Research article Special Issues

Predicting the prognosis of HER2-positive breast cancer patients by fusing pathological whole slide images and clinical features using multiple instance learning

  • In 2022, breast cancer will become an important factor affecting women's public health and HER2 positivity for approximately 15–20% invasive breast cancer cases. Follow-up data for HER2-positive patients are rare, and research on prognosis and auxiliary diagnosis is still limited. In light of the findings obtained from the analysis of clinical features, we have developed a novel multiple instance learning (MIL) fusion model that integrates hematoxylin-eosin (HE) pathological images and clinical features to accurately predict the prognostic risk of patients. Specifically, we segmented the HE pathology images of patients into patches, clustered them by K-means, aggregated them into a bag feature-level representation through graph attention networks (GATs) and multihead attention networks, and fused them with clinical features to predict the prognosis of patients. We divided West China Hospital (WCH) patients (n = 1069) into a training cohort and internal validation cohort and used The Cancer Genome Atlas (TCGA) patients (n = 160) as an external test cohort. The 3-fold average C-index of the proposed OS-based model was 0.668, the C-index of the WCH test set was 0.765, and the C-index of the TCGA independent test set was 0.726. By plotting the Kaplan-Meier curve, the fusion feature (P = 0.034) model distinguished high- and low-risk groups more accurately than clinical features (P = 0.19). The MIL model can directly analyze a large number of unlabeled pathological images, and the multimodal model is more accurate than the unimodal models in predicting Her2-positive breast cancer prognosis based on large amounts of data.

    Citation: Yifan Wang, Lu Zhang, Yan Li, Fei Wu, Shiyu Cao, Feng Ye. Predicting the prognosis of HER2-positive breast cancer patients by fusing pathological whole slide images and clinical features using multiple instance learning[J]. Mathematical Biosciences and Engineering, 2023, 20(6): 11196-11211. doi: 10.3934/mbe.2023496

    Related Papers:

    [1] Zuhal Küçükarslan Yüzbașı, Dae Won Yoon . On geometry of isophote curves in Galilean space. AIMS Mathematics, 2021, 6(1): 66-76. doi: 10.3934/math.2021005
    [2] Haiming Liu, Jiajing Miao . Geometric invariants and focal surfaces of spacelike curves in de Sitter space from a caustic viewpoint. AIMS Mathematics, 2021, 6(4): 3177-3204. doi: 10.3934/math.2021192
    [3] Rawya A. Hussein, Ali A. Ali . Geometry of the line space associated to a given dual ruled surface. AIMS Mathematics, 2022, 7(5): 8542-8557. doi: 10.3934/math.2022476
    [4] M. Khalifa Saad, Nural Yüksel, Nurdan Oğraş, Fatemah Alghamdi, A. A. Abdel-Salam . Geometry of tubular surfaces and their focal surfaces in Euclidean 3-space. AIMS Mathematics, 2024, 9(5): 12479-12493. doi: 10.3934/math.2024610
    [5] Ibrahim AL-Dayel, Emad Solouma, Meraj Khan . On geometry of focal surfaces due to B-Darboux and type-2 Bishop frames in Euclidean 3-space. AIMS Mathematics, 2022, 7(7): 13454-13468. doi: 10.3934/math.2022744
    [6] Yanlin Li, Kemal Eren, Soley Ersoy . On simultaneous characterizations of partner-ruled surfaces in Minkowski 3-space. AIMS Mathematics, 2023, 8(9): 22256-22273. doi: 10.3934/math.20231135
    [7] Mehmet Önder . Non-null slant ruled surfaces. AIMS Mathematics, 2019, 4(3): 384-396. doi: 10.3934/math.2019.3.384
    [8] Ayman Elsharkawy, Clemente Cesarano, Abdelrhman Tawfiq, Abdul Aziz Ismail . The non-linear Schrödinger equation associated with the soliton surfaces in Minkowski 3-space. AIMS Mathematics, 2022, 7(10): 17879-17893. doi: 10.3934/math.2022985
    [9] Fatemah Mofarreh, Rashad A. Abdel-Baky . Singularities of swept surfaces in Euclidean 3-space. AIMS Mathematics, 2024, 9(9): 26049-26064. doi: 10.3934/math.20241272
    [10] Hassan Al-Zoubi, Bendehiba Senoussi, Mutaz Al-Sabbagh, Mehmet Ozdemir . The Chen type of Hasimoto surfaces in the Euclidean 3-space. AIMS Mathematics, 2023, 8(7): 16062-16072. doi: 10.3934/math.2023819
  • In 2022, breast cancer will become an important factor affecting women's public health and HER2 positivity for approximately 15–20% invasive breast cancer cases. Follow-up data for HER2-positive patients are rare, and research on prognosis and auxiliary diagnosis is still limited. In light of the findings obtained from the analysis of clinical features, we have developed a novel multiple instance learning (MIL) fusion model that integrates hematoxylin-eosin (HE) pathological images and clinical features to accurately predict the prognostic risk of patients. Specifically, we segmented the HE pathology images of patients into patches, clustered them by K-means, aggregated them into a bag feature-level representation through graph attention networks (GATs) and multihead attention networks, and fused them with clinical features to predict the prognosis of patients. We divided West China Hospital (WCH) patients (n = 1069) into a training cohort and internal validation cohort and used The Cancer Genome Atlas (TCGA) patients (n = 160) as an external test cohort. The 3-fold average C-index of the proposed OS-based model was 0.668, the C-index of the WCH test set was 0.765, and the C-index of the TCGA independent test set was 0.726. By plotting the Kaplan-Meier curve, the fusion feature (P = 0.034) model distinguished high- and low-risk groups more accurately than clinical features (P = 0.19). The MIL model can directly analyze a large number of unlabeled pathological images, and the multimodal model is more accurate than the unimodal models in predicting Her2-positive breast cancer prognosis based on large amounts of data.



