Loading [MathJax]/jax/element/mml/optable/GeneralPunctuation.js
Research article Special Issues

Abdomen tissues segmentation from computed tomography images using deep learning and level set methods


  • Accurate abdomen tissues segmentation is one of the crucial tasks in radiation therapy planning of related diseases. However, abdomen tissues segmentation (liver, kidney) is difficult because the low contrast between abdomen tissues and their surrounding organs. In this paper, an attention-based deep learning method for automated abdomen tissues segmentation is proposed. In our method, image cropping is first applied to the original images. U-net model with attention mechanism is then constructed to obtain the initial abdomen tissues. Finally, level set evolution which consists of three energy terms is used for optimize the initial abdomen segmentation. The proposed model is evaluated across 470 subsets. For liver segmentation, the mean dice are 96.2 and 95.1% for the FLARE21 datasets and the LiTS datasets, respectively. For kidney segmentation, the mean dice are 96.6 and 95.7% for the FLARE21 datasets and the LiTS datasets, respectively. Experimental evaluation exhibits that the proposed method can obtain better segmentation results than other methods.

    Citation: Zhaoxuan Gong, Jing Song, Wei Guo, Ronghui Ju, Dazhe Zhao, Wenjun Tan, Wei Zhou, Guodong Zhang. Abdomen tissues segmentation from computed tomography images using deep learning and level set methods[J]. Mathematical Biosciences and Engineering, 2022, 19(12): 14074-14085. doi: 10.3934/mbe.2022655

    Related Papers:

    [1] Shumoua F. Alrzqi, Fatimah A. Alrawajeh, Hany N. Hassan . An efficient numerical technique for investigating the generalized Rosenau–KDV–RLW equation by using the Fourier spectral method. AIMS Mathematics, 2024, 9(4): 8661-8688. doi: 10.3934/math.2024420
    [2] Xiaoli Wang, Lizhen Wang . Traveling wave solutions of conformable time fractional Burgers type equations. AIMS Mathematics, 2021, 6(7): 7266-7284. doi: 10.3934/math.2021426
    [3] Zhengang Zhao, Yunying Zheng, Xianglin Zeng . Finite element approximation of fractional hyperbolic integro-differential equation. AIMS Mathematics, 2022, 7(8): 15348-15369. doi: 10.3934/math.2022841
    [4] Ailing Zhu, Yixin Wang, Qiang Xu . A weak Galerkin finite element approximation of two-dimensional sub-diffusion equation with time-fractional derivative. AIMS Mathematics, 2020, 5(5): 4297-4310. doi: 10.3934/math.2020274
    [5] Jian-Gen Liu, Jian Zhang . A new approximate method to the time fractional damped Burger equation. AIMS Mathematics, 2023, 8(6): 13317-13324. doi: 10.3934/math.2023674
    [6] Weiwen Wan, Rong An . Convergence analysis of Euler and BDF2 grad-div stabilization methods for the time-dependent penetrative convection model. AIMS Mathematics, 2024, 9(1): 453-480. doi: 10.3934/math.2024025
    [7] Shanhao Yuan, Yanqin Liu, Yibin Xu, Qiuping Li, Chao Guo, Yanfeng Shen . Gradient-enhanced fractional physics-informed neural networks for solving forward and inverse problems of the multiterm time-fractional Burger-type equation. AIMS Mathematics, 2024, 9(10): 27418-27437. doi: 10.3934/math.20241332
    [8] Xin Zhao, Xin Liu, Jian Li . Convergence analysis and error estimate of finite element method of a nonlinear fluid-structure interaction problem. AIMS Mathematics, 2020, 5(5): 5240-5260. doi: 10.3934/math.2020337
    [9] Zhichao Fang, Ruixia Du, Hong Li, Yang Liu . A two-grid mixed finite volume element method for nonlinear time fractional reaction-diffusion equations. AIMS Mathematics, 2022, 7(2): 1941-1970. doi: 10.3934/math.2022112
    [10] Muhammad Asim Khan, Norma Alias, Umair Ali . A new fourth-order grouping iterative method for the time fractional sub-diffusion equation having a weak singularity at initial time. AIMS Mathematics, 2023, 8(6): 13725-13746. doi: 10.3934/math.2023697
  • Accurate abdomen tissues segmentation is one of the crucial tasks in radiation therapy planning of related diseases. However, abdomen tissues segmentation (liver, kidney) is difficult because the low contrast between abdomen tissues and their surrounding organs. In this paper, an attention-based deep learning method for automated abdomen tissues segmentation is proposed. In our method, image cropping is first applied to the original images. U-net model with attention mechanism is then constructed to obtain the initial abdomen tissues. Finally, level set evolution which consists of three energy terms is used for optimize the initial abdomen segmentation. The proposed model is evaluated across 470 subsets. For liver segmentation, the mean dice are 96.2 and 95.1% for the FLARE21 datasets and the LiTS datasets, respectively. For kidney segmentation, the mean dice are 96.6 and 95.7% for the FLARE21 datasets and the LiTS datasets, respectively. Experimental evaluation exhibits that the proposed method can obtain better segmentation results than other methods.



    In this article, we consider the following time-fractional generalized Rosenau-RLW-Burgers equation:

    utC0Dαtuxx+C0Dβtuxxxx+uxuxx+f(u)x=g(x,t), (x,t)Ω×J, (1.1)

    with boundary conditions

    u(x,t)=uxx(x,t)=0, (x,t)Ω×ˉJ, (1.2)

    and initial condition

    u(x,0)=u0(x), xΩ, (1.3)

    where Ω=(a,b) is the spatial domain, J=(0,T] is the time interval with T(0,), and g(x,t) is a known source term function. The nonlinear term f(u) satisfies the assumption condition |f(u)|cf(u)|u|, where cf(u) is a positive constant on u. C0Dαtu and C0Dβtu are both Caputo fractional derivatives with 0<α,β<1. Since C0Dγtu=γ(uu0)tγ, all of the above Caputo fractional derivatives can be converted into the Riemann-Liouville fractional derivative, note that

    γutγ=1Γ(1γ)tt0u(x,s)(ts)γds,0<γ<1. (1.4)

    Specifically, when α=1, β=1, (1.1) degenerates into the generalized Rosenau-RLW-Burgers equation which can be seen as the combined system between the generalized Rosenau-RLW equation and the generalized Rosenau-Burgers equation.

    The RLW equation, the Rosenau equation, and their combined systems with other equations are significant mathematical and physical equations that effectively describe nonlinear wave behaviors. These equations have become interesting topics in the study of nonlinear dispersion dynamics. Since obtaining analytical solutions for these equations is challenging, studying their numerical methods is paramount. Over the years, there has been extensive research on numerical methods for solving this type of equation. In [1], Atouani and Omrani discussed the numerical solution of the Rosenau-RLW (RRLW) equation based on the Galerkin finite element method. In [2], He and Pan developed a three-level, linearly implicit finite difference method for solving the generalized Rosenau-Kawahara-RLW equation. In [3], Wongsaijai and Poochinapan developed a pseudo-compact finite difference scheme for solving the generalized Rosenau-RLW-Burgers equation. In [4], Mouktonglang et al. analyzed a generalized Rosenau-RLW-Burgers equation with periodic initial-boundary value. For more papers on related equations, please refer to [5,6,7,8]. It is worth noting that the literature on the fractional generalized Rosenau-RLW-Burgers equation is relatively scarce, and its analytical solution is difficult to obtain. Therefore, we have to consider effective numerical methods such as finite element methods [9,10,11,12], finite difference methods [13,14,15], finite volume methods [16], spectral methods [17,18,19], and mixed finite element methods [20,21,22]. In addition, the existence of time-fractional derivatives increases the difficulty of studying numerical methods. Therefore, it is crucial to choose an appropriate high-order approximation formula for the fractional derivative to establish a stable numerical scheme for (1.1).

    In 1986, Lubich [23] proposed the convolution quadrature (CQ) formula for Riemann Liouville fractional operators using the discrete convolution. In [24], Chen et al. developed an alternating direction implicit fractional trapezoidal rule type to solve a two-dimensional fractional evolution equation. In [25], Jin et al. proposed a corrected approximation formula for high-order BDFs through appropriate initial modifications to discretize fractional evolution equations. Based on the CQ formula, in [26], Liu et al. developed the shifted convolution quadrature (SCQ) theory, which extended the CQ formula at xnθ and discussed the constraints of parameter θ. In [27], Yin et al. studied the generalized BDF2-θ with the finite element method for solving the fractional mobile/immobile transport model, and also developed a correction scheme by adding the starting part to restore convergence order. For more related papers, please refer to [28,29,30,31,32,33].

