
As one of the critical branches of medical image processing, the task of segmentation of breast cancer tumors is of great importance for planning surgical interventions, radiotherapy and chemotherapy. Breast cancer tumor segmentation faces several challenges, including the inherent complexity and heterogeneity of breast tissue, the presence of various imaging artifacts and noise in medical images, low contrast between the tumor region and healthy tissue, and inconsistent size of the tumor region. Furthermore, the existing segmentation methods may not fully capture the rich spatial and contextual information in small-sized regions in breast images, leading to suboptimal performance. In this paper, we propose a novel breast tumor segmentation method, called the transformer and graph convolutional neural (TS-GCN) network, for medical imaging analysis. Specifically, we designed a feature aggregation network to fuse the features extracted from the transformer, GCN and convolutional neural network (CNN) networks. The CNN extract network is designed for the image's local deep feature, and the transformer and GCN networks can better capture the spatial and context dependencies among pixels in images. By leveraging the strengths of three feature extraction networks, our method achieved superior segmentation performance on the BUSI dataset and dataset B. The TS-GCN showed the best performance on several indexes, with Acc of 0.9373, Dice of 0.9058, IoU of 0.7634, F1 score of 0.9338, and AUC of 0.9692, which outperforms other state-of-the-art methods. The research of this segmentation method provides a promising future for medical image analysis and diagnosis of other diseases.
Citation: Haiyan Song, Cuihong Liu, Shengnan Li, Peixiao Zhang. TS-GCN: A novel tumor segmentation method integrating transformer and GCN[J]. Mathematical Biosciences and Engineering, 2023, 20(10): 18173-18190. doi: 10.3934/mbe.2023807
[1] | Miguel Vivas-Cortez, Muhammad Aamir Ali, Artion Kashuri, Hüseyin Budak . Generalizations of fractional Hermite-Hadamard-Mercer like inequalities for convex functions. AIMS Mathematics, 2021, 6(9): 9397-9421. doi: 10.3934/math.2021546 |
[2] | Hasan Kara, Hüseyin Budak, Mehmet Eyüp Kiriş . On Fejer type inequalities for co-ordinated hyperbolic ρ-convex functions. AIMS Mathematics, 2020, 5(5): 4681-4701. doi: 10.3934/math.2020300 |
[3] | Yanping Yang, Muhammad Shoaib Saleem, Waqas Nazeer, Ahsan Fareed Shah . New Hermite-Hadamard inequalities in fuzzy-interval fractional calculus via exponentially convex fuzzy interval-valued function. AIMS Mathematics, 2021, 6(11): 12260-12278. doi: 10.3934/math.2021710 |
[4] | Sabila Ali, Shahid Mubeen, Rana Safdar Ali, Gauhar Rahman, Ahmed Morsy, Kottakkaran Sooppy Nisar, Sunil Dutt Purohit, M. Zakarya . Dynamical significance of generalized fractional integral inequalities via convexity. AIMS Mathematics, 2021, 6(9): 9705-9730. doi: 10.3934/math.2021565 |
[5] | Thabet Abdeljawad, Muhammad Aamir Ali, Pshtiwan Othman Mohammed, Artion Kashuri . On inequalities of Hermite-Hadamard-Mercer type involving Riemann-Liouville fractional integrals. AIMS Mathematics, 2021, 6(1): 712-725. doi: 10.3934/math.2021043 |
[6] | Ghulam Farid, Hafsa Yasmeen, Hijaz Ahmad, Chahn Yong Jung . Riemann-Liouville Fractional integral operators with respect to increasing functions and strongly (α,m)-convex functions. AIMS Mathematics, 2021, 6(10): 11403-11424. doi: 10.3934/math.2021661 |
[7] | Maimoona Karim, Aliya Fahmi, Shahid Qaisar, Zafar Ullah, Ather Qayyum . New developments in fractional integral inequalities via convexity with applications. AIMS Mathematics, 2023, 8(7): 15950-15968. doi: 10.3934/math.2023814 |
[8] | Shuang-Shuang Zhou, Saima Rashid, Muhammad Aslam Noor, Khalida Inayat Noor, Farhat Safdar, Yu-Ming Chu . New Hermite-Hadamard type inequalities for exponentially convex functions and applications. AIMS Mathematics, 2020, 5(6): 6874-6901. doi: 10.3934/math.2020441 |
[9] | Jamshed Nasir, Shahid Qaisar, Saad Ihsan Butt, Hassen Aydi, Manuel De la Sen . Hermite-Hadamard like inequalities for fractional integral operator via convexity and quasi-convexity with their applications. AIMS Mathematics, 2022, 7(3): 3418-3439. doi: 10.3934/math.2022190 |
[10] | XuRan Hai, ShuHong Wang . Hermite-Hadamard type inequalities based on the Erdélyi-Kober fractional integrals. AIMS Mathematics, 2021, 6(10): 11494-11507. doi: 10.3934/math.2021666 |
As one of the critical branches of medical image processing, the task of segmentation of breast cancer tumors is of great importance for planning surgical interventions, radiotherapy and chemotherapy. Breast cancer tumor segmentation faces several challenges, including the inherent complexity and heterogeneity of breast tissue, the presence of various imaging artifacts and noise in medical images, low contrast between the tumor region and healthy tissue, and inconsistent size of the tumor region. Furthermore, the existing segmentation methods may not fully capture the rich spatial and contextual information in small-sized regions in breast images, leading to suboptimal performance. In this paper, we propose a novel breast tumor segmentation method, called the transformer and graph convolutional neural (TS-GCN) network, for medical imaging analysis. Specifically, we designed a feature aggregation network to fuse the features extracted from the transformer, GCN and convolutional neural network (CNN) networks. The CNN extract network is designed for the image's local deep feature, and the transformer and GCN networks can better capture the spatial and context dependencies among pixels in images. By leveraging the strengths of three feature extraction networks, our method achieved superior segmentation performance on the BUSI dataset and dataset B. The TS-GCN showed the best performance on several indexes, with Acc of 0.9373, Dice of 0.9058, IoU of 0.7634, F1 score of 0.9338, and AUC of 0.9692, which outperforms other state-of-the-art methods. The research of this segmentation method provides a promising future for medical image analysis and diagnosis of other diseases.
Chaos theory has developed over decades, especially in nonlinear dynamical systems, since its discovery by [1]. It has been applied in a variety of fields, such as disease control and prevention [2], mechanics [1,3], biology [4], cryptography [5], secure communications [6], and other application areas. Various published works [7,8,9] have presented the concept of two different types of attractors in continuous nonlinear dynamic systems, particularly regarding the behavior and the emergence or disappearance of equilibrium (fixed) points. An attractor is called a self-excited attractor if it has a basin of attraction that is associated with an unstable equilibrium. Conversely, an attractor is called hidden if its basin of attraction does not intersect with a small neighborhood of any equilibrium point [10]. The presence of multiple attractors, known as multistability, is a common phenomenon observed in many scientific fields and in nature [11]. Generally, the convergence to one of these attractors depends solely on the initial conditions or a specific set of parameters. Bifurcation analysis tools, particularly bifurcation diagrams, are essential for investigating the multiple attractors and qualitative properties of nonlinear dynamical systems. A wide range of software applications have been developed to calculate and display bifurcations, such as MATCONT [12], AUTO-07p [13], and PyDSTool [14]. In this context, the authors of [15] presented an advanced algorithm and specialized software for plotting bifurcation diagrams, which used graphics processing units (GPUS) accelerated computations in conjunction with a highly efficient semi-implicit ordinary differential equation (ODE) solver.
Hidden attractors have recently emerged as an important topic in a variety of theoretical and applied areas of continuous dynamic systems. These attractors can be identified within dynamic systems, categorized into the following four types: dynamic systems with no equilibrium points [16,17,18], systems that only have steady equilibrium points [19,20], systems with an infinite number of equilibrium points [21], systems with coexisting self-excited attractors [22].
The Jerk system has emerged as one of the chaotic systems that have recently attracted significant interest. Many researchers have studied the Jerk system in the presence of continuous behavior [23], where some researchers have looked at the Jerk-like system in the presence of discontinuous behavior [24]. Other authors [25,26], have directed their attention towards exploring the presence of periodic orbits, multistability, and the coexistence of attractors within specific parameter regions. Futhermore, [22] has demonstrated that both versions of the Jerk system exhibit multistability, where different types of attractors, such as periodic, chaotic, or a combination of both, can exist together in the phasespace, regardless of whether they are self-excited or hidden. Recently, the authors of [27] highlighted the presence of both single- and dual-scroll expansions in the three-dimensional (3D) and four-dimensional (4D) Jerk systems.
Moreover, there have been a few studies that examined the Jerk system in the presence of discontinuous behavior. Linz and Sprott [28] developed a comprehensive range of piecewise-linear Jerk functions. The dependence of these systems is exclusively on the switching between two vectors, dictated by the sign of one variable, thereby inhibiting the occurrence of a sliding mode. Hosham [24] proposed a novel chaotic Jerk system that is defined across four distinct domains designated by codimension-2 discontinuity surfaces, and it has been discovered that the chaotic behavior can exhibit a segment of sliding motion. In general, the existence of discontinuous behavior is responsible for the complex phenomena observed in real systems [29,30], particularly in ecological contexts with epidemic models [31].
This study is motivated by the desire to propose and analyze a nonlinear Jerk-like system with discontinuous behavior, with a particular emphasis on the detection of hidden and self-excited attractors. One advantage of the proposed system is its ability to reveal the existence and absence of equilibrium points on the basis of defined arbitrary functions. We explore nonsmooth properties because they are critical for identifying and facilitating rapid transient responses to dynamic events that influence a system's behaviors. To obtain periodic orbits numerically, a system of nonlinear algebraic equations needs to be set up, based on analytical solutions. A variety of Jerk-like system classes, including both crossing and sliding solutions, are investigated with the goal of detecting self-excited and hidden attractors. To examine the dynamic properties of these systems, various tools are used, including analytical criteria for discontinuous solutions, the Poincarè return map, the spectrum of Lyapunov exponents, and bifurcation diagrams.
The novelty of this work compared with prior studies [16,23,32] is summarized as follows.
● We present a novel Jerk system that incorporates discontinuities, enabling the exhibition of both sliding and crossing modes. This highlights the unique contribution and significance of the present work in advancing the study of discontinuous Jerk systems.
● Flexibility in equilibrium behavior is shown, where the proposed systems may (or may not) exhibit equilibrium points on one or both sides of the discontinuity surface.
● A novel and interesting result is the discovery of a periodic orbit completely limited to the sliding mode, which is an uncommon and noteworthy phenomenon in the study of discontinuous Jerk systems.
The structure of this paper is as follows: Section 2 introduces the proposed discontinuous Jerk-like system and investigates its nonsmooth properties, focusing on the fundamental principles governing flow interactions on a codimension-1 switching surface. Section 3 includes a theoretical analysis as well as numerical simulations of various classes of Jerk-like systems. This includes examining crossing solutions in order to identify self-excited and hidden attractors. Section 4 presents hidden attractors with a sliding mode. Finally, the conclusion of this paper is given in Section 5.
In physics, a Jerk equation takes the general form
⃛x=J(x,˙x,¨x), | (2.1) |
where J is a smooth function in R3, and the term "Jerk" refers to the definition of successive derivatives in mechanical systems. Moreover, x, ˙x, ¨x, and ⃛x are the displacement, velocity, acceleration, and Jerk, respectively. Jerk systems are frequently considered to be the simplest sort of chaotic systems [26] and possess several applications in scientific and engineering fields [32]. The most commonly reported Jerk systems use quadratic, cubic, exponential, and hyperbolic nonlinearities, and they attempt to design and implement circuits for these systems [33]. Jerk systems with some type of discontinuity have recently gained substantial interest and have been extensively reported in the pertinent literature because of their simplicity and complex dynamics; see, for instance, [24].
In our research, we analyze Jerk-like systems, which are described by piecewise differential systems composed of two differential systems separated by a switching surface ℏ(x,˙x,¨x)=0, as follows:
⃛x={Ja(x,˙x,¨x),ℏ(x,˙x,¨x)<0;Jb(x,˙x,¨x),ℏ(x,˙x,¨x)>0. | (2.2) |
The motivation for this work is to study the discontinuous behavior of the nonlinear Jerk-like system above in the presence of hidden and self-excited attractors. Recent research on discontinuous dynamic systems, as presented in [34], demonstrates that a planar linear nonsmooth system devoid of equilibria in each subsystem, whether real or virtual, can nonetheless display a minimum of one limit cycle. In our recent study [35], the hidden dynamics of the proposed three-dimensional discontinuous system are related to the fact that only one subsystem in the whole system lacks an equilibrium point. The investigation of the dynamics and bifurcations of the discontinuous system (2.2) has significantly increased, as researchers have come to understand that discontinuous systems can exhibit a diverse range of complicated dynamic phenomena that require attention in their applications. The grazing bifurcation framework for closed orbits [10], which includes the collision of closed orbits with the switching manifold, has been thoroughly explained and applied in systems containing dry friction and mechanical impact oscillators [36]. In general, bifurcations in discontinuous systems are categorized into two types: nonstandard bifurcations, which are unique to discontinuous systems such as sliding bifurcation or discontinuity-induced bifurcations, and standard bifurcations, which are identical to those found in continuous systems such as saddle-node and Hopf bifurcations.
The equation ℏ(x,˙x,¨x)=0 is considered to establish a surface of two dimensions, known as the switching manifold, which is indicated by D. The system (2.2) is written as a 3D first-order differential equation as follows:
˙η={Ja(η),ℏ(η)<0,Jb(η),ℏ(η)>0, | (2.3) |
where η=<x,˙x,¨x> is a vector in a 3D space.
Assume that U⊂R3 is open, ℏ(η):U→Rd, η∈R3, and d≤3 is a continuously differentiable function with a regular value of 0. To ensure there is only a codimension-1 discontinuity manifold, assume that the gradient vector ∇ℏ(η) does not vanish in any part of D. If ℏ(η)=˙ℏ(η)=....=ℏr−1(η)=0, then the motion of (2.2) is called an r-sliding mode; for example, see [37]. Let σa=(∇ℏ(η))TJa(η) and σb=(∇ℏ(η))TJb(η), where the elements of Ja(η) and Jb(η) are oriented towards ∇ℏ(η).
When a trajectory intersects with a discontinuity surface, the system's dynamics cannot be represented by a single continuous vector field. In other words, as the trajectory approaches a discontinuity surface, two options may arise: Cross the surface or stay on it. If we choose to stay on the surface, we must describe the motion that occurs there, specifically a sliding motion. Therefore, it becomes necessary to define additional rules to to characterize distinct potential behaviors.
Definition 1. The partial region of D is known as [36]
● A direct crossing zone Dc:={η∈R3∣σa(η)σb(η)>0}, which is divided into Dca:={η∈R3∣σa(η)<0,σb(η)<0}, and Dcb:={η∈R3∣σa(η)>0,σb(η)>0};
● An attracting sliding zone Dsa:={η∈R3∣σa(η)>0,andσb(η)<0};
● A repelling sliding zone Dsr:={η∈R3∣σa(η)<0,andσb(η)>0},
where D=Dc∪Dsa∪Dsr.
The transition between these modes highlights the complex dynamics inherent in the proposed nonsmooth Jerk system. When the orbit of (2.3) approaches an attracting sliding region, it may touch the sliding surface and proceed to slide along it. This behavior is characterized by the Fillipov extension [36,38] as follows:
˙η=Fs(η)=Sa(η)Ja(η)+Sb(η)Jb(η),η∈D, | (2.4) |
where Sa(η)=1−Sb(η), and Sb(η)=∇ℏ(η)TJb(η)∇ℏ(η)T(Jb(η)−Ja(η)). The borders of these regions are formed by the equations σa(η0)=0 and σb(η0)=0, where η0∈D are called the tangent points. This indicates that at any point of η0∈D, the orbit of Ja and Jb typically exhibits a quadratic tangency with D.
Definition 2. The point ˉη∈R3 is an equilibrium point for the subsystem ˙η=Ja(η) [˙η=Jb(η)] if Ja(ˉη)=0 [Jb(ˉη)=0]), and it is termed [24]:
● An admissible point if ˉη∈D− [ˉη∈D+];
● Virtual if ˉη∈D+ [ˉη∈D−];
● A boundary equilibrium if ℏ(ˉη)=0.
In addition, the sliding flow (2.4) exhibits a pseudo-equilibrium point if ℏ(ˉη)=Fs(ˉη)=0.
The bifurcation that emerges from these fixed points results in what is known as a discontinuous bifurcation, in which the generalized set-valued Jacobian is defined as follows:
DJ(ˉη)=(DJb(ˉη)−DJa(ˉη))α+DJa(ˉη),0≤α≤1. |
This formula describes the Jacobian's behavior at ˉη∈D. The literature [36,39] discussed discontinuous bifurcation and discontinuity-induced bifurcation in various examples.
Assume that ˜η∈Dca. The flow generated by ˙η=Ja(η) intersects D at time ta, after which it continues according to the flow defined by ˙η=Jb(η,ta), reaching D again at time tb. Hence, the variational equations for the trajectories for the piecewise smooth system (2.3) are
{˙η=Ja(η),η(0)=˜η:˙Ya=∂∂ηJa(˜η)Ya,Ya(0)=I;˙η=Jb(η),η(0)=˜η:˙Yb=∂∂ηJb(˜η)Yb,Yb(0)=I. | (2.5) |
At the intersection point and time (ˉta,˜η), we have Yb=ˉSYa, where ˉS is called the jump or saltation matrix, defined as. ˉS=I+(Jb(ˉta,˜η)−Ja(ˉta,˜η))(∇ℏ(˜η))T(∇ℏ(˜η))TJa(ˉta,˜η); see [39]. These properties serve as valuable tools for addressing discontinuities and refine the flow trajectory on D.
We consider the vector fields of (2.3) defined as follows:
˙η={Ja(η)=(yz−az−y2+Ga(η)),ℏ(η)<0,Jb(η)=(yz−az+y2+Gb(η)),ℏ(η)>0, | (2.6) |
where a>0 and the phase space is divided into two domains D±={X∈U∣±ℏ(η)>0} separated by a hyperplane D={X∈R3∣ℏ(η)=x=0}, and η=(x,y,z)T. Furthermore, this system has only the crossing mode and cannot exhibit the sliding mode due to Ds={∅}. The divergence of the system (2.6) is calculated using the following formula:
∇J=−a+∂Gi∂z,i=a,b. |
The choice of the function Gi is then determined by whether or not both subsystems in (2.6) are dissipative. The functions Gi(η) and ℏ(η) are crucial in defining the behavior of the proposed discontinuous systems, leading to different dynamic scenarios such as direct crossing and sliding modes, as discussed below.
This section aims to investigate the diverse range of dynamic behaviors exhibited by the system (2.6) through the utilization of phase portraits, time-domain waveforms, Poincaré maps, Lyapunov exponents, and bifurcation diagrams.
The equilibrium points of the system (2.6) can be determined by solving the subsequent equations Ja(η)=0andJb(η)=0, and η∈R3.
Definition 3. The subsystems described by (2.6) possess an equilibrium point denoted ˉη if the conditions Ga(ˉη)=0 and Gb(ˉη)=0 are satisfied. Consequently, all resulting attractors are classified as self-excited. Otherwise, the subsystems (2.6) have no equilibrium point, and hence the resulting attractors are all hidden.
The unknown functions within the system described by (2.6) are defined as follows: Ga(η)=Gb(η)=−x. Consequently, the origin emerges as the unique equilibrium point of this system. Furthermore, the Jerk system (2.6) is rewritten as
˙η=Aη+{−y2e3,x<0,y2e3,x≥0, | (3.1) |
where e3∈R3 is the third basis vector of R3, and the continuity property can be clearly identified, since the only variation between the two subsystems is found in the third row.
Therefore, at the point η0∈D, only a direct crossing region Dc is present, as both subsystems reflect the same value. Unlike Filippov systems (2.4), the systems defined by Eq (3.1), in conjunction with the continuity condition, cannot exhibit sliding behavior on D. Therefore, trajectories that approach D will cross it immediately. A coefficient matrix A∈R3×3 is a matrix consisting of the coefficients of the variables in a set of linearized systems (3.1). The characteristic equation for these linearized systems is provided as a depressed cubic equation λ3+aλ2+1=0. Since the discriminant, which is derived from the coefficients of this cubic equation, is negative, it implies that the cubic has a single real root λ1=μ∈R and two complex conjugate roots λ2,3=α±iβ, such that α=12μ2>0, and μ=−(a+2α)=−1/(α2+β2)<0 (i.e., the origin point is a saddle-focus point, indicating the potential of self-excited strange attractors). According to the Routh-Hurwitz criterion, the equilibrium point exhibits instability in both subsystems. Additionally, the Routh-Hurwitz criterion, which is derived from the Sturm series, provides the essential conditions for the occurrence of a Hopf bifurcation in the system (3.1).
Lemma 1. The dynamic system described by (3.1) exhibits a Hopf bifurcation if a→∞.
To prove the Lemma above, let us consider the Sturm series denoted by Si,i=0,1,2,...,k, which is associated with a characteristic equation λ3+aλ2+1=0. This sequence is generated using a recursive process that begins with the polynomial and its derivative. It is used to determine the number of real roots of the characteristic equation within a given interval by observing the sign variations throughout the sequence. The series Si is defined as follows:
Si(σ)=Ci,0σk−i+Ci,1σk−i−2+... |
where Ci,0 are indicative of what are termed Hurwitz determinants; for an in-depth discussion, see [40]. We find that the Hurwitz determinants C0,0=1>0, C1,0=a>0, and C1,1=−1<0. In addition, condition C2,0=−1a must approach zero to ensure that the roots of the characteristic equation possess negative real parts, with the exception of a purely imaginary conjugate pair, which arises as a approaches infinity.
The matrix A has a real eigenvalue μ associated with the eigenvector v1 and a pair of complex conjugate eigenvalues α±iβ, which correspond to the eigenvectors v2∓iv3. The general solution of the linear system ˙η=Aη can be expressed as follows:
η(t)=eμtξ1+eαt(ξ2(cos(βt)v2+sin(βt)v3)+ξ3(cos(βt)v3−sin(βt)v2)), | (3.2) |
where ξi,i=1,2,3 are determined by the initial condition η0. Since μ<0 and α>0, at t→∞, the term eμt governs a decay component, while eαt governs a growth component, which modifies the oscillatory terms. The oscillations may cause periodic behavior.
The general solution of the system (3.1) can be described using an exponential matrix as follows:
ˇP(η)=exp(At)η0±∫t0exp(A(t−s))y(s)2e3ds,for±eT1η0≥0, | (3.3) |
where ei∈R3,i=1,2,3; are the standard basis of a coordinate vector space. Note that the integration term in Eq (3.3) can be written explicitly; however, the resulting formulas are rather complex. Consequently, the dynamics of the system (3.1) can be analyzed using a discrete Poincarè map ˇP, with the Poincarè section established at the hypersurface D. Starting with the initial condition η0∈Dc−, the orbit that begins at η0 at zero time will re-enter Dc+ after the duration t=ta(η), reaching the point ηb=η(ta)∈Dc+. After the crossing event, the orbit continues to oscillate with the subsystem ˙η=Jb(η) until it reaches Dc− again, following the return time tb(η) at the crossing point ηa∈Dc−. The trajectories associated with the Poincarè half-maps ˇPa,b are specified below, and their combination gives rise to the first recurrence map, termed the generalized Poincarè map ˇP of the whole system (3.1).
ˇPa:Dc−→Dc+η0→ηb:=η(ta),ˇPb:Dc+→Dc−ηb→ηa:=η(η(ta),tb),ˇP(η):=ˇPb(ˇPa(η)). | (3.4) |
The existence and computation of the return intersection times are indeed challenging, as they remain unknown a priori and are implicitly characterized as nonlinear functions of η.
tr(η):=inf{t>0∣eT1ηr(t(η),η0)=0},r=a,b. | (3.5) |
The essential criterion for the existence of an invariant half-line on D that characterizes a periodic orbit is the equation ˇP(η)=η, which implies the existence of the quantities η,ta,tb. The identification of periodic orbits for the system (3.1) using Poincarè mapping notation requires the application of a root-finding algorithm [30] to resolve the subsequent nonlinear algebraic equations
0=S(X)=(ˇP(η)−ηeT1ˇPa(η)eT1ˇPb(η)ηTη−1), | (3.6) |
where X=(η,ta,tb,a)T are six independent variables, the fourth equation is a normalization condition. If the Jacobian matrix is nonsingular at a particular solution X0, it follows that the nonlinear system can be addressed through a Newton method as follows:
DSΔXn=−S(Xn),Xn+1=Xn+ΔXn. | (3.7) |
For example, assume that the initial guess for the Newton method is : η0=(0,−1.0,1.0)T∈Dc−. This technique computes the fixed point of the Poincarè mapping at η=(0,1.1970,−0.8483)T, ta=5.7313, tb=6.4437, and a=1.9; see Figure 2(a). Moreover, diverse strategies related to the continuation algorithm can be employed to examine families of periodic orbits as they vary with a system parameter.
The Lyapunov exponents of a nonlinear system characterize the convergence and divergence of its states. The presence of a positive Lyapunov exponent indicates that the system exhibits chaotic behavior. The Lyapunov exponents as a function of the varying parameter a are shown in Figure 1(a) for the dynamic system (3.1), with the initial condition (6.32,1,−1) and the time interval t∈[0,200]. For example, it has been observed that the Lyapunov characteristic exponents of the Jerk system (3.1) are estimated to be L1=0.3645,L2=0.00005, and L3=−1.8651. The Kaplan-Yorke dimension is obtained as:
DKY=2+L1+L2|L3|=2.1957. |
This result supports the hypothesis that the Jerk system described in Eq (3.1) is chaotic, as indicated by the presence of one positive Lyapunov exponent.
In addition, the bifurcation diagram, a critical tool in nonlinear theory, is used to numerically investigate the dynamic behavior of the system. This exploration is carried out using a fourth-order Runge-Kutta algorithm, which is supported by a robust event detection method that reliably computes the switching points. In Figure 1(b), the bifurcation diagrams of the variable z versus the parameter a. In more detail, as the value of a decreases from a=1.9, the system transitions from a steady state of period-1 (illustrated in Figure 2(a)) to a scenario of period doubling (depicted in Figure 2(b)), subsequently leading to a chaotic state, as evidenced by the chaotic attractor presented in a three-dimensional space in Figure 2(c).
In this context, we characterize the unknown functions present in the system represented by (2.6) as follows:
˙η={Ja(η)=(yz−az−y2+a−y),x<0;Jb(η)=(yz−az+y2−x),x≥0. | (3.8) |
It is clear that the first subsystem of (3.8) does not have an equilibrium point (that is, when x<0), while the origin appears to be a unique equilibrium point within the second subsystem (that is, when x≥0). In addition, at η0∈D, a direct crossing region arises because both subsystems exhibit the same value at η0; that is, ∇ℏ(η0))TJa(η0)=∇ℏ(η0))TJb(η0)=y0. Note that the explicit solution ηb(t) is given by (3.3) when x≥0. However, the second subsystem of (3.8) is rewritten as
˙ηa=Ja=ˉAη+(a−y2)e3,x<0, | (3.9) |
where the matrix ˉA possesses the eigenvalues given by λ1=0, λ2,3=12(−a+√a2−4), with the associated eigenvectors V1=(1,0,0)T, V2=(λ2λ3,λ2,1)T, and V3=(λ3/λ2,λ3,1)T. We observe that the eigenvalue λ1=0 is relevant to the x-axis direction (x<0) in the dynamic system (3.9). This implies that perturbations along the x-axis remain unchanged in terms of exponential growth or decay, whereas perturbations in the y- and z-directions are pivotal in shaping the system's behavior. The discriminant a2−4 contributes to our understanding of the behavior of the subsystem indicated in (3.9). We are particularly concerned with the scenario in which a2<4, since this leads to complex eigenvalues and enables the system to display oscillatory behavior in proximity to the discontinuity surface. These eigenvalues and eigenvectors are utilized to derive the analytical solution of the linear system ˙ˉηa=ˉAˉη, enabling us to obtain the explicit solution for the nonlinear system (3.8). The solution ¯ηa(t) under the initial condition η0=(0,y0,z0)T∈D is given by
ˉηa(t)=(eαt(−C1cos(βt)−aC1−2y02βsin(βt))+C1eαt(C1+z02βsin(βt)+y0cos(βt))eαt(−az0+2y02βsin(βt)+z0cos(βt))), | (3.10) |
where α=−a2,β=√4−a22, and C1=ay0+z0.
Assume that the flow represented by the earlier solution is redefined as ˉηa(t)=ϕ(η0,t). Hence, the general solution of the nonlinear system (3.9) is expressed as follows:
ˇPa=ˉηa(t)+∫t0ˉηa(t−s)(a−y(s)2)e3ds. | (3.11) |
The transition time ta is determined through numerical methods by solving the equation
eT1ˇPa(η)=0 | (3.12) |
As an example, one can utilize the technique established in Eqs (3.4), (3.6), and (3.7) to compute the transition times and the fixed point that results in a periodic orbit. According to the findings derived from Newton's method, the values are ta=1.9925, tb=7.649, η0=(0,0.8485,0.5544)T, and a=1.85. Figure 3(a) demonstrates the periodic orbit of the system outlined in the Eq (3.8), as derived from these results. Figures 3(a), (b), (c), and (d) depict the system transitioning from a period-1 steady state to a chaotic state.
In order to achieve chaotic behavior, it is essential that the solution to the initial value problem (3.8), η0∈Dc, exhibits oscillatory characteristics determined by the eigenvalues. The parameter range 0<a<2 ensures that the solution sets for both subsystems are oscillatory in nature, a consequence of having two pairs of complex-conjugate eigenvalues.
The dynamical behavior of the system (3.8) is analyzed using bifurcation diagrams and Lyapunov exponents. For the values of the chosen parameters a=1.6 and η0=(0,−1.7031,2.0477)T, the Lyapunov exponents of the chaotic system (3.8) are obtained as follows : L1=0.3527, L2≈0, and L3=−1.9527. The negative sum of Li for i=1,2,3 confirms that the novel system defined by (3.8) has dissipative properties. The Kaplan–Yorke dimension is calculated as
DKY=2+L1+L2|L3|=2.1806. |
The Lyapunov exponents as a function of the varying parameter a are illustrated in Figure 4 (left). The bifurcation diagram is also obtained for the range 1.55≤a≤1.85, and is shown in Figure 4 (right). The system is demonstrated to transition into chaos via a series of period doublings, confirming the presence of chaos; see Figure 4.
This section investigates the dynamic behavior that occurs when neither subsystem in (2.6) has any equilibrium points. It is worth noting that the study in [34] examined a discontinuous linear planar system separated by a nonregular line and proved that piecewise-linear systems without equilibria can exhibit limit cycles. These results indicate that the discontinuous system can exhibit unique behavior compared with continuous systems.
The unknown system functions Ga(η) and Gb(η) are illustrated in the following discontinuous system as follows:
˙η={Ja(η)=(yz−az−y2+a−y),x<0;Jb(η)=(yz−az+y2−(1+xz)),x≥0. | (3.13) |
It is obvious that there are no equilibrium points in both vector fields. Traditional methods of stability analysis, such as linearization around fixed points, cannot be used directly because neither region has any equilibrium points. Nonetheless, we can analyze the overall flow characteristics to better understand the system's behavior. At η0∈D, (i.e., x=0), it is evident that the system transitions between Ja(η) and Jb(η), with the behaviour near this boundary being dictated by the trajectory that approaches it.
● If x→0−, the dynamics are governed by Ja(η)
Ja(η)=(yz−az−y2+a−y). |
● If x→0+, the dynamics are governed by Jb(η)
Jb(η)=(yz−az+y2−(1+xy)). |
The discontinuities in ˙z at x=0 can cause abrupt changes in the system's trajectory. This behavior introduces complex switching dynamics, which depend on the trajectory's direction and velocity as it crosses the boundary x=0. The bifurcation diagram and the Lyapunov exponent spectrum are employed to analyze the discontinuous system (3.13) as a function of its control parameter to establish the type of scenario that produces complex dynamics. When we set a=0.77 and the initial conditions (0.0,−1.7031,2.0477), the Lyapunov exponents are calculated as: L1=0.4165,L2≈0,L3=−1.1865, and the Kaplan-Yorke dimension in the 3D case is calculated to be:
DKY=2+L1+L2|L3|=2.3511. |
The dynamics of the Lyapunov exponents are shown in Figure 5 (left). We also observe a cascade of period doubling through the bifurcation diagram (Figure 5 (right)).
The trajectories of the dynamical systems (3.1), (3.8), and (3.13) do not involve the sliding mode. We are redefining the discontinuity surface as D={X∈U∣ℏ(η)=x+z=0}. Thus, the discontinuous system indicated in (3.14) is revised to
˙η={Ja(η)=(yz−az−y2+a−y),x+z<0,Jb(η)=(yz−az+y2−(1+xz)),x+z>0. | (4.1) |
Since ℏ(η)=x+z, then σa=a(1−z)−y2 and σb=(y+0.5)2+(z−0.5a)2−a2+54. Therefore, the discontinuity set is divided into separate regions. The crossing and sliding regions are given as follows:
Dc−:={η∈R3∣a(1−z)<y2,(y+0.5)2+(z−0.5a)2<a2+54}, |
Dc+:={η∈R3∣a(1−z)>y2,(y+0.5)2+(z−0.5a)2>a2+54}, |
Dsa:={η∈R3∣a(1−z)>y2,(y+0.5)2+(z−0.5a)2<a2+54}, |
Dsr:={η∈R3∣a(1−z)<y2,(y+0.5)2+(z−0.5a)2>a2+54}. |
Figure 6(a) depicts the discontinuity surface ℏ(η), which is divided into the crossing and sliding regions.
Here, we focus on the case when the orbit of (4.1) gets close to an attracting sliding region Dsa, where it may come into interactions with the sliding surface and slide along it. The Fillipov extension describes the behavior of sliding motion in the following way:
˙η=Fs(η)=Sa(η)Ja(η)+Sb(η)Jb(η),η∈D, | (4.2) |
where Sa(η)=−(y+0.5)2+(z−0.5a)2−a2+54a+1−(z2+y(2y+1)) and Sb(η)=a(1−z)−y2a+1−(z2+y(2y+1)), and the system (4.2) is reduced to system of linear ordinary differential equations as follows:
˙η=Fs(η)=(yz−y),η∈D, | (4.3) |
Within this sliding flow, there is a pseudo-equilibrium point located at the origin that serves as the center point. The eigenvalues of the Jacobian matrix for the linear system (4.3) are λ1=0 and λ2,3=±i, which supports this characterization. In addition, the sliding system (4.3) behaves like a harmonic oscillator, with all variables oscillating in a periodic manner. It should be noted that the presence of a single zero eigenvalue is due to the sliding flow mode's singularity, which restricts its behavior to a two-dimensional domain.
The general solution for the sliding system (4.3), where η0∈Dsa, is derived as follows:
x(t)=−z0cos(t)+y0sin(t),y(t)=z0sin(t)+y0cos(t),z(t)=z0cos(t)−y0sin(t). | (4.4) |
Assuming η0∈Dsa, the sliding flow has two possible behaviours. In one case, the flow continues indefinitely within the sliding region, whereas in the other, the flow eventually reaches one of the boundaries. Solving one of the following equations yields the duration required for the flow of the sliding system to reach one of the boundaries σa=0 or σb=0.
E1=a(z0cos(t)−y0sin(t))+(z0sin(t)+y0cos(t))2−a=0,E2=(z0sin(t)+y0cos(t)+0.5)2+(z0cos(t)−y0sin(t)−0.5a)2−a2+54=0. |
Consider the scenario where η∈Dsa and the flow progresses towards the boundary σa=0. The sliding time flow represents the first positive zero of the function E1, as shown in Figure 6(b), as the parameter a changes.
Theorem 1. Assume that η0=(x0,0,z0)T∈Dsa and a−√a2+42<z0<1. The dynamics of the Jerk-like system (4.1), separated by the surface ℏ=x+z and lacking both real and virtual equilibria, reveal that within the sliding region surrounding the pseudo-equilibrium, a family of periodic orbits exists with the period t=2π.
Proof: The parameter a does not affect the general solution (4.4) of the sliding flow but does affect the identification of the attracting sliding region Dsa. Since η0=(x0,0,z0)∈Dsa, and according to the definition of the attracting sliding region Dsa (i.e., σ(η0)>0, and σb(η0)<0), it follows that a−√a2+42<z0<1.
We define the Poincarè map associated with the sliding flow (4.3) using the general solution (4.4), which is ˇPs(η):==Dsa→Dsa. The sliding flow is linear in η and homogeneous, allowing it to maintain linear homogeneity, as indicated by ˇPs(mη)=mˇP(η) and ∂ˇP∂η(mη)=∂ˇP∂η for some m∈R+. Evaluating the Poincarè map at the point η0=(x0,0,z0)T∈Dsa results in t=2π and ˇPs(mη0)=mη0. As a consequence, the sliding flow has a family of flat periodic orbits that occur only in the sliding region Dsa. Figure 7 depicts a family of periodic orbits that are exclusively in sliding mode, with a=0.9 and a=0.3.
Figure 8(a) and (b) highlights that the Jerk-like system (4.1) without equilibria establishes a single periodic orbit with a sliding mode at a=0.9 and η0=(0.2653,0.8251,−0.2653)T. Figure 8(c) and (d) highlights that when the parameter a is varied (that is, a=0.83), a period-doubling orbit appears with a sliding segment, with the initial condition set as η0=(0.3473,0.7280,−0.3473)T. Figure 9 shows a specific transition of behavior, including a sliding mode from the period-k orbit (k∈R) to chaos, which is detected by varying the parameter a. The Jerk-like system (4.1) exhibits a period-k orbit at a=0.8 with η0=(0.5829,−0.5083,−0.5829)T∈Dsa and chaotic behavior at a=0.75 with η0=(0.5208,−0.6836,−0.5208)T∈Dsa, as shown in Figure 9(a) and (b) and Figure 9(c) and (d), respectively.
We present a novel chaotic Jerk-like system distinguished by discontinuous vector fields separated by a codimension-1 switching surface D. A qualitative investigation was conducted on the proposed discontinuous systems generated from dissipative subsystems, both with and without equilibria. The study focused on the nonsmooth bifurcation characteristics of the switching surface D, where orbits may either cross or exhibit sliding behavior. While the new system may have equilibrium and nonequilibrium points, it generates periodic orbits, period-doubling, and chaotic dynamics that include both self-excited and hidden attractors. In particular, the intersection of attractors and sliding modes has been observed to exhibit an unusual phenomenon, resulting in the formation of a family of periodic orbits within the sliding region that surrounds the pseudo-equilibrium. The dynamic properties of the proposed system, both self-excited and hidden attractors, are confirmed by using phase portraits, geometric analysis, Poincaré maps, bifurcation diagrams, and Lyapunov exponents. Further analysis or numerical simulations are required to fully understand the impact of this switching on the system's overall behavior.
The Jerk systems with discontinuities exhibit sliding motion, a phenomenon in which trajectories slide over D. This sliding characteristic is particularly useful in control systems, mechanical systems with impacts, and electrical circuits with switching components. This paves the way for practical applications, including the development of robust sliding mode controllers, vibration isolation systems, and nonlinear oscillators for signal generation. Future research may benefit from employing continuation methods to obtain quantitative characterizations of the basins of attraction. Furthermore, the proposed system may exhibit various aspects of multistability due to its high sensitivity to the initial conditions and abrupt changes in behavior caused by the inequality of the Jacobians on either side of the discontinuity surface. This has motivated us to continue developing an accurate algorithm for efficiently managing the root identifying of the event function ℏ(η) in discontinuous Jerk systems (2.6) with multiple abrupt changes and recurrent reordering for the variational Eq (2.5).
The author declares that he has not used Artificial Intelligence (AI) tools in the creation of this article.
The author is thankful to the Deanship of Graduate Studies and Scientific Research at the University of Bisha for supporting this work through the Fast-Track Research Support Program.
The author declares no conflict of interest in this paper.
[1] |
M. H. Yap, G. Pons, J. Marti, S. Ganau, M. Sentis, R. Zwiggelaar, et al., Automated breast ultrasound lesions detection using convolutional neural networks, IEEE J. Biomed. Health Inf., 22 (2018), 1218–1226. https://doi.org/10.1109/JBHI.2017.2731873 doi: 10.1109/JBHI.2017.2731873
![]() |
[2] |
J. Gao, Q. Jiang, B. Zhou, D. Chen, Convolutional neural networks for computer-aided detection or diagnosis in medical image analysis: An overview, Math. Biosci. Eng., 16 (2019), 6536–6561. https://doi.org/10.3934/mbe.2019326 doi: 10.3934/mbe.2019326
![]() |
[3] |
C. Xu, Y. Qi, Y. Wang, M. Lou, J. Pi, Y. Ma, ARF-Net: An adaptive receptive gield network for breast mass segmentation in whole mammograms and ultrasound images, Biomed. Signal Process. Control, 71 (2022), 103178. https://doi.org/10.1016/j.bspc.2021.103178 doi: 10.1016/j.bspc.2021.103178
![]() |
[4] |
Y. Wang, N. Wang, M. Xu, J. Yu, C. Qin, X. Luo, et al., Deeply-supervised networks with threshold loss for cancer detection in automated breast ultrasound, IEEE Trans. Med. Imaging, 39 (2019), 866–876. https://doi.org/10.1109/TMI.2019.2936500 doi: 10.1109/TMI.2019.2936500
![]() |
[5] |
S. Jiang, J. Li, Z. Hua, Transformer with progressive sampling for medical cellular image segmentation, Math. Biosci. Eng., 19 (2022), 12104–12126. https://doi.org/10.3934/mbe.2022563 doi: 10.3934/mbe.2022563
![]() |
[6] |
A. Iqbal, M. Sharif, MDA-Net: Multiscale dual attention-based network for breast lesion segmentation using ultrasound images, J. King Saud Univ. Comput. Inf. Sci., 34 (2022), 7283–7299. http://dx.doi.org/10.1016/j.jksuci.2021.10.002 doi: 10.1016/j.jksuci.2021.10.002
![]() |
[7] |
R. Bi, C. Ji, Z. Yang, M. Qiao, P. Lv, H. Wang, Residual-based attention-Unet combing DAC and RMP modules for automatic liver tumor segmentation in CT, Math. Biosci. Eng., 19 (2022), 4703–4718. https://doi.org/10.3934/mbe.2022219 doi: 10.3934/mbe.2022219
![]() |
[8] |
B. Lei, S. Huang, R. Li, C. Bian, H. Li, Y. H. Chou, et al., Segmentation of breast anatomy for automated whole breast ultrasound images with boundary regularized convolutional encoder-decoder network, Neurocomputing, 321 (2018), 178–186. https://doi.org/10.1016/j.neucom.2018.09.043 doi: 10.1016/j.neucom.2018.09.043
![]() |
[9] |
Y. Ouyang, Z. Zhou, W. Wu, J. Tian, F. Xu, S. Wu, et al., A review of ultrasound detection methods for breast microcalcification, Math. Biosci. Eng., 16 (2019), 1761–1785. https://doi.org/10.3934/mbe.2019085 doi: 10.3934/mbe.2019085
![]() |
[10] |
M. N. S. K. B. Soulami, N. Kaabouch, A. Tamtaoui, Breast cancer: One-stage automated detection, segmentation, and classification of digital mammograms using U-net model based semantic segmentation, Biomed. Signal Process. Control, 2021 (2021), 102481. https://doi.org/10.1016/j.bspc.2021.102481 doi: 10.1016/j.bspc.2021.102481
![]() |
[11] |
Y. Wang, N. Wang, M. Xu, J. Yu, C. Qin, X. Luo, et al., Deeply-supervised networks with threshold loss for cancer detection in automated breast ultrasound, IEEE Trans. Med. Imaging, 39 (2019), 866–876. https://doi.org/10.1109/TMI.2019.2936500 doi: 10.1109/TMI.2019.2936500
![]() |
[12] |
E. H. Houssein, M. M. Emam, A. A. Ali, P. N. Suganthan, Deep and machine learning techniques for medical imaging-based breast cancer: A comprehensive review, Exp. Syst. Appl., 167 (2021), 114161. https://doi.org/10.1016/j.eswa.2020.114161 doi: 10.1016/j.eswa.2020.114161
![]() |
[13] |
M. Xian, Y. Zhang, H. D. Cheng, F. Xu, B. Zhang, J. Ding, Automatic breast ultrasound image segmentation: A survey, Pattern Recognit., 79 (2018), 340–355. https://doi.org/10.1016/j.patcog.2018.02.012 doi: 10.1016/j.patcog.2018.02.012
![]() |
[14] |
Y. Tong, Y. Liu, M. Zhao, L. Meng, J. Zhang, Improved U-net MALF model for lesion segmentation in breast ultrasound images, Biomed. Signal Process. Control, 68 (2021), 102721. https://doi.org/10.1016/j.bspc.2021.102721 doi: 10.1016/j.bspc.2021.102721
![]() |
[15] |
D. Mishra, S. Chaudhury, M. Sarkar, A. S. Soin, Ultrasound image segmentation: A deeply supervised network with attention to boundaries, IEEE Trans. Biomed. Eng., 66 (2018), 1637–1648. https://doi.org/10.1109/TBME.2018.2877577 doi: 10.1109/TBME.2018.2877577
![]() |
[16] |
G. Chen, Y. Dai, J. Zhang, C-Net: Cascaded convolutional neural network with global guidance and refinement residuals for breast ultrasound images segmentation, Comput. Methods Programs Biomed., 2022 (2022), 107086. https://doi.org/10.1016/j.cmpb.2022.107086 doi: 10.1016/j.cmpb.2022.107086
![]() |
[17] |
N. K. Tomar, D. Jha, M. A. Riegler, H. D. Johansen, D. Johansen, J. Rittscher, et al., Fanet: A feedback attention network for improved biomedical image segmentation, Trans. Neural Networks Learn. Syst., 2022 (2022). https://doi.org/10.1109/TNNLS.2022.3159394 doi: 10.1109/TNNLS.2022.3159394
![]() |
[18] |
L. C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, A. L. Yuille, DeepLab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs, Trans. Pattern Anal. Mach. Intell., 2018 (2018), 834–848. https://doi.org/10.1109/TPAMI.2017.2699184 doi: 10.1109/TPAMI.2017.2699184
![]() |
[19] | Y. Xie, J. Zhang, C. Shen, Y. Xia, Cotr: Efficiently bridging CNN and Transformer for 3d medical image segmentation, in Medical Image Computing and Computer Assisted Intervention MICCAI, (2021), 171–180. https://doi.org/10.1007/978-3-030-87199-4_16 |
[20] |
N. S. Punn, S. Agarwal, RCA-IUnet: A residual cross-spatial attention-guided inception U-Net model for tumor segmentation in breast ultrasound imaging, Mach. Vision Appl., 33 (2022), 1–10. ttps://doi.org/10.1007/s00138-022-01280-3 doi: 10.1007/s00138-021-01257-8
![]() |
[21] |
N. Abraham, N. M. B. T. Khan, A novel focal tversky loss function with improved attention U-Net for lesion segmentation, Int. Symp. Biomed. Imaging, 2019 (2019), 683–687. https://doi.org/10.1109/ISBI.2019.8759329 doi: 10.1109/ISBI.2019.8759329
![]() |
[22] | W. Jin, T. Derr, Y. Wang, Y. Ma, Z. Liu, J. Tang, Node similarity preserving graph convolutional networks, in Proceedings of the 14th ACM International Conference on Web Search and Data Mining, (2021), 148–156. https://doi.org/10.1145/3437963.3441735 |
[23] | B. Wu, X. Liang, X. Zheng, Y. Guo, H. Tang, Improving dynamic graph convolutional network with fine-grained attention mechanism, in ICASSP International Conference on Acoustics, Speech and Signal Processing (ICASSP), (2022), 3938–3942. https://doi.org/10.1109/ICASSP43922.2022.9746009 |
[24] | Y. Lu, Y. Chen, D. Zhao, J. Chen, Graph-FCN for image semantic segmentation, in Advances in Neural Networks–ISNN 2019: 16th International Symposium on Neural Networks, (2019), 97–105. https://doi.org/10.1007/978-3-030-22796-8_11 |
[25] | Y. Huang, Y. Sugano, Y. Sato, Improving action segmentation via graph-based temporal reasoning, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2020), 14024–14034. https://doi.org/10.1109/CVPR42600.2020.01404 |
[26] |
W. Al-Dhabyani, M. Gomaa, H. Khaled, A. Fahmy, Dataset of breast ultrasound images, Data Brief, 28 (2020), 104863. https://doi.org/10.1016/j.dib.2019.104863 doi: 10.1016/j.dib.2019.104863
![]() |
[27] |
Z. Fu, J. Zhang, R. Luo, Y. Sun, D. Deng, L. Xia, TF-Unet: An automatic cardiac MRI image segmentation method, Math. Biosci. Eng., 19 (2022), 5207–5222. https://doi.org/10.3934/mbe.2022244 doi: 10.3934/mbe.2022244
![]() |
[28] |
X. Xu, M. Zhao, P. Shi, R. Ren, X. He, X. Wei, et al., Crack detection and comparison study based on faster R-CNN and mask R-CNN, Sensors, 22 (2022), 1215. https://doi.org/10.3390/s22031215 doi: 10.3390/s22031215
![]() |
[29] | L. C. Chen, Y. Zhu, G. Papandreou, F. Schroff, H. Adam, Encoder-decoder with atrous separable convolution for semantic image segmentation, in Proceedings of the European Conference on Computer Vision (ECCV), (2018), 801–818. |
[30] |