    A surface that arises when a curve α(u) is translated over another curve β(v) is called a translation surface. A translation surface is defined to be the sum of two curves α(u) and β(v), which are called generating curves. Therefore, translation surfaces are made up of quadrilateral, that is, four sided, facets. Because of this property, translation surfaces are used in architecture to design and construct free-form glass roofing structures. A translation surface in Euclidean 3-space by two curves in orthogonal planes, is given in [2,3] as a graph of the function z(u,v)=f(u)+g(v), such that f(u) and g(v) are smooth functions defined on an interval of R.

    The translation surfaces in E3 studied by H. F. Scherk [1] are obtained as a graph of the sum z(u,v)=f(u)+g(v), and where he proved that the only minimal translation surfaces are the those denoted by

    z(u,v)=1alog|cos(au)cos(av)|=1alog|cos(au)|1alog|cos(av)|,

    where a is a non-zero constant. These surfaces[12,13] are nowadays said to be Scherk's minimal surfaces.

    Constant mean curvature and constant Gaussian curvature surfaces have been the main problem for the geometers for a long time. Translation surfaces with constant mean curvature, in particular, zero mean curvature, in the Euclidean space and the Minkowski space are stated in [7]. Lone and Karacan yield categorization of dual translation surface, which has constant dual isotropic mean curvature or constant dual isotropic Gaussian curvature [8]. Dillen, Goemans and Woestyne obtained a categorization of translation surfaces in the 3-dimensional Euclidean and Minkowski space such that the Weingarten circumstance is satisfying [5].

    Sipus and Divjak [11] explained translation surfaces in the Galilean space having constant Gaussian and mean curvatures together with translation Weingarten surfaces. Yoon searched coordinate finite type translation surfaces in G3 and explained some examples of new classes of translation surfaces [12]. He also studied a translation surface in G3 with a log-linear density and classified such a surface with vanishing weighted mean curvature[13].

    LN-surfaces, which were taken into account in [6] with enough elasticity to model smooth surfaces. Peternell and Odehnal generalized the idea of LN-surfaces to R4 [9]. Bulca calculated the Gaussian, normal and mean curvatures of LN- surfaces in E4. Further, she pointed out the flat and minimal points of the surfaces in [3]. Sampoli, Peternell and Jüttler indicate that even the convolution surface of an LN-surface and any rational surface admits rational parametrization [10]. Moreover, LN and LCN translation surfaces have been characterized in 3-dimensional affine and Euclidean spaces in [14,15].

    In this study, the dual translation surfaces defined by z=f(u)+g(v), are considered in the three dimensional Galilean space G3.