    In this article, we develop the generalized BDF2-θ in time combined with the mixed finite element method in space to solve (1.1). The focuses of this article are as follows:

    ● It is noted that the time-fractional generalized Rosenau-RLW-Burgers equation containing two time-fractional operators is studied.

    ● The stability of the time-fractional generalized Rosenau-RLW-Burgers equation (1.1) based on the mixed finite element method is given.

    ● Based on a comprehensive analysis of some numerical examples, the numerical method's feasibility and effectiveness have been extensively validated. Specifically, the issue of decreasing the convergence rate of nonsmooth solutions is solved by adding correction terms.

    The structure of this article is as follows: In Section 2, the generalized BDF2-θ is introduced, and the fully discrete mixed finite element scheme is provided. In Section 3, the existence and uniqueness theorem for the fully discrete mixed finite element scheme is given. In Section 4, the stability of the scheme is proved. In Section 5, some numerical examples with smooth and nonsmooth solutions based on the discrete scheme are presented. In Section 6, some conclusions are given.

    In this section, we present the fully discrete mixed finite element scheme for (1.1) in space, which combines the generalized BDF2-θ in time. The generalized BDF2-θ with the starting part is introduced in [27]. Further, we divide the time interval [0,T] into 0=t0<t1<<tN1<tN=T, and let tn=nτ(n=1,2,,N), where τ is time step length size and N is a positive integer.

    For the convenience of research, set ˆu:=uu0, and assume that ˆu has the following form:

    ˆu(x,t)=ˆu1(x,t)+ˆu2(x,t):=κj=1cjtσj+tσκ+1ϕ(x,t), (2.1)

    where cj=c(x), 1<σ1<σ2<<σκ<σκ+1 and ϕ(x,t) is sufficiently differentiable with respect to t.

    Using ˆu:=uu0, we can write (1.1)–(1.3) as

    ˆutαˆuxxtα+βˆuxxxxtβ+ˆuxˆuxx+f(ˆu)x=ˆg(x,t),(x,t)Ω×J, (2.2)

    with boundary conditions

    ˆu(x,t)=ˆuxx(x,t)=0,(x,t)Ω×ˉJ, (2.3)

    and initial condition

    ˆu(x,0)=0,xΩ, (2.4)

    where ˆg(x,t)=g(x,t)+(u0)x(u0)xx.

    Now, we introduce an auxiliary variable q=ˆuxx to obtain the following coupled system:

    ˆutαˆuxxtα+βqxxtβ+ˆuxˆuxx+f(ˆu)x=ˆg(x,t), (2.5)

    and

    q=ˆuxx. (2.6)

    Multiplying (2.5) and (2.6) by vH10 and wH10, respectively, integrating the result equations, and using integration by parts, we obtain the following weak form:

    (ˆut,v)+(αˆuxtα,vx)(βqxtβ,vx)(ˆu,vx)+(ˆux,vx)(f(ˆu),vx)=(ˆg,v),vH10, (2.7)

    and

    (q,w)+(ˆux,wx)=0,wH10. (2.8)

    To provide the fully discrete numerical scheme, we first introduce the relevant formulas and lemmas for the generalized BDF2-θ.

    For smooth functions ˆu and q in [0,T], we let ˆun=ˆu(,tn), qn=(,tn). The approximation formula for the Riemann-Liouville fractional derivative at time tnθ with the generalized BDF2-θ is

    γˆunθtγ=τγnj=0ω(γ)jˆunj+τγκj=1ω(γ)n,jˆuj+Rnθγ:=Ψγ,nτˆu+Sγ,nτ,κˆu+Rnθγ, (2.9)

    where |Rnθγ|Cτ2.

    The discrete convolution part is denoted as

    Ψγ,nτˆu:=τγnj=0ω(γ)jˆunj, (2.10)

    and the starting part is

    Sγ,nτ,κˆu:=τγκj=1ω(γ)n,jˆuj. (2.11)

    The convolution weights {ω(γ)j}j=0 in (2.10) are generated by the following generating function:

    ω(γ)(ξ)=(3γ2θ2γ2γ2θγξ+γ2θ2γξ2)γ. (2.12)

    Lemma 2.1. [27] We give the convolution weights {ω(γ)j}j=0 of the generalized BDF2-θ as follows:

    ω(γ)0=(3γ2θ2γ)γ,ω(γ)1=2(θγ)(2γ3γ2θ)1γ,ω(γ)j=2γj(3γ2θ)[2(γθ)(j1γ1)ω(γ)j1+(γ2θ)(1j22γ)ω(γ)j2],j2. (2.13)

    Lemma 2.2. [27] The starting weights {ω(γ)n,j}κj=1 of the generalized BDF2-θ are given as the following:

    κj=1ω(γ)n,jj=Γ(+1)Γ(γ+1)(nθ)γnj=1ω(γ)njj,=σ1,σ2,,σκ. (2.14)

    Lemma 2.3. [12,15] For ˆuC4[0,π], the following two approximate formulas at tnθ hold:

    g(tnθ)=gnθ+O(τ2),f(ˆu(tnθ))=f(ˆunθ)+O(τ2), (2.15)

    where gnθ:=(1θ)gn+θgn1 and f(ˆunθ):=(2θ)f(ˆun1)(1θ)f(ˆun2).

    Next, we have the following approximate formula:

    ˆu(tnθ)=ˆunθ+S0,nτ,κˆu+O(τ2):=(1θ)ˆun+θˆun1+S0,nτ,κˆu+O(τ2). (2.16)

    Without considering the starting part, we can obtain the weak form of (2.5) and (2.6) at tnθ:

    (Ψ1,nτˆu,v)+(Ψα,nτˆux,vx)(Ψβ,nτqx,vx)(ˆunθ,vx)+(ˆunθx,vx)=(f(ˆunθ),vx)+(ˆgnθ,v)(Rnθ1,v), (2.17)

    and

    (qnθ,w)+(ˆunθx,wx)=(Rnθ2,w), (2.18)

    where Rnθ1=O(τ2) and Rnθ2=O(τ2).

    To establish the fully discrete mixed finite element scheme, we introduce the following finite element space:

    Vh={vh|vhH10,vh|IiPk(Ii),IiTh,k1},

    where Th is a subdivision of ˉΩ=[a,b] into M subintervals Ii=[xi1,xi], with hi=xixi1, h=max1iMhi, and Pk(Ii) represent the polynomials with a degree less than or equal to k in Ii.

    Next, we provide linear basis functions {φi}Mi=1 of finite element space Vh as follows:

    φi(x)={1+xxihi,xIi,1xxihi+1,xIi+1,0,others, (2.19)
    φM(x)={1+xxMhM,xIM,0,others. (2.20)

    Based on the above finite element space, we find {Unθ,Qnθ}Vh×Vh satisfying

    (Ψ1,nτU,V)+(Ψα,nτUx,Vx)(Ψβ,nτQx,Vx)(Unθ,Vx)+(Unθx,Vx)=(f(Unθ),Vx)+(ˆgnθ,V),VVh, (2.21)

    and

    (Qnθ,W)+(Unθx,Wx)=0,WVh. (2.22)

    Theorem 3.1. The solution of the fully discrete mixed finite element scheme (2.21) and (2.22) is uniquely solvable.

    Proof. Taking basis functions {φi}Mi=1 of finite element space Vh, we have

    Un=Mi=1uniφi,Qn=Mi=1qniφi. (3.1)

    Taking V=φj and W=φj from (2.21) and (2.22), we have

    τ1ω(1)0AUn+ταω(α)0BUn+(1θ)BUn(1θ)CUnτβω(β)0BQn=Fnθ+Gnθτ1nk=1ω(1)kAUnkταnk=1ω(α)kBUnkθBUn1+θCUn1+τβnk=1ω(β)kBQnk, (3.2)

    and

    (1θ)BUn+(1θ)AQn=θBUnθAQn, (3.3)

    where

    A=[(φi,φj)]T1i,jM,B=[(φix,φjx)]T1i,jM,C=[(φi,φjx)]T1i,jM,Fnθ=[(f(Unθ),φ1x),,(f(Unθ),φMx)]T,Gnθ=[(gnθ,φ1),,(gnθ,φM)]T.

    Obviously, A and B are symmetric and positive definite. Further, processing the boundary and simplifying the right-hand term, we have

    (τ1ω(1)0˜A+ταω(α)0˜B+(1θ)˜B(1θ)˜C)Unτβω(β)0˜BQn=Hn1, (3.4)

    and

    (1θ)˜BUn+(1θ)˜AQn=Hn2, (3.5)

    where

    Hn1=Fnθ+Gnθτ1nk=1ω(1)kAUnkταnk=1ω(α)kBUnkθBUn1+θCUn1+τβnk=1ω(β)kBQnk,Hn2=θBUnθAQn.

    Multiplying (3.4) by τ˜A1, we have

    (ω(1)0E+τ1αω(α)0˜A1˜B+τ(1θ)˜A1˜Bτ(1θ)˜A1˜C)Unτ1βω(β)0˜A1˜BQn=τ˜A1Hn1. (3.6)

    Further, rewrite (3.5) as

    Qn=Hn3, (3.7)

    where Hn3=(1θ)1˜A1Hn2˜A1˜BUn.

    Substitute (3.7) into (3.6) to obtain

    KUn=Hn4, (3.8)

    where

    K=ω(1)0E+τ1αω(α)0˜A1˜B+τ(1θ)˜A1˜Bτ(1θ)˜A1˜C+τ1βω(β)0˜A1˜B˜A1˜B,
    Hn4=τ˜A1Hn1+τ1β(1θ)1ω(β)0˜A1˜B˜A1Hn2.

    It is easy to see that (3.7) and (3.8) are equivalent to (3.4) and (3.5). Due to τ being small enough and E being an identity matrix, the matrix K is invertible. Additionally, since Uk(k=0,1,,n1) is known, after multiple iterations, (3.7) and (3.8) have a unique solution.

    Remark 3.1. Since we introduce the auxiliary variable q=ˆuxx to transform (2.2) into a first-order system (2.5) and (2.6), according to [34,35], the mixed finite element scheme (2.21) and (2.22) do not need to satisfy the LBB condition. In [36], the LBB condition is a condition for the problem to be well posed. From this perspective, typically satisfying the LBB condition is to obtain the existence and uniqueness of a solution. Although the mixed finite element scheme in this article does not need to satisfy the LBB condition, it still satisfies the existence and uniqueness of a solution.

    Lemma 4.1. [12,14] For UmVh, satisfying Um=0(m<0), we have

    (Ψ1,mtU,Umθ)14τ(H[Um]H[Um1]), m1,

    where

    H[Um]=(32θ)

    and

    \mathbb{H}[U^m]\ge\frac{1}{1-\theta}\|U^m\|^2,\ m\ge1.

    Lemma 4.2. [27] } {For any vector (v^0, v^1, \cdots, v^{n-1})\in \mathbb{R}^n , defining \{\omega_k^{(\gamma)}\}_{k = 0}^{\infty}\; (0 < \gamma < 1) be a sequence of coefficients of the generating function \omega^{(\gamma)}(\xi) in (2.12) and 0\le\theta\le\min\{\gamma, \frac{1}{2}\} , we have

    \sum\limits_{m = 1}^{n-1}v^m\sum\limits_{k = 1}^{m}\omega_{m-k}^{(\gamma)} v^k\ge0,\ n\ge1.

    Theorem 4.1. Let u^n_h = U^n+\bar{u}_h^0 , where \bar{u}_h^0 is an approximation of u_0 , the following stability of the fully discrete scheme (2.21) and (2.22) holds:

    \begin{equation} \|u_h^L\|^2\ \le\ C\left(\|\bar{u}_h^0 \|^2+\tau\sum\limits_{n = 1}^{L}\|g^{n-\theta} \|^2\right),\ 1\le L \le N, \end{equation} (4.1)

    where C is a positive constant independent of h and \tau .

    Proof. Taking V = U^{n-\theta} , W = \Psi_{\tau}^{\beta, n}Q , (2.21) and (2.22) can be written as

    \begin{equation} \begin{split} &(\Psi_{\tau}^{1,n} U,U^{n-\theta})+(\Psi_{\tau}^{\alpha,n} U_{x},U^{n-\theta}_x)-(\Psi_{\tau}^{\beta,n}Q_{x},U^{n-\theta}_x)+\|U_{x}^{n-\theta }\|^2\\ = &(U^{n-\theta},U^{n-\theta}_x)+(f(U^{n-\theta}),U^{n-\theta}_x)+(\hat{g}^{n-\theta},U^{n-\theta}), \end{split} \end{equation} (4.2)

    and

    \begin{equation} (Q^{n-\theta},\Psi_{\tau}^{\beta,n}Q)+(U^{n-\theta}_{x},\Psi_{\tau}^{\beta,n}Q_x) = 0. \end{equation} (4.3)

    Adding (4.2) and (4.3), we have

    \begin{equation} \begin{split} &(\Psi_{\tau}^{1,n} U,U^{n-\theta})+(\Psi_{\tau}^{\alpha,n} U_{x},U^{n-\theta}_x)+(Q^{n-\theta},\Psi_{\tau}^{\beta,n}Q)+\|U_{x}^{n-\theta }\|^2\\ = &(U^{n-\theta},U^{n-\theta}_x)+(f(U^{n-\theta}),U^{n-\theta}_x)+(\hat{g}^{n-\theta},U^{n-\theta}). \end{split} \end{equation} (4.4)

    Using Lemma 4.1, we obtain

    \begin{equation} \begin{split} &\frac{1}{4\tau}(\mathbb{H}[U^n]-\mathbb{H}[U^{n-1}])+(\Psi_{\tau}^{\alpha,n} U_{x},U^{n-\theta}_x)+(Q^{n-\theta},\Psi_{\tau}^{\beta,n}Q)+\|U_{x}^{n-\theta }\|^2\\ \le&(U^{n-\theta},U^{n-\theta}_x)+(f(U^{n-\theta}),U^{n-\theta}_x)+(\hat{g}^{n-\theta},U^{n-\theta}). \end{split} \end{equation} (4.5)

    Multiply (4.5) by 4\tau and sum it with respect to n from 1 to L to get

    \begin{equation} \begin{split} &\mathbb{H}[U^L]-\mathbb{H}[U^{0}]+ 4\tau\sum\limits_{n = 1}^{L}(\Psi_t^{\alpha,n} U_{x},U^{n-\theta}_x)+ 4\tau\sum\limits_{n = 1}^{L}(Q^{n-\theta},\Psi_{\tau}^{\beta,n}Q)+ 4\tau\sum\limits_{n = 1}^{L}\|U_{x}^{n-\theta }\|^2\\ \le&4\tau\left(\sum\limits_{n = 1}^{L}(U^{n-\theta},U^{n-\theta}_x)+\sum\limits_{n = 1}^{L}(f(U^{n-\theta}),U^{n-\theta}_x)+\sum\limits_{n = 1}^{L}(\hat{g}^{n-\theta},U^{n-\theta})\right). \end{split} \end{equation} (4.6)

    By the Hölder inequality and Young inequality, the three terms on the right-hand side of (4.6) can be expanded to

    \begin{equation} \sum\limits_{n = 1}^{L}(U^{n-\theta},U^{n-\theta}_x)\le\frac{1}{2}\sum\limits_{n = 1}^{L}\|U^{n-\theta}\|^2+\frac{1}{2}\sum\limits_{n = 1}^{L}\|U_{x}^{n-\theta }\|^2, \end{equation} (4.7)
    \begin{equation} \begin{split} \sum\limits_{n = 1}^{L}(f(U^{n-\theta}),U^{n-\theta}_x) \le&\sum\limits_{n = 1}^{L}\|c_f(U^{n-\theta})\|_{\infty}\|U^{n-\theta}\|\|U^{n-\theta}_x\| \\ \le& C\sum\limits_{n = 1}^{L}\|U^{n-\theta}\|^2+\frac{1}{2}\sum\limits_{n = 1}^{L}\|U_{x}^{n-\theta }\|^2, \end{split} \end{equation} (4.8)
    \begin{equation} \sum\limits_{n = 1}^{L}(\hat{g}^{n-\theta},U^{n-\theta})\le\frac{1}{2}\sum\limits_{n = 1}^{L}\|\hat{g}^{n-\theta}\|^2 +\frac{1}{2}\sum\limits_{n = 1}^{L}\|U^{n-\theta}\|^2, \end{equation} (4.9)

    where we use the bounded condition \|c_f(U^{n-\theta})\|_{\infty}\leq C .

    Substituting (4.7)–(4.9) into (4.6), we arrive at

    \begin{equation} \begin{split} & \mathbb{H}[U^L]-\mathbb{H}[U^{0}]+ 4\tau\sum\limits_{n = 1}^{L}(\Psi_t^{\alpha,n} U_{x},U^{n-\theta}_x)+ 4\tau\sum\limits_{n = 1}^{L}(Q^{n-\theta},\Psi_{\tau}^{\beta,n}Q)\\ \le& C\tau \left(\sum\limits_{n = 1}^{L}\|\hat{g}^{n-\theta}\|^2+\sum\limits_{n = 1}^{L}\|U^{n-\theta}\|^2\right). \end{split} \end{equation} (4.10)

    In what follows, using Lemmas 4.1 and 4.2 and the Gronwall inequality, we have

    \begin{equation} \|U^L\|^2-\|U^0\|^2\le C\tau\sum\limits_{n = 1}^{L}\|\hat{g}^{n-\theta}\|^2. \end{equation} (4.11)

    Since U^0 = 0 , we obtain

    \begin{equation} \|U^L\|^2\le C\tau\sum\limits_{n = 1}^{L}\|\hat{g}^{n-\theta}\|^2. \end{equation} (4.12)

    Noting that U^L = u^L_h-\bar{u}_h^0 and using the triangle inequality, the conclusion of this theorem is derived.

    In this section, we present numerical simulation results for both smooth and nonsmooth solutions to verify the effectiveness of the numerical scheme. Next, we set the nonlinear term f(u) = u^2 , the spatial domain \Omega = (0, 1) , and the time interval J = (0, 1] .

    Example 5.1 The exact solution is u(x, t) = t^{2}\sin(2\pi x) satisfying u(x, 0) = 0 , and the known source function g(x, t) is given by

    \begin{equation} g(x,t) = \sin(2\pi x)\left(2t+\frac{8\pi ^2t^{2-\alpha}}{\Gamma(3-\alpha)}+\frac{32\pi^4t^{2-\beta}}{\Gamma(3-\beta)}+4\pi^2t^2\right)+2\pi t^2\cos(2\pi x)+2\pi t^4\sin(4\pi x ). \end{equation} (5.1)

    In Table 1, fixing \tau = 1/1000 and choosing h = 1/10, 1/20, 1/40, 1/80 , we provide the L^2 -errors and the spatial convergence rates for u and q with different parameters \alpha , \beta , and \theta , where \theta\le\min\{\alpha, \beta, \frac{1}{2}\} . Similarly, in Table 2, taking h = 1/1000 , we calculate the L^2 -errors and the time convergence rates with \tau = 1/10, 1/20, 1/40, 1/80 . From Tables 1 and 2, one can see that the convergence rates in both space and time are close to 2 when the exact solution is smooth. In Table 3, if \theta > \min\{\alpha, \beta, \frac{1}{2}\} , the convergence accuracy will be unstable, which verifies the range of \theta values from a numerical perspective. To observe the effect of numerical simulation more clearly, we provide the comparison images between numerical solutions and exact solutions. In Figure 1, we show distinct comparison images of the numerical solutions of u_h and q_h and the exact solutions of u and q with \tau = 1/1000 , h = 1/80 , \alpha = 0.2 , \beta = 0.8 , and \theta = 0.2 .

    Table 1.  Spatial convergence results with \tau = 1/1000 .
    \alpha \beta \theta h \|u_h-u\| Rate \|q_h-q\| Rate
    1/10 2.2165E-02 - 2.6449E-02 -
    0.2 1/20 5.6375E-03 1.9752 6.1401E-03 2.1069
    1/40 1.4151E-03 1.9941 1.5121E-03 2.0217
    1/80 3.5399E-04 1.9992 3.8250E-04 1.9830
    1/10 2.2165E-02 - 2.6470E-02 -
    0.2 0.8 -0.5 1/20 5.6369E-03 1.9753 6.1607E-03 2.1032
    1/40 1.4146E-03 1.9945 1.5327E-03 2.0070
    1/80 3.5346E-04 2.0008 4.0308E-04 1.9269
    1/10 2.2164E-02 - 2.6490E-02 -
    -1 1/20 5.6364E-03 1.9754 6.1810E-03 2.0996
    1/40 1.4141E-03 1.9949 1.5530E-03 1.9928
    1/80 3.5293E-04 2.0024 4.2346E-04 1.8747
    1/10 2.1828E-02 - 4.0463E-02 -
    0.5 1/20 5.5499E-03 1.9756 9.6844E-03 2.0629
    1/40 1.3932E-03 1.9941 2.3964E-03 2.0148
    1/80 3.4861E-04 1.9987 5.9910E-04 2.0000
    1/10 2.1828E-02 - 4.0465E-02 -
    0.5 0.5 0.2 1/20 5.5499E-03 1.9757 9.6868E-03 2.0626
    1/40 1.3931E-03 1.9941 2.3988E-03 2.0137
    1/80 3.4854E-04 1.9989 6.0151E-04 1.9956
    1/10 2.1826E-02 - 4.0523E-02 -
    -1 1/20 5.5484E-03 1.9759 9.7445E-03 2.0561
    1/40 1.3916E-03 1.9953 2.4564E-03 1.9880
    1/80 3.4706E-04 2.0035 6.5921E-04 1.8977
    1/10 2.1399E-02 - 5.8275E-02 -
    0.2 1/20 5.4386E-03 1.9763 1.4195E-02 2.0375
    1/40 1.3651E-03 1.9943 3.5279E-03 2.0085
    1/80 3.4156E-04 1.9988 8.8248E-04 1.9992
    1/10 2.1399E-02 - 5.8276E-02 -
    0.8 0.2 0 1/20 5.4386E-03 1.9763 1.4196E-02 2.0374
    1/40 1.3650E-03 1.9943 3.5290E-03 2.0082
    1/80 3.4153E-04 1.9989 8.8360E-04 1.9978
    1/10 2.1396E-02 - 5.8410E-02 -
    -1 1/20 5.4352E-03 1.9769 1.4329E-02 2.0272
    1/40 1.3616E-03 1.9970 3.6619E-03 1.9683
    1/80 3.3812E-04 2.0097 1.0167E-03 1.8487

     | Show Table
    DownLoad: CSV
    Table 2.  Time convergence results with h = 1/1000 .
    \alpha \beta \theta \tau \|u_h-u\| Rate \|q_h-q\| Rate
    1/10 1.9614E-03 - 7.7510E-02 -
    0.2 1/20 5.0017E-04 1.9714 1.9814E-02 1.9679
    1/40 1.2703E-04 1.9773 5.0533E-03 1.9712
    1/80 3.2128E-05 1.9832 1.2867E-03 1.9736
    1/10 7.2565E-03 - 2.8650E-01 -
    0.2 0.8 -0.5 1/20 1.8309E-03 1.9867 7.2335E-02 1.9858
    1/40 4.6032E-04 1.9919 1.8206E-02 1.9903
    1/80 1.1551E-04 1.9946 4.5778E-03 1.9917
    1/10 1.2197E-02 - 4.8151E-01 -
    -1 1/20 3.1215E-03 1.9662 1.2329E-01 1.9655
    1/40 7.8542E-04 1.9907 3.1091E-02 1.9875
    1/80 1.9579E-04 2.0042 7.8135E-03 1.9924
    1/10 5.3041E-04 - 2.1042E-02 -
    0.5 1/20 1.3386E-04 1.9863 5.3622E-03 1.9724
    1/40 3.3676E-05 1.9910 1.3637E-03 1.9753
    1/80 8.4582E-06 1.9933 3.4761E-04 1.9720
    1/10 1.1457E-03 - 4.5326E-02 -
    0.5 0.5 0.2 1/20 2.8884E-04 1.9879 1.1461E-02 1.9836
    1/40 7.2679E-05 1.9907 2.8911E-03 1.9871
    1/80 1.8260E-05 1.9929 7.2972E-04 1.9862
    1/10 1.5622E-02 - 6.1668E-01 -
    -1 1/20 4.0111E-03 1.9615 1.5838E-01 1.9611
    1/40 1.0095E-03 1.9904 3.9868E-02 1.9901
    1/80 2.5362E-04 1.9929 1.0017E-02 1.9928
    1/10 5.2861E-04 - 2.0885E-02 -
    0.5 1/20 1.3465E-04 1.9730 5.3245E-03 1.9717
    1/40 3.4107E-05 1.9811 1.3530E-03 1.9765
    1/80 8.5857E-06 1.9901 3.4270E-04 1.9812
    1/10 2.2763E-03 - 8.9851E-02 -
    0.8 0.5 0 1/20 5.7402E-04 1.9875 2.2660E-02 1.9874
    1/40 1.4460E-04 1.9890 5.7086E-03 1.9889
    1/80 3.6405E-05 1.9899 1.4372E-03 1.9898
    1/10 1.5534E-02 - 6.1313E-01 -
    -1 1/20 3.9902E-03 1.9609 1.5749E-01 1.9609
    1/40 1.0033E-03 1.9917 3.9605E-02 1.9916
    1/80 2.5177E-04 1.9946 9.9391E-03 1.9945

     | Show Table
    DownLoad: CSV
    Table 3.  Time convergence results with h = 1/1000 .
    \alpha \beta \theta \tau \|u_h-u\| Rate \|q_h-q\| Rate
    1/10 2.4519E-03 - 9.6803E-02 -
    0.1 0.9 0.11 1/20 6.2095E-04 1.9813 2.4519E-02 1.9811
    1/40 5.2080E-04 0.2538 2.0647E-02 0.2480
    1/80 4.0699E+02 -19.5758 1.6068E+04 -19.5699
    1/10 1.8093E-03 - 7.1425E-02 -
    0.5 0.5 0.51 1/20 4.5109E-04 2.0039 1.7808E-02 2.0039
    1/40 2.0018E-04 1.1721 7.8154E-03 1.1881
    1/80 5.1099E-04 -1.3520 2.0090E-02 -1.3621
    1/10 4.8552E-04 - 1.9119E-02 -
    0.8 0.2 0.21 1/20 1.2438E-04 1.9647 4.8398E-03 1.9820
    1/40 5.8298E-05 1.0933 2.2123E-03 1.1294
    1/80 2.3759E-02 -8.6708 9.3790E-01 -8.7277

     | Show Table
    DownLoad: CSV
    Figure 1.  u_h , q_h and u , q with \tau = 1/1000 , h = 1/80 , \alpha = 0.2 , \beta = 0.8 , \theta = 0.2 .

    Example 5.2 In this example, we consider the case where the nonsmooth solution is taken as u = (t^{\alpha+\beta}+t^3) \sin(2\pi x) , and the known source term g(x, t) is

    \begin{equation} \begin{split} g(x,t) = &\sin(2\pi x)\left[(\alpha+\beta)t^{\alpha+\beta-1}+3t^2+4\pi^2\left(\frac{t^{\beta}\Gamma(\alpha+\beta+1)}{\Gamma(\beta+1)}+\frac{6t^{3-\alpha}}{\Gamma(4-\alpha)}\right)\right]\\ &+\sin(2\pi x)\left[16\pi^4\left(\frac{t^{\alpha}\Gamma(\alpha+\beta+1)}{\Gamma(\alpha+1)}+\frac{6t^{3-\beta}}{\Gamma(4-\beta)}\right)+4\pi^2(t^{\alpha+\beta}+t^3)\right]\\ &+2\pi (t^{\alpha+\beta}+t^3)\cos(2\pi x)+2\pi(t^{\alpha+\beta} +t^3)^2\sin(4\pi x ). \end{split} \end{equation} (5.2)

    Table 4 presents the L^2 -errors and the spatial convergence rates of u and q before and after adding the starting parts with h = 1/10, 1/20, 1/40, 1/80 , \tau = 1/2000 , where Erroro denotes the error before adding the starting parts and Errorc denotes the error after adding the starting parts.

    Table 4.  Spatial convergence results with \alpha = 0.9 , \beta = 0.2 , \tau = 1/2000 .
    \|u_h-u\| \|q_h-q\|
    \theta h Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 4.2694E-02 - 4.2693E-02 - 1.2162E-01 - 1.2163E-01 -
    0.2 1/20 1.0850E-02 1.9763 1.0850E-02 1.9763 2.9675E-02 2.0351 2.9679E-02 2.0349
    1/40 2.7235E-03 1.9942 2.7234E-03 1.9942 7.3708E-03 2.0094 7.3746E-03 2.0088
    1/80 6.8165E-04 1.9984 6.8155E-04 1.9985 1.8361E-03 2.0051 1.8400E-03 2.0029
    1/10 4.2693E-02 - 4.2692E-02 - 1.2165E-01 - 1.2168E-01 -
    -0.5 1/20 1.0849E-02 1.9764 1.0849E-02 1.9764 2.9706E-02 2.0340 2.9728E-02 2.0331
    1/40 2.7227E-03 1.9945 2.7221E-03 1.9947 7.4017E-03 2.0048 7.4237E-03 2.0016
    1/80 6.8085E-04 1.9996 6.8028E-04 2.0005 1.8671E-03 1.9871 1.8891E-03 1.9744
    1/10 4.2691E-02 - 4.2690E-02 - 1.2172E-01 - 1.2177E-01 -
    -1 1/20 1.0848E-02 1.9766 1.0846E-02 1.9767 2.9776E-02 2.0314 2.9822E-02 2.0297
    1/40 2.7209E-03 1.9952 2.7197E-03 1.9957 7.4714E-03 1.9947 7.5178E-03 1.9880
    1/80 6.7905E-04 2.0025 6.7786E-04 2.0044 1.9368E-03 1.9477 1.9833E-03 1.9224

     | Show Table
    DownLoad: CSV

    The spatial convergence rate is almost unaffected before and after correction, based on a comparison of the data in Table 5. In Tables 6 and 7, we present the L^2 -errors and the time convergence rates of u and q before and after adding the starting parts. Without the addition of the starting parts, the time convergence rates are unstable and cannot reach the second-order convergence results computed by the generalized BDF2- \theta . After adding the starting parts, the time convergence rates keep around 2 , indicating that the starting part plays a major role in correcting the time convergence rates.

    Table 5.  Spatial convergence results with \alpha = 0.5 , \beta = 0.7 , \tau = 1/2000 .
    \|u_h-u\| \|q_h-q\|
    \theta h Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 4.3949E-02 - 4.3948E-02 - 6.8838E-02 - 6.8891E-02 -
    0.5 1/20 1.1177E-02 1.9753 1.1176E-02 1.9754 1.6266E-02 2.0814 1.6318E-02 2.0778
    1/40 2.8070E-03 1.9934 2.8057E-03 1.9940 3.9696E-03 2.0348 4.0219E-03 2.0205
    1/80 7.0358E-04 1.9963 7.0222E-04 1.9984 9.4708E-04 2.0674 9.9935E-04 2.0088
    1/10 4.3949E-02 - 4.3948E-02 - 6.8842E-02 - 6.8897E-02 -
    0.2 1/20 1.1177E-02 1.9753 1.1176E-02 1.9754 1.6270E-02 2.0811 1.6324E-02 2.0774
    1/40 2.8069E-03 1.9935 2.8055E-03 1.9940 3.9739E-03 2.0336 4.0279E-03 2.0189
    1/80 7.0346E-04 1.9964 7.0206E-04 1.9986 9.5140E-04 2.0624 1.0054E-03 2.0023
    1/10 4.3948E-02 - 4.3946E-02 - 6.8877E-02 - 6.8974E-02 -
    -1 1/20 1.1176E-02 1.9754 1.1174E-02 1.9756 1.6304E-02 2.0788 1.6401E-02 2.0722
    1/40 2.8061E-03 1.9938 2.8035E-03 1.9948 4.0077E-03 2.0244 4.1049E-03 1.9984
    1/80 7.0259E-04 1.9978 7.0007E-04 2.0017 9.8520E-04 2.0243 1.0824E-03 1.9231

     | Show Table
    DownLoad: CSV
    Table 6.  Time convergence results with \alpha = 0.9 , \beta = 0.2 , h = 1/2000 .
    \|u_h-u\| \|q_h-q\|
    \theta \tau Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 2.1916E-03 - 4.2694E-02 - 8.6515E-02 - 1.2162E-01 -
    0.2 1/20 1.6194E-03 0.4365 1.0850E-02 1.9763 6.3929E-02 0.4365 2.9675E-02 2.0351
    1/40 1.0890E-03 0.5725 2.7235E-03 1.9942 4.2990E-02 0.5725 7.3710E-03 2.0093
    1/80 6.8790E-04 0.6627 6.8165E-04 1.9984 2.7157E-02 0.6627 1.8364E-03 2.0050
    1/10 2.5540E-03 - 4.2693E-02 - 1.0123E-01 - 1.2164E-01 -
    0 1/20 1.1111E-03 1.2007 1.0850E-02 1.9763 4.3864E-02 1.2065 2.9692E-02 2.0345
    1/40 7.7616E-04 0.5176 2.7231E-03 1.9944 3.0641E-02 0.5176 7.3878E-03 2.0069
    1/80 5.0234E-04 0.6277 6.8122E-04 1.9990 1.9831E-02 0.6277 1.8531E-03 1.9952
    1/10 2.4652E-02 - 4.2689E-02 - 9.7312E-01 - 1.2182E-01 -
    -0.5 1/20 7.1878E-03 1.7781 1.0845E-02 1.9768 2.8376E-01 1.7779 2.9872E-02 2.0279
    1/40 1.9420E-03 1.8880 2.7184E-03 1.9962 7.6702E-02 1.8874 7.5673E-03 1.9809
    1/80 5.5906E-04 1.7965 6.7658E-04 2.0064 2.2071E-02 1.7971 2.0329E-03 1.8962

     | Show Table
    DownLoad: CSV
    Table 7.  Time convergence results with \alpha = 0.5 , \beta = 0.7 , h = 1/2000 .
    \|u_h-u\| \|q_h-q\|
    \theta \tau Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 1.1456E-02 - 4.3948E-02 - 4.5228E-01 - 6.8880E-02 -
    0.5 1/20 5.0680E-03 1.1767 1.1176E-02 1.9754 2.0008E-01 1.1767 1.6307E-02 2.0786
    1/40 2.2184E-03 1.1919 2.8060E-03 1.9939 8.7577E-02 1.1919 4.0110E-03 2.0235
    1/80 9.6796E-04 1.1965 7.0250E-04 1.9979 3.8213E-02 1.1965 9.8850E-04 2.0207
    1/10 5.4375E-03 - 4.3948E-02 - 2.1466E-01 - 6.8904E-02 -
    0.2 1/20 2.5220E-03 1.1084 1.1176E-02 1.9754 9.9562E-02 1.1084 1.6332E-02 2.0769
    1/40 1.1387E-03 1.1472 2.8053E-03 1.9941 4.4952E-02 1.1472 4.0351E-03 2.0170
    1/80 5.0160E-04 1.1828 7.0188E-04 1.9989 1.9802E-02 1.1827 1.0126E-03 1.9946
    1/10 2.8636E-02 - 4.3948E-02 - 1.1304E+00 - 6.8904E-02 -
    -1 1/20 7.7097E-03 1.8931 1.1176E-02 1.9754 3.0439E-01 1.8929 1.6332E-02 2.0769
    1/40 1.9098E-03 2.0133 2.8053E-03 1.9941 7.5437E-02 2.0126 4.0351E-03 2.0170
    1/80 6.3574E-04 1.5869 7.0188E-04 1.9989 2.5098E-02 1.5877 1.0126E-03 1.9946

     | Show Table
    DownLoad: CSV

    In Figure 2, we obtain the comparison images between the numerical solution and the exact solution with \tau = 1/1000 , h = 1/80 , \alpha = 0.9 , \beta = 0.2 , and \theta = 0.2 . In Figures 3 and 4, we present the space and time convergence rate images of u_h and q_h under different parameters \alpha , \beta , and \theta . From Figure 4, one can see that the corrected scheme with the starting parts can effectively restore the second-order convergence rate for the nonsmooth problem.

    Figure 2.  u_h , q_h and u , q with \tau = 1/2000 , h = 1/80 , \alpha = 0.9 , \beta = 0.2 , \theta = 0.2 .
    Figure 3.  The spatial convergence rates in L^2 -errors with different parameters \alpha , \beta , and \theta .
    Figure 4.  The time convergence rates in L^2 -errors with different parameters \alpha , \beta , and \theta .

    Example 5.3. To better investigate the effect of changes of two fractional parameters \alpha and \beta on the convergence rates, we introduce the numerical example with two nonsmooth terms. Here, we take the nonsmooth solution u with

    u = (t^{1+\alpha}+t^{1+\beta}+t^3) \sin(2\pi x),

    and the known source term

    \begin{equation} \begin{split} g(x,t) = &\sin(2\pi x)\left[(1+\alpha)t^\alpha+(1+\beta)t^\beta+3t^2+4\pi^2\left(t\Gamma(2+\alpha)+\frac{t^{1+\beta-\alpha}\Gamma(2+\beta)}{\Gamma(2+\beta-\alpha)}+\frac{6t^{3-\alpha}}{\Gamma(4-\alpha)}\right)\right]\\ &+\sin(2\pi x)\left[16\pi^4\left(\frac{t^{1+\alpha-\beta}\Gamma(2+\alpha)}{\Gamma(2+\alpha-\beta)}+t\Gamma(2+\beta)+\frac{6t^{3-\beta}}{\Gamma(4-\beta)}\right)+4\pi^2(t^{1+\alpha}+t^{1+\beta}+t^3)\right]\\ &+2\pi (t^{1+\alpha}+t^{1+\beta}+t^3)\cos(2\pi x)+2\pi(t^{1+\alpha}+t^{1+\beta}+t^3)^2\sin(4\pi x ). \end{split} \end{equation} (5.3)

    In Table 8, we provide the errors of \|u_h-u\| and \|q_h-q\| and the spatial convergence rates under different parameters, which indicate that the corrected term hardly affects the spatial convergence rate.

    Table 8.  Spatial convergence results with \alpha = 0.5 , \beta = 0.6 , \tau = 1/2000 .
    \|u_h-u\| \|q_h-q\|
    \theta h Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 6.5710E-02 - 6.5710E-02 - 1.1335E-01 - 1.1335E-01 -
    0.5 1/20 1.6709E-02 1.9755 1.6709E-02 1.9755 2.7037E-02 2.0678 2.7037E-02 2.0678
    1/40 4.1946E-03 1.9940 4.1946E-03 1.9940 6.6773E-03 2.0176 6.6775E-03 2.0175
    1/80 1.0498E-03 1.9984 1.0498E-03 1.9984 1.6615E-03 2.0068 1.6617E-03 2.0066
    1/10 6.5710E-02 - 6.5710E-02 - 1.1336E-01 - 1.1336E-01 -
    0.2 1/20 1.6708E-02 1.9755 1.6708E-02 1.9755 2.7042E-02 2.0676 2.7042E-02 2.0676
    1/40 4.1944E-03 1.9940 4.1944E-03 1.9940 6.6826E-03 2.0167 6.6825E-03 2.0167
    1/80 1.0497E-03 1.9986 1.0497E-03 1.9985 1.6668E-03 2.0033 1.6667E-03 2.0034
    1/10 6.5709E-02 - 6.5709E-02 - 1.1339E-01 - 1.1339E-01 -
    -0.5 1/20 1.6708E-02 1.9756 1.6708E-02 1.9756 2.7078E-02 2.0661 2.7077E-02 2.0662
    1/40 4.1935E-03 1.9943 4.1935E-03 1.9943 6.7188E-03 2.0109 6.7178E-03 2.0110
    1/80 1.0487E-03 1.9995 1.0487E-03 1.9995 1.7031E-03 1.9801 1.7020E-03 1.9808

     | Show Table
    DownLoad: CSV

    In Tables 911, fixing \tau = 1/4000 , choosing h = 1/10, 1/20, 1/40, 1/80 , and changing parameters \alpha , \beta , and \theta , we provide the L^2 -errors and the time convergence rates for u and q based on the corrected scheme and uncorrected scheme. The impact of different fractional parameters on the time convergence rates of nonsmooth problems is evident from Tables 911. Furthermore, one can see that the corrected scheme with the starting part can effectively restore the second-order convergence rate.

    Table 9.  Time convergence results with \alpha = 0.1 , \beta = 0.9 , h = 1/4000 .
    \|u_h-u\| \|q_h-q\|
    \theta \tau Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 8.9204E-03 - 6.2820E-03 - 3.5216E-01 - 2.4809E-01 -
    0.1 1/20 5.2261E-03 0.7714 1.8535E-03 1.7609 2.0632E-01 0.7714 7.3209E-02 1.7608
    1/40 2.6562E-03 0.9764 4.9663E-04 1.9001 1.0486E-01 0.9764 1.9628E-02 1.8992
    1/80 1.2855E-03 1.0470 1.2782E-04 1.9580 5.0749E-02 1.0470 5.0645E-03 1.9544
    1/10 1.1291E-02 - 9.1163E-03 - 4.4575E-01 - 3.5998E-01
    0 1/20 6.8182E-03 0.7277 2.7085E-03 1.7510 2.6917E-01 0.7277 1.0696E-01 1.7509
    1/40 3.4949E-03 0.9641 7.2773E-04 1.8960 1.3797E-01 0.9641 2.8751E-02 1.8954
    1/80 1.6987E-03 1.0408 1.8766E-04 1.9553 6.7062E-02 1.0408 7.4268E-03 1.9528
    1/10 2.1457E-02 - 2.6030E-02 - 1.1105E+00 - 1.7696E+00 -
    -0.5 1/20 1.4552E-02 0.5603 8.3150E-03 1.6464 8.5207E-01 0.3822 6.2053E-01 1.5118
    1/40 7.7887E-03 0.9017 2.2980E-03 1.8554 4.7736E-01 0.8359 1.7762E-01 1.8047
    1/80 3.8350E-03 1.0222 6.0067E-04 1.9357 2.3888E-01 0.9988 4.7205E-02 1.9118

     | Show Table
    DownLoad: CSV
    Table 10.  Time convergence results with \alpha = 0.5 , \beta = 0.6 , h = 1/4000 .
    \|u_h-u\| \|q_h-q\|
    \theta \tau Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 4.3621E-03 - 3.6704E-03 - 1.7226E-01 - 1.4498E-01 -
    0.5 1/20 1.0876E-03 2.0039 9.5152E-04 1.9476 4.2941E-02 2.0042 3.7576E-02 1.9480
    1/40 3.3133E-04 1.7148 2.4178E-04 1.9765 1.3080E-02 1.7150 9.5356E-03 1.9784
    1/80 1.1132E-04 1.5735 6.1138E-05 1.9835 4.3948E-03 1.5735 2.3985E-03 1.9912
    1/10 1.4643E-03 - 1.0614E-03 - 5.8318E-02 - 4.2803E-02 -
    0.2 1/20 4.8956E-04 1.5806 3.0032E-04 1.8214 1.9326E-02 1.5934 1.2126E-02 1.8197
    1/40 1.6670E-04 1.5542 7.8625E-05 1.9335 6.5811E-03 1.5541 3.1895E-03 1.9267
    1/80 5.7051E-05 1.5470 1.9768E-05 1.9918 2.2523E-03 1.5470 8.1546E-04 1.9676
    1/10 8.7609E-03 - 6.8360E-03 - 3.4601E-01 - 2.7011E-01 -
    0 1/20 2.2404E-03 1.9673 1.9129E-03 1.8374 8.8510E-02 1.9669 7.5599E-02 1.8371
    1/40 5.6418E-04 1.9895 5.0061E-04 1.9340 2.2304E-02 1.9886 1.9798E-02 1.9330
    1/80 1.5985E-04 1.8194 1.2738E-04 1.9746 6.3107E-03 1.8214 5.0505E-03 1.9709

     | Show Table
    DownLoad: CSV
    Table 11.  Time convergence results with \alpha = 0.9 , \beta = 0.1 , h = 1/4000 .
    \|u_h-u\| \|q_h-q\|
    \theta \tau Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 1.2779E-03 - 4.4043E-04 - 5.0452E-02 - 2.5629E-02 -
    0.1 1/20 1.0267E-03 0.3157 1.1804E-04 1.8996 4.0534E-02 0.3158 7.0942E-03 1.8531
    1/40 7.5884E-04 0.4362 3.0793E-05 1.9386 2.9957E-02 0.4362 1.8662E-03 1.9265
    1/80 5.0089E-04 0.5993 8.1141E-06 1.9241 1.9774E-02 0.5993 4.7828E-04 1.9642
    1/10 4.3621E-03 - 1.4545E-03 - 7.7482E-02 - 6.1261E-02 -
    0 1/20 1.8873E-03 0.6822 3.9496E-04 1.8807 4.6434E-02 0.7387 1.6763E-02 1.8697
    1/40 9.2034E-04 0.3539 1.0228E-04 1.9492 3.6333E-02 0.3539 4.3730E-03 1.9386
    1/80 6.1337E-04 0.5854 2.5740E-05 1.9904 2.4215E-02 0.5854 1.1169E-03 1.9692
    1/10 9.0746E-03 - 6.1035E-03 - 3.5896E-01 - 2.4209E-01 -
    -0.1 1/20 2.3761E-03 1.9332 1.7550E-03 1.7982 9.4044E-02 1.9324 6.9621E-02 1.7980
    1/40 7.1647E-04 1.7296 4.6664E-04 1.9111 2.8284E-02 1.7333 1.8526E-02 1.9100
    1/80 5.9682E-04 0.2636 1.1984E-04 1.9612 2.3561E-02 0.2636 4.7711E-03 1.9572

     | Show Table
    DownLoad: CSV

    To further validate the performance of the parameter \theta in numerical simulations with nonsmooth solutions, we provide the computing data in Table 12, from which one can see that the parameter \theta still needs to satisfy \theta\le \min\{\alpha, \beta, \frac{1}{2}\} , whether before or after correction. Notably, when \theta is negative, as long as it is not much less than 0, we can still obtain second-order convergence accuracy.

    Table 12.  Time convergence results with \alpha = 0.7 , \beta = 0.3 , h = 1/4000 .
    \|u_h-u\| \|q_h-q\|
    \theta \tau Erroro Rate Errorc Rate Erroro Rate Errorc Rate
    1/10 3.3675E-03 - 2.0604E-03 - 1.3333E-01 - 8.2027E-02 -
    0.31 1/20 1.2977E-03 1.3757 5.3018E-04 1.9584 5.1287E-02 1.3783 2.1127E-02 1.9570
    1/40 1.9442E-03 -0.5833 1.4145E-04 1.9062 7.6738E-02 -0.5813 5.6232E-03 1.9096
    1/80 6.1733E-02 -4.9888 1.3510E-04 0.0663 2.4372E+00 -4.9892 5.3212E-03 0.0797
    1/10 4.5472E-02 - 3.0346E-02 - 1.7952E+00 - 1.1983E+00 -
    -0.5 1/20 1.2127E-02 1.9067 9.1861E-03 1.7240 4.7880E-01 1.9066 3.6273E-01 1.7240
    1/40 3.1344E-03 1.9520 2.4851E-03 1.8861 1.2376E-01 1.9518 9.8141E-02 1.8859
    1/80 7.9608E-04 1.9772 6.4304E-04 1.9504 3.1447E-02 1.9766 2.5407E-02 1.9496
    1/10 1.1762E+00 - 3.4223E-01 - 4.6430E+01 - 1.3512E+01 -
    -5 1/20 3.9521E-01 1.5734 1.9704E-01 0.7964 1.5601E+01 1.5734 7.7788E+00 0.7966
    1/40 1.2069E-01 1.7113 7.7857E-02 1.3396 4.7643E+00 1.7113 3.0736E+00 1.3396
    1/80 3.3458E-02 1.8509 2.4673E-02 1.6579 1.3208E+00 1.8509 9.7402E-01 1.6579

     | Show Table
    DownLoad: CSV

    The time convergence rates of u and q are compared before and after correction with different parameters \alpha , \beta , and \theta in Figure 5, where the slope of the line segment indicates the convergence rate. The slope of each line segment in the corrected images is the same regardless of the parameters chosen, indicating that the introduction of the starting part has a significant effect on the time convergence rates for the case with nonsmooth solutions.

    Figure 5.  The time convergence rates in L^2 -errors with different parameters \alpha , \beta , and \theta .

    In this article, the spatial mixed finite element method with the generalized BDF2- \theta for solving the time-fractional generalized Rosenau-RLW-Burgers equation was presented. Detailed proofs of stability were shown. The numerical scheme's effectiveness and feasibility were verified by conducting numerical examples that included both smooth and nonsmooth solutions. The numerical examples with good regularity indicated that our algorithm with changed parameters \alpha , \beta , and \theta can maintain second-order convergence in time. Especially, the nonsmooth examples demonstrated that adding the correction term could effectively solve the problem of reduced order caused by weak singularity.

    N. Yang: Writing–original draft, Formal analysis, Software; Y. Liu: Methodology, Validation, Formal analysis, Funding acquisition, Supervision, Writing–review & editing. All authors have read and agreed to the published version of the manuscript.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    This work was supported by the National Natural Science Foundation of China (12061053), Young Innovative Talents Project of Grassland Talents Project and Program for Innovative Research Team in Universities of Inner Mongolia Autonomous Region (NMGIRT2413).

    The authors declare that they have no conflicts of interest.



    [1] O. Abd-Elaziz, M. Sayed, M. Abdullah, Liver tumors segmentation from abdominal CT images using region growing and morphological processing, in International Conference on Engineering and Technology (ICET), 2015. https://doi.org/10.1109/ICEngTechnol.2014.7016813
    [2] S. Rafiei, N. Karimi, B. Mirmahboub, K. Najarian, S. Soroushmehr, Liver segmentation in abdominal CT images using probabilistic atlas and adaptive 3D region growing, in 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2019. https://doi.org/10.1109/EMBC.2019.8857835
    [3] S. Tran, C. Cheng, D. Liu, A multiple layer U-Net, Un-Net, for liver and liver tumor segmentation in CT, IEEE Access, 9 (2020), 3752–3764. https://doi.org/10.1109/ACCESS.2020.3047861 doi: 10.1109/ACCESS.2020.3047861
    [4] P. Sofia, A. Juan, S. Manuel, A. Roberto, M. Alicia, D. Maceira, Automatic multi-atlas liver segmentation and couinaud classification from CT volumes, Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., 2021 (2021), 2826–2829. https://doi.org/10.1109/EMBC46164.2021.9630668 doi: 10.1109/EMBC46164.2021.9630668
    [5] L. Song, H. Wang, Z. Wang, Bridging the gap between 2D and 3D contexts in CT volume for liver and tumor segmentation, IEEE J. Biomed. Health Inf., 9 (2021), 3450–3459. https://doi.org/10.1109/JBHI.2021.3075752 doi: 10.1109/JBHI.2021.3075752
    [6] J. Zhang, B. Ji, Z. Jiang, J. Qin, CR-UNet: Context-rich UNet for liver segmentation from CT volumes, in 2021 International Conference on Electronic Information Engineering and Computer Science (EIECS), 2021. https://doi.org/10.1109/EIECS53707.2021.9588086
    [7] T. Lei, R. Wang, Y. Zhang, A. Nandi, DefED-Net: Deformable encoder-decoder network for liver and liver tumor segmentation, IEEE Trans. Radiat. Plasma Med. Sci., 6 (2022), 68–78. https://doi.org/10.1109/TRPMS.2021.3059780 doi: 10.1109/TRPMS.2021.3059780
    [8] X. Fang, S. Xu, B. Wood, P. Yan, Deep learning-based liver segmentation for fusion-guided intervention, Int. J. Comput. Assisted Radiol. Surg., 15 (2020), 963–972. https://doi.org/10.1007/s11548-020-02147-6 doi: 10.1007/s11548-020-02147-6
    [9] T. Amina, L. Lakhdar, B. Hakim, M. Abdallah, Improved active contour model through automatic initialisation: Liver segmentation, in 2021 IEEE 1st International Maghreb Meeting of the Conference on Sciences and Techniques of Automatic Control and Computer Engineering MI-STA, 2021. https://doi.org/10.1109/MI-STA52233.2021.9464516
    [10] M. N. U. Haq, A. Irtaza, N. Nida, M. A. Shah, L. Zubair, Liver tumor segmentation using resnet based mask-R-CNN, in 2021 International Bhurban Conference on Applied Sciences and Technologies (IBCAST), 2021. https://doi.org/10.1109/IBCAST51254.2021.9393194
    [11] X. Wang, Y. Zheng, G. Lan, W. Xuan, X. Sang, X. Kong, et al., Liver segmentation from CT images using a sparse priori statistical shape model (SP-SSM), PLoS One, 12 (2017), 1–23. https://doi.org/10.1371/journal.pone.0185249 doi: 10.1371/journal.pone.0185249
    [12] W. Qin, J. Wu, F. Han, Y. Yuan, W. Zhao, B. Ibragimov, et al., Superpixel-based and boundary-sensitive convolutional neural network for automated liver segmentation, Phys. Med. Biol., 9 (2018), 1–19. https://doi.org/10.1088/1361-6560/aabd19 doi: 10.1088/1361-6560/aabd19
    [13] Y. Yao, Y. Sang, Z. Zhao, Y. Cao, Research on segmentation and recognition of liver CT image based on multi-scale feature fusion, in 2021 2nd International Symposium on Computer Engineering and Intelligent Communications (ISCEIC), 2021. https://doi.org/10.1109/ISCEIC53685.2021.00075
    [14] S. Shao, X. Zhang, R. Cheng, C. Deng, Semantic segmentation method of 3D liver image based on contextual attention model, in 2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2021. https://doi.org/10.1109/SMC52423.2021.9659018
    [15] C. Li, Y. Tan, W. Chen, X. Luo, Y. Gao, X. Jia, et al., Attention Unet++: A nested attention-aware U-Net for liver CT image segmentation, in 2020 IEEE International Conference on Image Processing (ICIP), 2020. https://doi.org/10.1109/ICIP40778.2020.9190761
    [16] X. Yan, K. Yuan, W. Zhao, S. Wang, Z. Li, S. Cui, An efficient hybrid model for kidney tumor segmentation in CT images, in 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI), 2020. https://doi.org/10.1109/ISBI45749.2020.9098325
    [17] N. Thein, A. Nugroho, T. Bharata, K Hamamoto, An image preprocessing method for kidney stone segmentation in CT scan images, in 2018 International Conference on Computer Engineering, Network and Intelligent Multimedia (CENIM), 2018. https://doi.org/10.1109/CENIM.2018.8710933
    [18] G. Yang, G. Li, T. Pan, Y. Kong, X. Zhu, Automatic segmentation of kidney and renal tumor in CT images based on 3D fully convolutional neural network with pyramid pooling module, in 2018 24th International Conference on Pattern Recognition (ICPR), 2018. https://doi.org/10.1109/ICPR.2018.8545143
    [19] M. Arafat, G. Hamarne, R. Garbi, Cascaded regression neural nets for kidney localization and segmentation-free volume estimation, IEEE Trans. Med. Imaging, 40 (2021), 1555–1567. https://doi.org/10.1109/TMI.2021.3060465 doi: 10.1109/TMI.2021.3060465
    [20] J. Chen, X. Zhang, J. Wang, Coarse-to-fine deformable model-based kidney 3D segmentation, in 2019 WRC Symposium on Advanced Robotics and Automation (WRC SARA), 2019. https://doi.org/10.1109/WRC-SARA.2019.8931969
    [21] S. Yin, Z. Zhang, H. Li, Q. Peng, X. You, S. L. Furth, et al., Fully-automatic segmentation of kidneys in clinical ultrasound images using a boundary distance regression network, in 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019), 2019. https://doi.org/10.1109/ISBI.2019.8759170
    [22] R. Statkevych, S. Stirenko, Y. Gordienko, Human kidney tissue image segmentation by U-Net models, in IEEE EUROCON 2021-19th International Conference on Smart Technologies, 2021. https://doi.org/10.1109/EUROCON52738.2021.9535599
    [23] M. Li, Y. Chen, X. Zheng, K. Liu, Kidney region of interest extraction based on iterative convolution threshold method, in 2021 6th International Conference on Communication, Image and Signal Processing (CCISP), 2021. https://doi.org/10.1109/CCISP52774.2021.9639090
    [24] T. Les, T. Markiewcz, M. Dziekiewicz, M. Lorent, Kidney segmentation from computed tomography images using U-Net and batch-based synthesis, Comput. Biol. Med., 123 (2020), 103906. https://doi.org/10.1016/j.compbiomed.2020.103906 doi: 10.1016/j.compbiomed.2020.103906
    [25] H. R. Torres, S. Queirós, P. Morais, B. Oliveira, J. Vilaca, Kidney segmentation in 3-D ultrasound images using a fast phase-based approach, IEEE Trans. Ultrason., Ferroelectr., Freq. Control, 68 (2021), 1521–1531. https://doi.org/10.1109/TUFFC.2020.3039334 doi: 10.1109/TUFFC.2020.3039334
    [26] N. Weerasinghe, N. Lovell, A. Welsh, G. Stevenson, Multi-parametric fusion of 3D power doppler ultrasound for fetal kidney segmentation using fully convolutional neural networks, IEEE J. Biomed. Health Inf., 25 (2021), 2050–2057. https://doi.org/10.1109/JBHI.2020.3027318 doi: 10.1109/JBHI.2020.3027318
    [27] J. Guo, W. Zeng, S. Yu, J. Xiao, RAU-Net: U-Net model based on residual and attention for kidney and kidney tumor segmentation, in 2021 IEEE International Conference on Consumer Electronics and Computer Engineering (ICCECE), 2021. https://doi.org/10.1109/ICCECE51280.2021.9342530
    [28] T. M. Geethanjali, Minavathi, M. S. Dinesh, Semantic segmentation of tumors in kidneys using attention U-Net models, international conference on electrical, in 2021 5th International Conference on Electrical, Electronics, Communication, Computer Technologies and Optimization Techniques (ICEECCOT), 2021. https://doi.org/10.1109/ICEECCOT52851.2021.9708025
    [29] C. Li, C. Xu, C. Gui, M. Fox, Distance regularized level set evolution and its application to image segmentation, IEEE Trans. Image Process., 19 (2010), 3243–3254. https://doi.org/10.1109/TIP.2010.2069690 doi: 10.1109/TIP.2010.2069690
    [30] T. Chan, L. Vese, Active contours without edges, IEEE Trans. Image Process., 10 (2001), 266–277. https://doi.org/10.1109/83.902291 doi: 10.1109/83.902291
    [31] C. Feng, D. Zhao, M. Huang, Image segmentation and bias correction using local inhomogeneous iNtensity clustering (LINC): A region-based level set method, Neurocomputing, 219 (2017), 107–129. https://doi.org/10.1016/j.neucom.2016.09.008 doi: 10.1016/j.neucom.2016.09.008
    [32] C. Li, C. Kao, J. Gore, Z. Ding, Minimization of region scalable fitting energy for image segmentation, IEEE Trans. Image Process., 17 (2008), 1940–1949. https://doi.org/10.1109/TIP.2008.2002304 doi: 10.1109/TIP.2008.2002304
  • Reader Comments
  • © 2022 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(2517) PDF downloads(91) Cited by(0)

Figures and Tables

Figures(6)  /  Tables(2)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog