry. Our EEG asymmetry results suggest that the rtfMRI-nf training targeting the amygdala is beneficial to MDD patients, and that alpha-asymmetry EEG-nf would be compatible with the amygdala rtfMRI-nf. Combination of the two could enhance emotion regulation training and benefit MDD patients.

physics.med-ph↗

arXiv2016

Rare radiative decays of the $B_c$ meson

In this paper, we study the rare radiative processes $B_c\to D_{sJ} ^{(*)}γ$ within the Standard Model, where $D_{sJ}^{(*)}$ stands for the meson $D_s^*$,~$D_{s1}(2460,2536)$ or ~$D_{s2}^*(2573)$. During the investigations, we consider the contributions from the penguin, annihilation, color-suppressed and color-favored cascade diagrams. Our results show that: 1) the penguin and annihilation contributions are dominant in the branching fractions; 2) for the processes $B_c\to D_s^{*}γ$ and $B_c\to D_{s1}(2460,2536)γ$, the effects from the color-suppressed and color-favored cascade diagrams are un-negligible.

hep-ph↗

arXiv2015

The nonleptonic charmless decays of $B_c$ meson

In this paper, with the framework of (p)NRQCD and SCET, the processes $B_c\to M_1 M_2$ are investigated. Here $M_{1(2)}$ denotes the light charmless meson, such as $π$, $ρ$, $K$ or $K^*$. Based on the SCET power counting rules, the leading transition amplitudes are picked out, which include $A_{wA}^2$, $A_{wB}^2$, $A_{wC}^2$, $A_{wD}^2$ and $A_{c}^0$. From SCET, their factorization formulae are proven. Based on the obtained factorization formulae, in particular, the numerical calculation on $A_{wB}^2$ is performed.

hep-ph↗

arXiv2015

The Electromagnetic Decays of $B^{\pm}_c(2S)$

We calculate the electromagnetic (EM) decay widths of the $B^{\pm}_c(2S)$ meson, which is observed recently by the ATLAS Collaboration. The main EM decay channels of this particle are $1{^3S_1}γ$ and $1{P}γ$, which, in literature, are estimated to have the branching ratio of about $1/10$. In this work, we get the partial decay widths: $Γ(2{^1S_0}\rightarrow 1{^3S_1}γ)=0.192$ keV, $Γ(2{^1S_0}\rightarrow 1{P_1}γ) = 2.24$ keV and $Γ(2{^1S_0}\rightarrow 1{P_1^\prime}γ) = 11.4$ keV. In the calculation, the instantaneous approximated Bethe-Salpeter method is used. For the $P$-wave $B_c$ mesons, the wave functions are given by mixing the $^3P_1$ and $^1P_1$ states. Within the Mandelstam formalism, the decay amplitude is given, which includes the relativistic corrections.

hep-ph↗

arXiv2013

Self-regulation of human brain activity using simultaneous real-time fMRI and EEG neurofeedback

Neurofeedback is a promising approach for non-invasive modulation of human brain activity with applications for treatment of mental disorders and enhancement of brain performance. Neurofeedback techniques are commonly based on either electroencephalography (EEG) or real-time functional magnetic resonance imaging (rtfMRI). Advances in simultaneous EEG-fMRI have made it possible to combine the two approaches. Here we report the first implementation of simultaneous multimodal rtfMRI and EEG neurofeedback (rtfMRI-EEG-nf). It is based on a novel system for real-time integration of simultaneous rtfMRI and EEG data streams. We applied the rtfMRI-EEG-nf to training of emotional self-regulation in healthy subjects performing a positive emotion induction task based on retrieval of happy autobiographical memories. The participants were able to simultaneously regulate their BOLD fMRI activation in the left amygdala and frontal EEG power asymmetry in the high-beta band using the rtfMRI-EEG-nf. Our proof-of-concept results demonstrate the feasibility of simultaneous self-regulation of both hemodynamic (rtfMRI) and electrophysiological (EEG) activity of the human brain. They suggest potential applications of rtfMRI-EEG-nf in the development of novel cognitive neuroscience research paradigms and enhanced cognitive therapeutic approaches for major neuropsychiatric disorders, particularly depression.

physics.med-ph↗

arXiv2013

Lepton-number violating four-body decays of heavy mesons

Neutrinoless hadron Lepton Number Violating (LNV) decays can be induced by virtual Majorana neutrino, which in turn indubitably show the Majorana nature of neutrinos. Many three-body LNV processes and Lepton Flavour Violating (LFV) processes have been studied extensively in theory and by experiment. As a supplement, we here study 75 four-body LNV (LFV) processes from heavy pseudoscalar $B$ and $D$ decays. Most of these processes have not been studied in theory and searched for in experiment, while they may have sizable decay rates. Since the four-body decay modes have the same vertexes and mixing parameters with three-body cases, so their branching fractions are comparable with the corresponding three-body decays. We calculate their decay widths and branching fractions with current bounds on heavy Majorana neutrino mixing parameters, and estimate some channels' reconstruction events using the current experimental data from Belle.

hep-ph↗

arXiv2012

EEG-assisted retrospective motion correction for fMRI: E-REMCOR

We propose a method for retrospective motion correction of fMRI data in simultaneous EEG-fMRI that employs the EEG array as a sensitive motion detector. EEG motion artifacts are used to generate motion regressors describing rotational head movements with millisecond temporal resolution. These regressors are utilized for slice-specific motion correction of unprocessed fMRI data. Performance of the method is demonstrated by correction of fMRI data from five patients with major depressive disorder, who exhibited head movements by 1-3 mm during a resting EEG-fMRI run. The fMRI datasets, corrected using eight to ten EEG-based motion regressors, show significant improvements in temporal SNR (TSNR) of fMRI time series, particularly in the frontal brain regions and near the surface of the brain. The TSNR improvements are as high as 50% for large brain areas in single-subject analysis and as high as 25% when the results are averaged across the subjects. Simultaneous application of the EEG-based motion correction and physiological noise correction by means of RETROICOR leads to average TSNR enhancements as high as 35% for large brain regions. These TSNR improvements are largely preserved after the subsequent fMRI volume registration and regression of fMRI motion parameters. The proposed EEG-assisted method of retrospective fMRI motion correction (referred to as E-REMCOR) can be used to improve quality of fMRI data with severe motion artifacts and to reduce spurious correlations between the EEG and fMRI data caused by head movements. It does not require any specialized equipment beyond the standard EEG-fMRI instrumentation and can be applied retrospectively to any existing EEG-fMRI data set.

physics.med-ph↗