R. Huang, M. Lin, H. Dou, Z. Lin, Q. Ying, X. Jia, et al., Boundary-rendering network for breast lesion segmentation in ultrasound images, Med. Image Anal., 80 (2022), 102478. https://doi.org/10.1016/j.media.2022.102478 doi: 10.1016/j.media.2022.102478
![]() |
[31] | Z. Zhou, M. M. R. Siddiquee, N. Tajbakhsh, J. Liang, UNet++: A nested U-net architecture for medical image segmentation, in Lecture Notes in Computer Science, (2018), 3–11. http: /dx.doi.org/10.1007/ 978-3-030-00889-51 |
[32] | E. Sanderson, B. J. Matuszewski, FCN-Transformer feature fusion for polyp segmentation, in Medical Image Understanding and Analysis: 26th Annual Conference, Springer International Publishing, (2022), 892–907. https://doi.org/10.1007/978-3-031-12053-4_65 |
[33] |
X. Feng, T. Wang, X. Yang, M. Zhang, W. Guo, W. Wang, ConvWin-UNet: UNet-like hierarchical vision transformer combined with convolution for medical image segmentation, Math. Biosci. Eng., 20 (2023), 128–144. https://doi.org/10.3934/mbe.2023007 doi: 10.3934/mbe.2023007
![]() |
[34] | O. Oktay, J. Schlemper, L. L. Folgoc, M. Lee, M. Heinrich, K. Misawa, et al., Attention U-Net: learning where to look for the pancreas, preprint, arXiv: 1804.03999. https://doi.org/10.48550/arXiv.1804.03999 |
[35] |
X. Zhang, K. Liu, K. Zhang, X. Li, Z. Sun, B. Wei, SAMS-Net: Fusion of attention mechanism and multi-scale features network for tumor infiltrating lymphocytes segmentation, Math. Biosci. Eng., 20 (2023), 2964–2979. https://doi.org/10.3934/mbe.2023140 doi: 10.3934/mbe.2023140
![]() |
[36] |
R. K. Meleppat, P. Zhang, M. J. Ju, S. K. K. Manna, Y. Jian, E. N. Pugh, et al., Directional optical coherence tomography reveals melanin concentration-dependent scattering properties of retinal pigment epithelium, J. Biomed. Optics, 24 (2019), 066011. https://doi.org/10.1117/1.JBO.24.6.066011 doi: 10.1117/1.JBO.24.6.066011
![]() |
[37] |
R. K. Meleppat, C. R. Fortenbach, Y. Jian, E. S. Martinez, K. Wagner, B. S. Modjtahedi, et al., In Vivo imaging of retinal and choroidal morphology and vascular plexuses of vertebrates using swept-source optical coherence tomography, Trans. Vis. Sci. Tech., 11 (2022), 11. https://doi.org/10.1167/tvst.11.8.11 doi: 10.1167/tvst.11.8.11
![]() |
[38] |
R. K. Meleppat, K. E. Ronning, S. J. Karlen, K. K. Kothandath, M. E. Burns, E. N. Pugh, et al., In situ morphologic and spectral characterization of retinal pigment epithelium organelles in mice using multicolor confocal fluorescence imaging, Invest. Ophthalmol. Vis. Sci., 61 (2020), 1. https://doi.org/10.1167/iovs.61.13.1 doi: 10.1167/iovs.61.13.1
![]() |
[39] |
J. He, Q. Zhu, K. Zhang, P. Yu, J. Tang, An evolvable adversarial network with gradient penalty for COVID-19 infection segmentation, Appl. Soft Comput., 113 (2021), 107947. https://doi.org/10.1016/j.asoc.2021.107947 doi: 10.1016/j.asoc.2021.107947
![]() |
[40] |
X. Liu, D. Zhang, J. Yao, J. Tang, Transformer and convolutional based dual branch network for retinal vessel segmentation in OCTA images, Biomed. Signal Process. Control, 83 (2023), 104604. https://doi.org/10.1016/j.bspc.2023.104604 doi: 10.1016/j.bspc.2023.104604
![]() |
[41] |
C. Zhao, A. Vij, S. Malhotra, J. Tang, H. Tang, D. Pienta, et al., Automatic extraction and stenosis evaluation of coronary arteries in invasive coronary angiograms, Comput. Biol. Med., 136 (2021), 104667. https://doi.org/10.1016/j.compbiomed.2021.104667 doi: 10.1016/j.compbiomed.2021.104667
![]() |
1. | Tao Yan, Ghulam Farid, Ayşe Kübra Demirel, Kamsing Nonlaopon, Sun Young Cho, Further on Inequalities for α , h − m -Convex Functions via k -Fractional Integral Operators, 2022, 2022, 2314-4785, 1, 10.1155/2022/9135608 | |
2. | Muhammad Samraiz, Kanwal Saeed, Saima Naheed, Gauhar Rahman, Kamsing Nonlaopon, On inequalities of Hermite-Hadamard type via n-polynomial exponential type s-convex functions, 2022, 7, 2473-6988, 14282, 10.3934/math.2022787 | |
3. | Muhammad Samraiz, Muhammad Umer, Artion Kashuri, Thabet Abdeljawad, Sajid Iqbal, Nabil Mlaiki, On Weighted (k, s)-Riemann-Liouville Fractional Operators and Solution of Fractional Kinetic Equation, 2021, 5, 2504-3110, 118, 10.3390/fractalfract5030118 | |
4. | Peng Xu, Saad Ihsan Butt, Saba Yousaf, Adnan Aslam, Tariq Javed Zia, Generalized Fractal Jensen–Mercer and Hermite–Mercer type inequalities via h-convex functions involving Mittag–Leffler kernel, 2022, 61, 11100168, 4837, 10.1016/j.aej.2021.10.033 | |
5. | YONGFANG QI, QINGZHI WEN, GUOPING LI, KECHENG XIAO, SHAN WANG, DISCRETE HERMITE–HADAMARD-TYPE INEQUALITIES FOR (s,m)-CONVEX FUNCTION, 2022, 30, 0218-348X, 10.1142/S0218348X22501602 | |
6. | Muhammad Tariq, Sotiris K. Ntouyas, Asif Ali Shaikh, A Comprehensive Review on the Fejér-Type Inequality Pertaining to Fractional Integral Operators, 2023, 12, 2075-1680, 719, 10.3390/axioms12070719 | |
7. | Miguel Vivas-Cortez, Muzammil Mukhtar, Iram Shabbir, Muhammad Samraiz, Muhammad Yaqoob, On Fractional Integral Inequalities of Riemann Type for Composite Convex Functions and Applications, 2023, 7, 2504-3110, 345, 10.3390/fractalfract7050345 | |
8. | Miguel Vivas-Cortez, Muhammad Samraiz, Muhammad Tanveer Ghaffar, Saima Naheed, Gauhar Rahman, Yasser Elmasry, Exploration of Hermite–Hadamard-Type Integral Inequalities for Twice Differentiable h-Convex Functions, 2023, 7, 2504-3110, 532, 10.3390/fractalfract7070532 | |
9. | Muhammad Samraiz, Maria Malik, Saima Naheed, Gauhar Rahman, Kamsing Nonlaopon, Hermite–Hadamard-type inequalities via different convexities with applications, 2023, 2023, 1029-242X, 10.1186/s13660-023-02957-7 | |
10. | Muhammad Samraiz, Saima Naheed, Ayesha Gul, Gauhar Rahman, Miguel Vivas-Cortez, Innovative Interpolating Polynomial Approach to Fractional Integral Inequalities and Real-World Implementations, 2023, 12, 2075-1680, 914, 10.3390/axioms12100914 | |
11. | Miguel Vivas-Cortez, Muhammad Samraiz, Aman Ullah, Sajid Iqbal, Muzammil Mukhtar, A modified class of Ostrowski-type inequalities and error bounds of Hermite–Hadamard inequalities, 2023, 2023, 1029-242X, 10.1186/s13660-023-03035-8 | |
12. | XIAOHUA ZHANG, YUNXIU ZHOU, TINGSONG DU, PROPERTIES AND 2α̃-FRACTAL WEIGHTED PARAMETRIC INEQUALITIES FOR THE FRACTAL (m,h)-PREINVEX MAPPINGS, 2023, 31, 0218-348X, 10.1142/S0218348X23501347 |