    We emphasize that the Galilean space G3 is a Cayley-Klein space associated with a 3-dimensional projective space P(R3) with the absolute figure. The reader is referred to [11,12] for more details of

    d(P1,P2)={|x2x1|      ,  if   x1x2(y2y1)2+(z2z1)2 ,  if   x1=x2.   (2.1)

    The group of motions of G3 is a six-parameter group given in affine coordinates by

    ¯x=a+x,¯y=b+cx+ycosθ+zsinθ,¯z=d+exysinθ+zcosθ,

    with respect to the absolute figure, there are two types of lines in the Galilean space - isotropic lines which intersect the absolute line f and non-isotropic lines which do not. A plane is called Euclidean if it includes f, whereas it is said isotropic. In the given affine coordinates, isotropic vectors are of the form (0,y,z), but Euclidean planes are of the form x=k,k R. The induced geometry of a Euclidean plane is Euclidean, and of an isotropic plane isotropic (i.e., 2-dimensional Galilean or flag-geometry).

    A Cr-surface S, r1, immersed in the Galilean space, Ψ :UG3,UR2, Ψ(u,v)=(x(u,v),y(u,v),z(u,v)) has the following first fundamental form

    I=(g1du+g2dv)2+ϵ(h11du2+2h12dudv+h22dv2),

    where the symbols gi=xi, hij=Ψi.Ψj stand for derivatives of the first coordinate function x(u,v) with respect to u,v and for the Euclidean scalar product of the projections Ψk of vectors Ψk onto the yz -plane, respectively. Furthermore,

    ϵ={0,   ifdirectiondu:dvisnonisotropic,1,         ifdirectiondu:dvisisotropic.

    At every point of a surface, there exists a unique isotropic direction defined by g1du+g2dv=0. In that direction, the arc length is measured by

    ds2=h11du2+2h12dudv+h22dv2=h11g222h12g1g2+h22g21g21=W2g21dv2,

    where g10. A surface is called admissible if it has no Euclidean tangent planes. Therefore, for an admissible surface either g10 or g20 holds. The Gaussian curvature K and mean curvature H are Cr2 functions, r2, defined by

    K=LNM2W2, H=g22L2g1g2M+g21N2W2,

    where

    Lij=x1ΨijxijΨ1x1U,  x1=g10.

    We will use Lij, i,j=1,2, for L, M, N. The vector U defines a normal vector to a surface and given by

    U=1W(0,x1z2x2z1,x2y1x1y2),

    where W2=(x2Ψ1x1Ψ2)2 [11]. It is well known that in terms of local coordinates {u,v} of S, the Laplacian operator ΔII according to the second fundamental form on S is defined by [4]

    ΔIIΨ=1LNM2[u(NΨuMΨvLNM2)v(MΨuLΨvLNM2)]. (2.2)

    Consider a surface in G3 as the graph of a function z=h(u,v). We will consider translation surfaces of Type 1 that are obtained by translating two planar curves. Translation surfaces of Type 1 in the Galilean 3-space can be locally represented by

    z=f(u)+g(v), (2.3)

    which yields the parametrization

    Ψ(u,v)=(u,v,f(u)+g(v)). (2.4)

    One translated curve is a non-isotropic curve in the plane y=0 and given as

    α(u)=(u,0,f(u)),

    and the other is an isotropic curve in the plane x=0 and it is expressed as

    β(v)=(0,v,g(v)).

    Let S be a Monge patch parametrized by z=h(u,v), then the surface

    Ψ=(zu,zv,z+uzu+vzv), (2.5)

    is called a dual surface of  S in G3 and the normal vector of S is given by

    U=(±(zuuzvv(zuv)2)W)(0,v,1). (2.6)

    The Gaussian curvature of the dual surface is

    K=1(zuuzvv(zuv)2)(1+v2)2. (2.7)

    Definition 1. A surface S in G3 is said to be Weingarten surface if and only if there exists a non-trivial functional relation Φ(K,H)=0 between the Gaussian curvature and mean curvature of S. The condition Φ(K,H)=0 is equivalent to the vanishing of the corresponding Jacobian determinant, i.e., |(K,H)(u,v)|=0. A linear Weingarten surface in Galilean 3-space is a surface where there exists a relation aK+bH=c for some non-zero real numbers a,b,c [5,10].

    In [11], Sipus and Divjak analyzed the translation surface of Type 1, which has the constant Gaussian and mean curvatures in the Galilean 3-space G3 and they showed the following theorems:

    Theorem 1. If S is a translation surface of Type 1 of constant Gaussian curvature in the Galilean 3-space, then S is congruent to a particular surface with

    f(x)=ax2+bx+c,

    a,b,cR, and

    k(u)=12Au2+Bu+C,h(u)=Au+BA1(Au+B)2+12Aarcsin(Au+B)+C1

    for A,B,C,C1R, where K=f(x)k(u) and h2+k2=1.

    Theorem 2. If S is a translation surface of Type 1 of constant mean curvature H 0 in the Galilean 3-space, then S is congruent to a surface

    z=f(u)12H1(2Hv+c1)2+c2,

    where c1,c2R.

    The main aim of this section is to give a classification of dual translation surfaces with constant Gaussian and mean curvatures in the Galilean 3-space. Let S  be a dual translation surface. Using (2.3) and (2.5), we can define dual translation surfaces as

    Ψ(u,v)=(f(u),g(v),uf(u)+vg(v)f(u)g(v)). (3.1)

    The unit normal vector field U of the surface (3.1) is

    U=11+v2(0,v,1). (3.2)

    By a straightforward computation, the Gaussian and mean curvatures of the dual translation surface are given by

    K=1fg(1+v2)2,H=12(1+v2)32g, (3.3)

    where f,g0. Thus we have the following result.

    Corollary 1. If S is a dual translation surface in the Galilean 3-space, then there are no flat and minimal dual translation surfaces.

    Proof. The proof is obvious by Eq (3.3).

    Theorem 3. Let S be a dual translation surface which has constant Gaussian curvature in G3. Then it is congruent to an open part of the surface (3.1) with

     z=f(u)+(c1+c2v+c32varctanv).

    Proof. Dual translation surface with constant curvature K0 is described by the ordinary differential equation

    K0fg(1+v2)21=0, K0R. (3.4)

    By partial differentiation of the previous expression with respect to v, we get

    f(4v(1+v2)g+(1+v2)2g)=0.

    Therefore either f=0 or

    4v(1+v2)g+(1+v2)2g=0. (3.5)

    If f=0, then we have a contradiction. The general solution of (3.5) is given by

    g(v)=c1+c2v+c32varctanv,

    where ciR. Therefore, S is determined by

     z=f(u)+(c1+c2v+c32varctanv), (3.6)

    where f0 (see Figure 1).

    Figure 1.  z=cosu+(c1+c2v+c32varctanv).

    Theorem 4. Let S be a dual translation surface which has constant mean curvature in G3. Then it is congruent to an open part of the surface (3.1) with

     z=f(u)+(c1+c2v1+v22H0).

    Proof. We suppose that dual translation surface has constant mean curvature. Then we obtain

    2(1+v2)32gH01=0, H0R. (3.7)

    After solving (3.7), we get

    g(v)=c1+c2v1+v22H0.

    Therefore S is determined by

     z=f(u)+(c1+c2v1+v22H0), (3.8)

    where ciR (see Figure 2).

    Figure 2.  z=cosu+(c1+c2v1+v22H0).

    Theorem 5. Let S be a Weingarten dual translation surface in G3. Then it is congruent to an open part of the surface (3.1) with

     z=(c1+c2u+c3u2)+g(v),

    or

     z=f(u)+(c4+c5v+c61+v2).

    Proof. The condition (K,H)(u,v)=0 must be satisfied for the Weingarten surface S, it leads to KuHvKvHu=0. Since the mean curvature Hv is a function of v only, the previous equation reduces KuHv=0. Therefore, either Ku =0 or Hv=0. The first condition Ku =0 describes that a dual translation surface satisfies

    f(1+v2)2f2g=0.

    Therefore f =0 which implies

    f(u)=c1+c2u+c3u2.

    If the second condition Hv=0, then we have

    3vg+(1+v2)g2(1+v2)52g2=0,

    and

    g(v)=c4+c5v+c61+v2,

    where ciR. Thus S is determined by either

     z=(c1+c2u+c3u2)+g(v), (3.9)

    see Figure 3,

    Figure 3.  z=cosv+(c1+c2u+c3u2).

    or

     z=f(u)+(c4+c5v+c61+v2), (3.10)

    see Figure 4.

    Figure 4.  z=eu+(c4+c5v+c61+v2).

    Theorem 6. Let S be a linear Weingarten dual translation surface satisfying aK+bH=c, where a,b,c are real constants. Then S is congruent to (3.1) with

    z=c1+c2u+m2u2+c3+c4v+bm1+v2+av arctanv2cm.

    Proof. Let us consider surface with a0,b0. A dual translation surface is linear Weingarten if and only if

    2a1+v2+f(bbv22cg(1+v2)(1+2v2+v4))=0, (3.11)

    which can be written as

    f=2a1+v2(bbv22cg(1+v2)(1+2v2+v4))=m,

    where mR. Therefore, we get

    f(u)=c1+c2u+m2u2,g(v)=c3+c4v+bm1+v2+av arctanv2cm, (3.12)

    where ci,mR and they satisfy the functional relationship aK+bH=c for any constants a,b,c, (see Figure 5).

    Figure 5.  z=c1+c2u+m2u2+c3+c4v+bm1+v2+av arctanv2cm.

    Theorem 7. Let S be a dual translation surface given by (3.1) in the three dimensional Galilean space G3. Then there is no surface S satisfies the condition ΔIIΨ=λiΨi, where λiR, i=1,2,3.

    Proof. We classify dual translation surfaces in G3 satisfying the equation

    ΔIIΨ=λixi, (3.13)

    where λiR, i=1,2,3 and

    ΔIIΨ=(ΔIIΨ1,ΔIIΨ2,ΔIIΨ3). (3.14)

    By a straightforward computation, using Eq (2.2), the Laplacian operator on S turns out to be

    ΔIIΨ=(1+v2f2f,1+v2g2g,21+v2+u1+v2f2f+v1+v2g2g). (3.15)

    So Eq (3.13) by means of (3.15) gives the following system of ordinary differential equations

    1+v2f2f=λ1f, (3.16)
    1+v2g2g=λ2g, (3.17)
    21+v2+u1+v2f2f+v1+v2g2g=λ3(uf(u)+vg(v)f(u)g(v)), (3.18)

    where λi R. On combining Eqs (3.16)–(3.18), we have

    21+v2+λ1uf+vλ2g=λ3(uf+vgfg), (3.19)

    which can be written as

    λ1ufλ3(uf+f)=m,λ3(vgg)λ2vg21+v2=m, (3.20)

    where λi,m R. If {λ1,λ2,λ3}={0,0,0}, hence we have a contradiction. In the cases {λ1,λ20,λ30}, there are no analytical solutions for the function g(v). We discuss three cases with respect to constants λ1,λ2,λ3. We summarized the solutions of (3.20) in the following:

    λ10,λ2=0,λ30{f(u)=mλ3+c1(u(λ1λ3))λ3λ1λ3g(v)=c2vλ3m21+v2+2varcsinvλ3},λ10,λ20,λ3=0{f(u)=c1+mln|u|λ1g(v)=c1λ221+v2(2+m)ln|v|+2ln|1+1+v2|λ2},λ1=0,λ2=0,λ30{f(u)=c1u+mλ3g(v)=c2vλ3m21+v2+2varcsinvλ3},

    for some constants ci0 and λi0. In the third row, we have a contradiction for the function f(u), because substituting the first and second rows into (3.14)–(3.16), respectively we can easily see that they do not satisfy these equations.

    Theorem 8. The dual translation surface H2K =0 is locally a generalized umbilical surface and it can be stated to be a surface (3.1) by

    z=c1+c2u+m2u2+c3+c4v+14m(v arctanv12ln|1+v2|).

    Proof. Next, we aim to classify the dual translation surfaces given by (3.1) in G3 that satisfy

    H2K =0. By considering Eq (3.3), we have

    f4g4v2g2(1+v2)3fg2=0. (3.21)

    We can rewrite (3.21) as

    f4g4v2g=0,

    which implies

    f=4g+4v2g=m.

    Solving this, we obtain

    f(u)=c1+c2u+m2u2,g(v)=c3+c4v+14m(v arctanv12ln|1+v2|), (3.22)

    where m,ci R (see Figure 6).

    Figure 6.  z=c1+c2u+m2u2+c3+c4v+14m(v arctanv12ln|1+v2|).

    A rational surface S in R3 is called LN-surface if its tangent planes admit the representation T(u,v)=ux+vy+z=h(u,v) with a rational function h(u,v)=f(u)+g(v). In the present study, we have defined these surfaces in Galilean 3-space and called these surfaces as dual translation surfaces.

    We also have calculated the Gaussian and mean curvatures of these surfaces. Further, we have pointed out the constant Gaussian and mean curvatures of these surfaces.

    We have discussed whether these surfaces satisfy the Laplacian operator with respect to the second fundamental form.

    The author states that she has no conflict of interests.



    [1] R. L. Siegel, K. D. Miller, H. E. Fuchs, A. Jemal, Cancer statistics, 2022, CA Cancer J. Clin., 72 (2022), 7–33. https://doi.org/10.3322/caac.21708
    [2] E. A. Perez, E. H. Romond, V. J. Suman, J. Jeong, G. Sledge, C. E. Geyer Jr, et al., Trastuzumab plus adjuvant chemotherapy for human epidermal growth factor receptor 2–Positive breast cancer: Planned joint analysis of overall survival from NSABP B-31 and NCCTG N9831, JCO, 32 (2014), 3744–3752. https://doi.org/10.1200/JCO.2014.55.5730 doi: 10.1200/JCO.2014.55.5730
    [3] C. L. Arteaga, M. X. Sliwkowski, C. K. Osborne, E. A. Perez, F. Puglisi, L. Gianni, Treatment of HER2-positive breast cancer: current status and future perspectives, Nat. Rev. Clin. Oncol., 9 (2012), 16–32. https://doi.org/10.1038/nrclinonc.2011.177 doi: 10.1038/nrclinonc.2011.177
    [4] J. N. Wang, B. H. Xu, Targeted therapeutic options and future perspectives for HER2-positive breast cancer, Sig. Transduct. Target Ther., 4 (2019), 34. https://doi.org/10.1038/s41392-019-0069-2 doi: 10.1038/s41392-019-0069-2
    [5] D. Cameron, M. J. Piccart-Gebhart, R. D. Gelber, M. Procter, A. Goldhirsch, E. de Azambuja, et al., 11 years' follow-up of trastuzumab after adjuvant chemotherapy in HER2-positive early breast cancer: Final analysis of the HERceptin Adjuvant (HERA) trial, Lancet, 389 (2017), 1195–1205. https://doi.org/10.1016/S0140-6736(16)32616-2 doi: 10.1016/S0140-6736(16)32616-2
    [6] Director's challenge consortium for the molecular classification of lung adenocarcinoma, Gene expression–based survival prediction in lung adenocarcinoma: a multi-site, blinded validation study, Nat. Med., 14 (2008), 822–827. https://doi.org/10.1038/nm.1790 doi: 10.1038/nm.1790
    [7] M. Y. Park, T. Hastie, L1-regularization path algorithm for generalized linear models, J Royal Statistical Soc B, 69 (2007), 659–677. https://doi.org/10.1111/j.1467-9868.2007.00607.x doi: 10.1111/j.1467-9868.2007.00607.x
    [8] E. Bair, R. Tibshirani, Semi-Supervised methods to predict patient survival from gene expression data, PLoS Biol., 2 (2004), 512–522. https://doi.org/10.1371/journal.pbio.0020108 doi: 10.1371/journal.pbio.0020108
    [9] A. Warth, T. Muley, M. Meister, A. Stenzinger, M. Thomas, P. Schirmacher, et al., The novel histologic international association for the study of lung cancer/American thoracic society/European respiratory society classification system of lung adenocarcinoma is a Stage-Independent predictor of survival, JCO, 30 (2012), 1438–1446. https://doi.org/10.1200/JCO.2011.37.2185 doi: 10.1200/JCO.2011.37.2185
    [10] B. Ehteshami Bejnordi, M. Veta, P. Johannes van Diest, B. van Ginneken, N. Karssemeijer, G. Litjens, et al., Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer, JAMA, 318 (2017), 2199. https://doi.org/10.1001/jama.2017.14585 doi: 10.1001/jama.2017.14585
    [11] Y. Yuan, H. Failmezger, O. M. Rueda, H. R. Ali, S. Gräf, S. Chin, et al., Quantitative image analysis of cellular heterogeneity in breast tumors complements genomic profiling, Sci. Transl. Med., 4 (2012). https://doi.org/10.1126/scitranslmed.3004330
    [12] J. Xu, Y. Cao, Y. Sun, J. Tang, Absolute exponential stability of recurrent neural networks with generalized activation function, IEEE Trans. Neural Networks, 19 (2008), 1075–1089, . https://doi.org/10.1109/TNN.2007.2000060 doi: 10.1109/TNN.2007.2000060
    [13] J. Tang, X. Liu, H. Cheng, K. M. Robinette, Gender recognition using 3-D human body shapes, IEEE Trans. Syst. Man Cybern. C, 41 (2011), 898–908. https://doi.org/10.1109/TSMCC.2011.2104950
    [14] X. Liu, J. Liu, X. Xu, L. Chun, J. Tang, Y. Deng, A robust detail preserving anisotropic diffusion for speckle reduction in ultrasound images, BMC Genom., 12 (2011), S14. https://doi.org/10.1186/1471-2164-12-S5-S14 doi: 10.1186/1471-2164-12-S5-S14
    [15] J. Tang, S. Millington, S. T. Acton, J. Crandall, S. Hurwitz, Ankle cartilage surface segmentation using directional gradient vector flow snakes 2004. IEEE Int. Conf. Inf. Process., 4 (2004), 2745–2748. https://doi.org/10.1109/ICIP.2004.1421672
    [16] J. Tang, S. Acton, An image retrieval algorithm using multiple query images, ISSPA 2003, 1 (2003), 193–196. https://doi.org/10.1109/ISSPA.2003.1224673 doi: 10.1109/ISSPA.2003.1224673
    [17] E. H. Cain, A. Saha, M. R. Harowicz, J. R. Marks, P. K. Marcom, M. A. Mazurowski, Multivariate machine learning models for prediction of pathologic response to neoadjuvant therapy in breast cancer using MRI features: a study using an independent validation set, Breast Cancer Res. Treat., 173 (2019), 455–463. https://doi.org/10.1007/s10549-018-4990-9 doi: 10.1007/s10549-018-4990-9
    [18] H. Wang, F. Xing, H. Su, A. Stromberg, L. Yang, Novel image markers for non-small cell lung cancer classification and survival prediction, BMC Bioinform., 15 (2014), 310. https://doi.org/10.1186/1471-2105-15-310 doi: 10.1186/1471-2105-15-310
    [19] Z. Hu, J. Tang, Z. Wang, K. Zhang, L. Zhang, Q. Sun Jr, Deep learning for image-based cancer detection and diagnosis-A survey, Pattern Recognition, 83 (2018), 134–149. https://doi.org/10.1016/j.patcog.2018.05.014 doi: 10.1016/j.patcog.2018.05.014
    [20] J. Yang, J. Ju, L. Guo, B. Ji, S. Shi, Z. Yang, et al., Prediction of HER2-positive breast cancer recurrence and metastasis risk from histopathological images and clinical information via multimodal deep learning, Comput. Struct. Biotel., 20 (2022), 333–342. https://doi.org/10.1016/j.csbj.2021.12.028 doi: 10.1016/j.csbj.2021.12.028
    [21] K. Yu, C. Zhang, G. J. Berry, R. B. Altman, C. Ré, D. L. Rubin, et al., Predicting non-small cell lung cancer prognosis by fully automated microscopic pathology image features, Nat. Commun., 7 (2016), 12474. https://doi.org/10.1038/ncomms12474 doi: 10.1038/ncomms12474
    [22] X. Liu, Z. Guo, J. Cao, J. Tang, MDC-net: A new convolutional neural network for nucleus segmentation in histopathology images with distance maps and contour information, Comput. Biol. Med., 135 (2021), 104543. https://doi.org/10.1016/j.compbiomed.2021.104543 doi: 10.1016/j.compbiomed.2021.104543
    [23] R. Yan, F. Ren, Z. Wang, L. Wang, T. Zhang, Y. Liu, Breast cancer histopathological image classification using a hybrid deep neural network, Methods, 173 (2020), 52–60. https://doi.org/10.1016/j.ymeth.2019.06.014 doi: 10.1016/j.ymeth.2019.06.014
    [24] X. Zhu, J. Yao, F. Zhu, J. Huang, Wsisa: Making survival prediction from whole slide histopathological images, in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2017), 7234–7242. https://doi.org/10.1109/CVPR.2017.725
    [25] R. Li, Graph CNN for survival analysis on whole slide pathological images, in Medical Image Computing and Computer Assisted Intervention, Springer International Publishing, (2018), 174–182. https://doi.org/10.1007/978-3-030-00934-2_20
    [26] J. Yao, X. Zhu, J. Jonnagaddala, N. Hawkins, J. Huang, Whole slide images based cancer survival prediction using attention guided deep multiple instance learning networks, Med. Image Anal., 65 (2022), 101789. https://doi.org/10.1016/j.media.2020.101789 doi: 10.1016/j.media.2020.101789
    [27] M. Y. Lu, D. F. K. Williamson, T. Y. Chen, R. J. Chen, M. Barbieri, F. Mahmood, Data-efficient and weakly supervised computational pathology on whole-slide images, Nat. Biomed. Eng., 5 (2021), 555–570. https://doi.org/10.1038/s41551-020-00682-w doi: 10.1038/s41551-020-00682-w
    [28] F. Wu, P. Liu, B. Fu, Y. Ye, DeepGCNMIL: Multi-head attention guided multi-instance learning approach for whole-slide images survival analysis using graph convolutional networks, ICMLC 2022, (2022), 67–-73. https://doi.org/10.1145/3529836.3529942
    [29] G. Campanella, M. G. Hanna, L. Geneslaw, A. Miraflor, V. Silva, K. J. Busam, et al., Clinical-grade computational pathology using weakly supervised deep learning on whole slide images, Nat. Med., 25 (2019), 1301–1309. https://doi.org/10.1038/s41591-019-0508-1 doi: 10.1038/s41591-019-0508-1
    [30] R. J. Chen, M. Y. Lu, J. Wang, D. F. K. Williamson, S. J. Rodig, N. I. Lindeman, et al., Pathomic fusion: An integrated framework for fusing histopathology and genomic features for cancer diagnosis and prognosis., IEEE Trans. Med. Imaging, 4 (2022), 757–770. https://doi.org/10.1109/TMI.2020.3021387 doi: 10.1109/TMI.2020.3021387
    [31] C. Kandoth, M. D. McLellan, F. Vandin, K. Ye, B. Niu, C. Lu, et al., Mutational landscape and significance across 12 major cancer types, Nature, 502 (2013), 333–339. https://doi.org/10.1038/nature12634 doi: 10.1038/nature12634
    [32] K. Simonyan, A. Zisserman, Very deep convolutional networks for large-scale image recognition, preprint, arXiv: 1409.1556.
    [33] K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2016), 770–778. https://doi.org/10.1109/CVPR.2016.90
    [34] T. N. Kipf, M. Welling, Semi-supervised classification with graph convolutional networks, preprint, arXiv: 1609.02907.
    [35] W. Hamilton, Z. Ying, J. Leskovec, Inductive representation learning on large graphs, Adv. Neural Inform. Proc. Syst., (2017), 30. https://doi.org/10.5555/3294771.3294869
    [36] P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò, Y. Bengio, Graph attention networks, preprint, arXiv: 1710.10903.
    [37] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, et al., Attention is all you need, preprint, arXiv: 1706.03762.
    [38] J. Devlin, M. W. Chang, K. Lee, K. Toutanova, Bert: Pre-training of deep bidirectional transformers for language understanding, preprint, arXiv: 1810.04805.
    [39] P. J. Heagerty, T. Lumley, M. S. Pepe, Time-dependent ROC curves for censored survival data and a diagnostic marker, Biometrics, 56 (2000), 337–344. https://doi.org/10.1111/j.0006-341X.2000.00337.x doi: 10.1111/j.0006-341X.2000.00337.x
  • This article has been cited by:

    1. Nural Yüksel, Burçin Saltık, On inextensible ruled surfaces generated via a curve derived from a curve with constant torsion, 2023, 8, 2473-6988, 11312, 10.3934/math.2023573
    2. Esra Damar, Adjoint curves of special Smarandache curves with respect to Bishop frame, 2024, 9, 2473-6988, 35355, 10.3934/math.20241680
  • Reader Comments
  • © 2023 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(2737) PDF downloads(175) Cited by(6)

Figures and Tables

Figures(4)  /  Tables(5)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog