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

Lung cancer diagnosis from computed tomography scans using convolutional neural network architecture with Mavage pooling technique

  • Received: 01 September 2024 Revised: 23 October 2024 Accepted: 08 November 2024 Published: 09 January 2025
  • Background 

    Lung cancer is a deadly disease. An early diagnosis can significantly improve the patient survival and quality of life. One potential solution is using deep learning (DL) algorithms to automate the diagnosis using patient computed tomography (CT) scans. However, the limited availability of training data and the computational complexity of existing algorithms, as well as their reliance on high-performance systems, limit the potential of DL algorithms. To improve early lung cancer diagnoses, this study proposes a low-cost convolutional neural network (CNN) that uses a Mavage pooling technique to diagnose lung cancers.

    Methods 

    The DL-based model uses five convolution layers with two residual connections and Mavage pooling layers. We trained the CNN using two publicly available datasets comprised of the IQ_OTH/NCCD dataset and the chest CT scan dataset. Additionally, we integrated the Mavage pooling in the AlexNet, ResNet-50, and GoogLeNet architectures to analyze the datasets. We evaluated the performance of the models based on accuracy and the area under the receiver operating characteristic curve (AUROC).

    Results 

    The CNN model achieved a 99.70% accuracy and a 99.66% AUROC when the scans were classified as either cancerous or non-cancerous. It achieved a 90.24% accuracy and a 94.63% AUROC when the scans were classified as containing either normal, benign, or malignant nodules. It achieved a 95.56% accuracy and a 99.37% AUROC when lung cancers were classified. Additionally, the results indicated that the diagnostic abilities of AlexNet, ResNet-50, and GoogLeNet were improved with the introduction of the Mavage pooling technique.

    Conclusions 

    This study shows that a low-cost CNN can effectively diagnose lung cancers from patient CT scans. Utilizing Mavage pooling technique significantly improves the CNN diagnostic capabilities.

    Citation: Ayomide Abe, Mpumelelo Nyathi, Akintunde Okunade. Lung cancer diagnosis from computed tomography scans using convolutional neural network architecture with Mavage pooling technique[J]. AIMS Medical Science, 2025, 12(1): 13-27. doi: 10.3934/medsci.2025002

    Related Papers:

    [1] Panpan Jia, Jizhu Nan, Yongsheng Ma . Separating invariants for certain representations of the elementary Abelian $ p $-groups of rank two. AIMS Mathematics, 2024, 9(9): 25603-25618. doi: 10.3934/math.20241250
    [2] Hengbin Zhang . Automorphism group of the commuting graph of $ 2\times 2 $ matrix ring over $ \mathbb{Z}_{p^{s}} $. AIMS Mathematics, 2021, 6(11): 12650-12659. doi: 10.3934/math.2021729
    [3] Huaqing Gong, Shilin Yang . The representation ring of a non-pointed bialgebra. AIMS Mathematics, 2025, 10(3): 5110-5123. doi: 10.3934/math.2025234
    [4] Hicham Saber, Tariq Alraqad, Rashid Abu-Dawwas . On graded $ s $-prime submodules. AIMS Mathematics, 2021, 6(3): 2510-2524. doi: 10.3934/math.2021152
    [5] Fatma Zehra Uzekmek, Elif Segah Oztas, Mehmet Ozen . $ (\theta_i, \lambda) $-constacyclic codes and DNA codes over $ \mathbb{Z}_{4}+u\mathbb{Z}_{4}+u^{2}\mathbb{Z}_{4} $. AIMS Mathematics, 2024, 9(10): 27908-27929. doi: 10.3934/math.20241355
    [6] Pengcheng Ji, Jialei Chen, Fengxia Gao . Projective class ring of a restricted quantum group $ \overline{U}_{q}(\mathfrak{sl}^{*}_2) $. AIMS Mathematics, 2023, 8(9): 19933-19949. doi: 10.3934/math.20231016
    [7] Dong Su, Shilin Yang . Automorphism groups of representation rings of the weak Sweedler Hopf algebras. AIMS Mathematics, 2022, 7(2): 2318-2330. doi: 10.3934/math.2022131
    [8] Junyong Zhao . On the number of unit solutions of cubic congruence modulo $ n $. AIMS Mathematics, 2021, 6(12): 13515-13524. doi: 10.3934/math.2021784
    [9] Fareeha Jamal, Muhammad Imran . Distance spectrum of some zero divisor graphs. AIMS Mathematics, 2024, 9(9): 23979-23996. doi: 10.3934/math.20241166
    [10] Nikken Prima Puspita, Indah Emilia Wijayanti . Bi-clean and clean Hopf modules. AIMS Mathematics, 2022, 7(10): 18784-18792. doi: 10.3934/math.20221033
  • Background 

    Lung cancer is a deadly disease. An early diagnosis can significantly improve the patient survival and quality of life. One potential solution is using deep learning (DL) algorithms to automate the diagnosis using patient computed tomography (CT) scans. However, the limited availability of training data and the computational complexity of existing algorithms, as well as their reliance on high-performance systems, limit the potential of DL algorithms. To improve early lung cancer diagnoses, this study proposes a low-cost convolutional neural network (CNN) that uses a Mavage pooling technique to diagnose lung cancers.

    Methods 

    The DL-based model uses five convolution layers with two residual connections and Mavage pooling layers. We trained the CNN using two publicly available datasets comprised of the IQ_OTH/NCCD dataset and the chest CT scan dataset. Additionally, we integrated the Mavage pooling in the AlexNet, ResNet-50, and GoogLeNet architectures to analyze the datasets. We evaluated the performance of the models based on accuracy and the area under the receiver operating characteristic curve (AUROC).

    Results 

    The CNN model achieved a 99.70% accuracy and a 99.66% AUROC when the scans were classified as either cancerous or non-cancerous. It achieved a 90.24% accuracy and a 94.63% AUROC when the scans were classified as containing either normal, benign, or malignant nodules. It achieved a 95.56% accuracy and a 99.37% AUROC when lung cancers were classified. Additionally, the results indicated that the diagnostic abilities of AlexNet, ResNet-50, and GoogLeNet were improved with the introduction of the Mavage pooling technique.

    Conclusions 

    This study shows that a low-cost CNN can effectively diagnose lung cancers from patient CT scans. Utilizing Mavage pooling technique significantly improves the CNN diagnostic capabilities.



    Tensor categories should be thought as counterparts of rings in the world of categories [1,2,3], i.e., the categorification of groups and rings [4,5,6]. They are ubiquitous in noncommutative algebra and representation theory. Tensor categories were introduced by Bénabou [7] in 1963 and Lane [8] as "categories with multiplication", and its related theories are now widely used in many fields of mathematics, including algebraic geometry [9], algebraic topology [10], number theory [11], operator algebraic theory [12], etc. The theory of tensor categories is also seen as a development following from that of Hopf algebras and their representation theory [13,14]. As an important invariant in the theory of tensor categories, the concept of a Z+-ring can be traced back to Lusztig's work [15] in 1987. Later, in [16,17], the notion of a Z+-module over a Z+-ring was introduced. Module categories over multitensor categories were first considered in [4,18], and then the notion of an indecomposable module category was introduced in [17]. As a categorification of irreducible Z+-modules, it is interesting to classify indecomposable exact module categories over a given tensor category. In this process, it is often necessary to first classify all irreducible Z+-modules over the Grothendieck ring of a given tensor category.

    Typical examples of Z+-rings are the representation rings of Hopf algebras [19,20,21,22,23,24]. Another example is the Grothendieck rings of tensor categories [25,26,27,28]. It is natural to consider the classification of all irreducible Z+-modules over them. For example, Etingof and Khovanov classified irreducible Z+-modules over the group ring ZG, and showed that indecomposable Z+-modules over the representation ring of SU(2), under certain conditions, correspond to affine and infinite Dynkin diagrams [16]. Also, there is a lot of related research in the context of near-group fusion categories. For instance, Tambara and Yamagami classified semisimple tensor categories with fusion rules of self-duality for finite abelian groups. Evans, Gannon, and Izumi have contributed to the classification of the near-group C-categories [29,30]. Yuan et al. [31] studied irreducible Z+-modules of the near-group fusion ring K(Z3,3) and so on.

    In this paper, we explore the problem of classifying irreducible based modules of rank up to 5 over the complex representation ring r(S4), and then discuss their categorification. Furthermore, we overcome the technical difficulty of solving a series of non-negative integer equations using MATLAB. In contrast with the representation ring of S3, r(Sn) is no longer a near-group fusion ring when n>3, and the classification of irreducible Z+-modules over general r(Sn) seems to be a hopeless task. Hence, our paper attempts to classify irreducible based modules for the non-near-group fusion ring r(S4). In fact, the fusion rule of r(Sn) is already a highly nontrivial open problem in combinatorics, namely counting the multiplicities of irreducible components of the tensor product of any two irreducible complex representations of Sn (so called the Kronecker coefficients).

    The paper is organized as follows. In Section 2, we recall some basic definitions and propositions. In Section 3, we discuss the irreducible based modules of rank up to 5 over r(S4) and give the classification of all these based modules (Propositions 3.1–3.5). In Section 4, we first show that any Z+-module over the representation ring r(G) of a finite group G categorified by a module category over the representation category Rep(G) should be a based module (Theorem 4.2), and then determine which irreducible based modules over r(S4) can be categorified (Theorem 4.12).

    Throughout this paper, all rings are assumed to be associative with unit 1. Let Z+ denote the set of nonnegative integers. First, we recall the definitions of Z+-rings and Z+-modules. For more details about these concepts, readers can refer to [17,32].

    In this section, we first recall some definitions, and then we exhibit a class of identities for transposed Poisson n-Lie algebras.

    Definition 2.1. Let A be a ring which is free as a Z-module:

    (i) A Z+-basis of A is a basis

    B={bi}iI,

    such that

    bibj=kIckijbk,

    where ckijZ+.

    (ii) A Z+-ring is a ring with a fixed Z+-basis and with unit 1 being a non-negative linear combination of the basis elements.

    (iii) A Z+-ring is unital if the unit 1 is one of its basis elements.

    Definition 2.2. Let A be a Z+-ring with basis {bi}iI. A Z+-module over A is an A-module M with a fixed Z-basis {ml}lL such that all the structure constants akil, defined by the equality

    biml=kakilmk

    are non-negative integers.

    A Z+-module has the following equivalent definition referring to [32, Section 3.4].

    Definition 2.3. Let A be a Z+-ring with basis {bi}iI. A Z+-module M over A means an assignment where each basis bi in A is in one-to-one correspondence with a non-negative integer square matrix Mi such that M forms a representation of A:

    MiMj=kIckijMk, i,j,kI,

    where the unit of A corresponds to the identity matrix. The rank of a Z+-module M is equal to the order of the matrix Mi.

    Definition 2.4. (i) Two Z+-modules M1,M2 over A with bases {m1i}iL1,{m2j}jL2 are equivalent if and only if there exists a bijection ϕ: L1L2 such that the induced Z-linear map ˜ϕ of abelian groups M1,M2 defined by

    ˜ϕ(m1i)=m2ϕ(i)

    is an isomorphism of A-modules. In other words, for aA, let aM1 and aM2 be the matrices with respect to the bases {m1i}iL1 and {m2j}jL2, respectively. Then, two Z+-modules M1, M2 of rank n are equivalent if and only if there exists an n×n permutation matrix P such that

    aM2=PaM1P1, aA.

    (ii) The direct sum of two Z+-modules M1,M2 over A is the module M1M2 over A whose basis is the union of the bases of M1 and M2.

    (iii) A Z+-module M over A is indecomposable if it is not equivalent to a nontrivial direct sum of Z+-modules.

    (iv) A Z+-submodule of a Z+-module M over A with basis {ml}lL is a subset JL such that the abelian subgroup of M generated by {mj}jJ is an A-submodule.

    (v) A Z+-module M over A is irreducible if any Z+-submodule of M is 0 or M. In other words, the Z-span of any proper subset of the basis of M is not an A-submodule.

    Let A be a Z+-ring with basis {bi}iI, and let I0 be the set of iI such that bi occurs in the decomposition of 1. Let τ: AZ denote the group homomorphism defined by

    τ(bi)={1, if  iI0,0, if  iI0.

    Definition 2.5. A Z+-ring with basis {bi}iI is called a based ring if there exists an involution ii of I such that the induced map

    a=iIaibia=iIaibi,aiZ,

    is an anti-involution of the ring A, and

    τ(bibj)={1,if i=j,0,if ij.

    A fusion ring is a unital based ring of finite rank.

    Definition 2.6. A based module over a based ring A with basis {bi}iI is a Z+-module M with basis {ml}lL over A such that

    akil=alik,

    where akil are defined as in Definition 2.2.

    Let A be a unital Z+-ring of finite rank with basis {bi}iI, and let M be a Z+-module over A with Z-basis {ml}lL. Take

    b=iIbi.

    For any fixed ml0, the Z+-submodule of M generated by ml0 is the Z-span of {mk}kY, where the set Y consists of kL such that mk is a summand of bml0. Also, we need the following facts.

    Proposition 2.1. [32, Proposition 3.4.6] Let A be a based ring of finite rank over Z. Then there exist only finitely many irreducible Z+-modules over A.

    Proposition 2.2. [17, Lemma 2.1] Let M be a based module over a based ring A. If M is decomposable as a Z+-module over A, then M is irreducible as a Z+-module over A.

    As a result, any Z+-module of finite rank over a fusion ring is completely reducible, and then only irreducible Z+-modules need to be classified.

    In general, the rank of an irreducible Z+-module over a fusion ring A may be larger than the rank of A; e.g., A=r(D5) for the dihedral group D5 ([33, Remark 1]). In this paper, we explore which irreducible based modules over r(S4) can be categorified by indecomposable exact module categories over the representation category Rep(S4). Since all these module categories are of rank not greater than 5, we only deal with based modules of rank up to 5 correspondingly.

    In this section, we will classify the irreducible based modules over the complex representation ring r(S4) of S4 up to equivalence. r(S4) is a commutative fusion ring having a Z+-basis {1,Vψ,Vρ1,Vρ2,Vρ3} with the fusion rule.

    V2ψ=1,VψVρ1=Vρ1,VψVρ2=Vρ3,V2ρ1=1+Vψ+Vρ1,Vρ1Vρ2=Vρ2+Vρ3,V2ρ2=1+Vρ1+Vρ2+Vρ3, (3.1)

    where 1, Vψ, and Vρ1 denote the trivial representation, sign representation, and 2-dimensional irreducible representation, respectively, while Vρ2 stands for the 3-dimensional standard representation and Vρ3 denotes its conjugate representation. Then we have the following Table 1.

    Table 1.  The complex character table of S4.
    (1) (12) (123) (1234) (12)(34)
    χ1 1 1 1 1 1
    χψ 1 1 1 1 1
    χρ1 2 0 1 0 2
    χρ2 3 1 0 1 1
    χρ3 3 1 0 1 1

     | Show Table
    DownLoad: CSV

    Let M be a based module of r(S4) with the basis {ml}lL. Let T,Q,U, and W be the matrices representing the action of Vψ,Vρ1,Vρ2, and Vρ3 on M respectively. They are all symmetric matrices with nonnegative integer entries by Definition 2.6. Let E be the identity matrix. By the fusion rule of r(S4), we have

    T2=E, (3.2)
    TQ=QT=Q, (3.3)
    TU=UT=W, (3.4)
    Q2=E+T+Q, (3.5)
    QU=U+TU, (3.6)
    U2=E+Q+U+TU. (3.7)

    In particular, since T2=E and T has nonnegative integer entries, we know that T is a symmetric permutation matrix.

    Convention 3.1. Let Pn be the group of n×n permutation matrices. Since there is naturally a group isomorphism between Sn and Pn, we will use the cycle notation of permutations to represent permutation matrices.

    We define a Z+-module M1,1 of rank 1 over r(S4) by letting

    Vψ1,Vρ12,Vρ23,Vρ33. (3.8)

    Proposition 3.1. Any irreducible based module of rank 1 over r(S4) is equivalent to M1,1.

    Proof. Note that any integral fusion ring A has the unique character FPdim: AZ, which takes non-negative values on the Z+-basis, so there exists a unique Z+-module M of rank 1 over it. Clearly, such M is a based module. Now this argument is available for the situation A=r(S4).

    Next, we consider irreducible based modules of rank 2,3. According to the fusion rule of r(S4) given in (3.1), it is sufficient to only list the representation matrices of Vψ, Vρ1, and Vρ2 acting on them. For simplicity, we choose to present our result for the cases of small rank 2 and 3 directly, and then analyze the cases of higher rank 4 and 5 with details.

    Proposition 3.2. Let M be an irreducible based module of rank 2 over r(S4). Then M is equivalent to one of the based modules M2,i,1i3, listed in Table 2.

    Table 2.  Inequivalent irreducible based modules of rank 2 over r(S4).
    Vψ Vρ1 Vρ2
    M2,1 (1001) (2002) (1221)
    M2,2 (0110) (1111) (2112)
    M2,3 (0110) (1111) (1221)

     | Show Table
    DownLoad: CSV

    Proposition 3.3. Let M be an irreducible based module of rank 3 over r(S4). Then M is equivalent to one of the based modules M3,i,1i3, listed in Table 3.

    Table 3.  Inequivalent irreducible based modules of rank 3 over r(S4).
    Vψ Vρ1 Vρ2
    M3,1 (100010001) (011101110) (111111111)
    M3,2 (010100001) (001001111) (101011112)
    M3,3 (010100001) (001001111) (011101112)

     | Show Table
    DownLoad: CSV

    Proposition 3.4. Let M be an irreducible based module of rank 4 over r(S4). Then M is equivalent to one of the based modules M4,i,1i7, listed in Table 4.

    Table 4.  Inequivalent irreducible based modules of rank 4 over r(S4).
    Vψ Vρ1 Vρ2
    M4,1 (1000010000100001) (0011020010011010) (0100121101000100)
    M4,2 (1000010000100001) (2000020000200002) (0111101111011110)
    M4,3 (0100100000100001) (1100110000200002) (1011011111011110)
    M4,4 (0100100000100001) (1100110000200002) (0111101111011110)
    M4,5 (0100100000010010) (1100110000110011) (0111101111011110)
    M4,6 (0100100000010010) (1100110000110011) (0111101111101101)
    M4,7 (0100100000010010) (1100110000110011) (1011011111101101)

     | Show Table
    DownLoad: CSV

    Proof. Before giving its detailed proof, we provide the following proof outline first.

    (i) The symmetric group S4 has 3 conjugacy classes of permutations of order 2, so there are 3 representatives for matrix T up to conjugation as follows:

    T1=E4,T2=(12),T3=(12)(34).

    Consequently, we can take T=Tr for some r=1,2,3 as the representation matrix of Vψ for the based module M up to equivalence.

    (ii) Use MATLAB to search all solutions of the representation matrices Q and U in the group of nonnegative integer matrix Eqs (3.3)–(3.7) by constraint satisfaction.

    (iii) Distinguish all conjugacy classes of tuples (T,Q,U) without simultaneous block decomposition. They correspond to the equivalence classes of irreducible based modules over r(S4).

    Proof. Let M be a based module of rank 4 over r(S4), with the action of r(S4) on it given by

    VψT,Vρ1Q=(aij)1i,j4,Vρ2U=(bij)1i,j4,Vρ3W=TU,

    where aij=aji, bij=bji.

    The symmetric group S4 has two conjugacy classes of permutations of order 2. One conjugacy class of 6 permutations includes (12), and the other one of 3 permutations includes (12)(34). As previously seen, T is the unit or an element of order 2 in P4, so we have 10 candidates for T, and each of them is conjugate to one of the following 3 matrices:

    T1=E4,T2=(12),T3=(12)(34).

    Hence, for the based module M determined by the pair (T,Q,U), there exists a 4×4 permutation matrix P such that

    T=PTP1

    is one of the above Tr's (1r3). Correspondingly, let

    Q=PQP1,U=PUP1.

    Then we get a based module M determined by the pair (T,Q,U) and equivalent to M as based modules by Definition 2.4 (ⅰ). So, we have reduced the proof to the situation when T=Tr.

    Case 1. T=T1=E4.

    Since Q satisfies Eq (3.5), we obtain the following system of integer equations:

    {a211+a212+a213+a214=2+a11,a11a12+a12a22+a13a23+a14a24=a12,a11a13+a12a23+a13a33+a14a34=a13,a11a14+a12a24+a13a34+a14a44=a14,a212+a222+a223+a224=2+a22,a12a13+a22a23+a23a33+a24a34=a23,a12a14+a22a24+a23a34+a24a44=a24,a213+a223+a233+a234=2+a33,a13a14+a23a24+a33a34+a34a44=a34,a214+a224+a234+a244=2+a44.

    We use MATLAB to figure out all the solutions of Q as follows:

    Q1=(0011020010011010),Q2=(0101100100201100),Q3=(0110101011000002),Q4=(2000001101010110),Q5=(2000020000200002).

    Next, we calculate U after taking Q as one Qk (1k5).

    Case 1.1. Q=Q1.

    Since U satisfies Eq (3.6), we get

    b12=b23=b24,b11=b13=b14=b33=b34=b44.

    Then, by Eq (3.7), we have

    {3b211+b212=2b11+1,3b11b12+b12b22=2b12,3b212+b222=2b22+3.

    The solutions of U given by MATLAB are as follows:

    U1=(0100121101000100),U2=(1011030010111011).

    It is easy to check that the based module determined by (T1,Q1,U1) is an irreducible based module denoted as M4,1, while the based module determined by (T1,Q1,U2) is reducible.

    Note that there exists a permutation matrix P=(14)(23) such that

    PQ1P1=Q2.

    Let

    U1=PU1P1.

    There is an irreducible based module N determined by the pair (T1,Q2,U1) and equivalent to M4,1 by Definition 2.4 (ⅰ). Conversely, any irreducible based module with representation matrices T1 and Q2 is equivalent to M4,1. The same analysis tells us that irreducible based modules with representation matrices T1 and Q3 (or Q4) are also equivalent to M4,1.

    Case 1.2. Q=Q5.

    Since U satisfies Eqs (3.6) and (3.7), we get a system of integer equations as follows:

    {b211+b212+b213+b214=2b11+3,b11b12+b12b22+b13b23+b14b24=2b12,b11b13+b12b23+b13b33+b14b34=2b13,b11b14+b12b24+b13b34+b14b44=2b14,b212+b222+b223+b224=2b22+3,b12b13+b22b23+b23b33+b24b34=2b23,b12b14+b22b24+b23b34+b24b44=2b24,b213+b223+b233+b234=2b33+3,b13b14+b23b24+b33b34+b34b44=2b34,b214+b224+b234+b244=2b44+3.

    Thus, the solutions of U by MATLAB are as follows:

    U1=(0111101111011110),U2=(1002012002102001),U3=(1020010220100201),U4=(1200210000120021),U5=(1020030020100003),U6=(1002030000302001),U7=(1200210000300003),U8=(3000030000120021),U9=(3000012002100003),U10=(3000010200300201),U11=(3000030000300003).

    Since T1 and Q5 are diagonal and the solutions Ut (2t11) are block diagonal with at least two blocks, only the based module determined by (T1,Q5,U1) is irreducible, denoted as M4,2.

    Case 2. T=T2=(12).

    Since Q satisfies Eq (3.3), we get

    Q=(a11a11a13a14a11a11a13a14a13a13a33a34a14a14a34a44).

    Since Q also satisfies Eq (3.5), we have the following system of integer equations:

    {2a211+a213+a214=a11+1,2a11a13+a13a33+a14a34=a13,2a11a14+a13a34+a14a44=a14,2a213+a233+a234=a33+2,2a13a14+a33a34+a34a44=a34,2a214+a234+a244=a44+2.

    Hence, the solutions of Q by MATLAB are as follows:

    Q1=(0001000100201101),Q2=(0010001011100002),Q3=(1100110000200002).

    Since U satisfies Eq (3.4), we get

    U=(b11b12b13b14b12b11b13b14b13b13b33b34b14b14b34b44).

    Next, we calculate U after taking Q as one Qk (1k3).

    Case 2.1. Q=Q1.

    Since U satisfies Eqs (3.6) and (3.7), the solutions of U given by MATLAB are as follows:

    U1=(0101100100301102),U2=(1001010100301102).

    Since T2, Q1 and all the solutions Ut for t=1,2 are block diagonal with at least two blocks, the based modules determined by each pair (T2,Q1,Ut) are reducible.

    Note that there exists a permutation matrix P=(12)(34) such that

    PQ1P1=Q2.

    Let

    Ut=PUtP1.

    Then each based module Nt determined by the pair (T2,Q2,Ut) is reducible. Namely, any based module with representation matrices T2 and Q2 is reducible.

    Case 2.2. Q=Q3.

    Since U satisfies Eqs (3.6) and (3.7), we have

    U1=(0111101111011110),U2=(1011011111011110),U3=(1200210000120021),U4=(1200210000300003),U5=(2100120000120021),U6=(2100120000300003).

    Since T2, Q3 and the solutions Us (3s6) are block diagonal with at least two blocks, only the based module determined by (T2,Q3,U1) and (T2,Q3,U2) are irreducible, denoted as M4,3 and M4,4, respectively. It is easy to check that M4,3 and M4,4 are inequivalent based modules.

    Case 3. T=T3=(12)(34).

    Since Q satisfies Eq (3.3), we get

    Q=(a11a11a13a13a11a11a13a13a13a13a33a33a13a13a33a33).

    Then, by Eq (3.5), we have the following system of integer equations:

    {2a211+2a213=a11+1,2a11a13+2a13a33=a13,2a213+2a233=a33+1.

    Q has the following unique solution:

    Q1=(1100110000110011).

    Since U satisfies Eq (3.4), we get

    U=(b11b12b13b14b12b11b14b13b13b14b33b34b14b13b34b33).

    Since U also satisfies Eqs (3.6) and (3.7), we obtain the solutions of U by MATLAB as follows:

    U1=(0111101111011110),U2=(0111101111101101),U3=(1011011111101101),U4=(1011011111011110),U5=(1200210000120021),U6=(1200210000210012),U7=(2100120000120021),U8=(2100120000210012).

    Clearly, T3, Q1 and the solutions Us are block diagonal with at least two blocks, but the based module determined by the pair (T3,Q1,Ut) is irreducible, denoted as M4,s, where 5s8,1t4. Define the Z-module isomorphism ϕ: M4,6M4,8 by

    ϕ(v11)=v24,ϕ(v12)=v23,ϕ(v13)=v22,ϕ(v14)=v21.

    It is easy to see that M4,6 is equivalent to M4,8 as based modules over r(S4) under ϕ. Then, we can check that {M4,s}5s7 are inequivalent irreducible based modules.

    Finally, we construct two based modules M5,i (i=1,2) over r(S4) with the actions of r(S4) on them presented in Table 5.

    Table 5.  Inequivalent irreducible based modules of rank 5 over r(S4).
    Vψ Vρ1 Vρ2
    M5,1 (0100010000000100010000001) (0000100001001100011011001) (0001000100011111011100110)
    M5,2 (0100010000000100010000001) (1100011000001100011000002) (0001100101010011000111111)

     | Show Table
    DownLoad: CSV

    Proposition 3.5. Let M be an irreducible based module of rank 5 over r(S4). Then M is equivalent to one of the based modules M5,i(i=1,2), listed in Table 5.

    Proof. Let M be a based module of rank 5 over r(S4), with the action of r(S4) on it given by

    VψT,Vρ1Q=(aij)1i,j5,Vρ2U=(bij)1i,j5,Vρ3W=TU,

    where aij=aji, bij=bji.

    First, by a similar argument applied in the case of rank 4, we only need to deal with one of the following 3 cases for T:

    T1=E5,T2=(12),T3=(12)(34).

    Case 1. T=T1=E5.

    There are 11 solutions of Q satisfying Eq (3.5), but only two conjugacy classes by permutation matrices with their representatives given as follows:

    Q1=(0110010100110000002000002),Q2=(2000002000002000002000002).

    Next, we calculate U after taking Q as one Qk (k=1,2).

    Case 1.1. Q=Q1.

    There are 4 solutions of U satisfying Eqs (3.6) and (3.7) as follows:

    U1=(0001000010000101112000003),  U2=(0000100001000010003011102),  U3=(1110011100111000003000003),  U4=(1110011100111000001200021).

    Case 1.2. Q=Q2.

    There are 31 solutions of U satisfying Eqs (3.6) and (3.7), but only 4 conjugacy classes by permutation matrices and their representatives as follows:

    U1=(0111010110110101110000003),  U2=(1002001200021002001000003),  U3=(1200021000003000003000003),  U4=(3000003000003000003000003).

    Each pair (T1,Qk,Ur) above determines a based module, but is not irreducible for any 1r4.

    Case 2. T=T2=(12).

    There are 5 solutions of Q satisfying Eq (3.5), but only 3 conjugacy classes with the following representatives:

    Q1=(0000100001002000002011001),Q2=(1100011000000110010100110),Q3=(1100011000002000002000002).

    Next, we calculate U after choosing Q.

    Case 2.1. Q=Q1.

    There are 4 solutions of U satisfying Eqs (3.6) and (3.7) as follows:

    U1=(0100110001001200021011002),  U2=(0100110001003000003011002),  U3=(1000101001001200021011002),  U4=(1000101001003000003011002).

    Case 2.2. Q=Q2.

    There are 2 solutions of U satisfying Eqs (3.6) and (3.7) as follows:

    U1=(1200021000001110011100111),  U2=(2100012000001110011100111).

    Case 2.3. Q=Q3.

    There are 14 solutions of U satisfying Eqs (3.6) and (3.7), but only 6 conjugacy classes by permutation matrices with their representatives given as follows:

    U1=(0111010110110101110000003),U2=(1011001110110101110000003),U3=(1200021000001200021000003),U4=(2100012000001200021000003),U5=(1200021000003000003000003),U6=(2100012000003000003000003).

    Through analysis, all based modules derived from Case 2 are reducible.

    Case 3. T=T3=(12)(34).

    There are 3 solutions of Q satisfying Eq (3.5) as follows:

    Q1=(0000100001001100011011001),Q2=(1100011000001100011000002),Q3=(1100011000000010000100111).

    Next, we calculate U after fixing Q.

    Case 3.1. Q=Q1.

    There are 6 solutions of U satisfying Eqs (3.6) and (3.7) as follows:

    U1=(0001000100011111011100110),U2=(0010000010101110111100110),U3=(0100110001001200021011002),U4=(0100110001002100012011002),U5=(1000101001001200021011002),U6=(1000101001002100012011002).

    Each pair (T3,Q1,Ur) (1r6) above determines a based module, but only the based modules with representation matrices U1 and U2 are irreducible. Such two irreducible based modules are denoted by M5,1 and M5,1, with the corresponding Z-basis {vk1,vk2,vk3,vk4,vk5} for k=1,2, respectively. Define the Z-module isomorphism ϕ: M5,1M5,1 by

    ϕ(v1s)=v2s,ϕ(v13)=v24,ϕ(v14)=v23,s=1,2,5.

    Then it is easy to see that M5,1 is equivalent to M5,1 as based modules over r(S4) under ϕ.

    Case 3.2. Q=Q2.

    There are 10 solutions of U satisfying Eqs (3.6) and (3.7), but only 7 conjugacy classes with their representatives given as follows:

    U1=(0001100101010011000111111),  U2=(0111010110110101110000003),  U3=(0111010110111001101000003),  U4=(1011001110111001101000003),
    U5=(1200021000001200021000003),  U6=(1200021000002100012000003),  U7=(2100012000002100012000003).

    Each pair (T3,Q2,Ut) (2t7) above determines a based module, but only the based module with representation matrix U1 is irreducible. We denote it by M5,2.

    Also, the based modules obtained by taking Q=Q3 are equivalent to the based module M5,1 found in Case 3.1.

    In this section, we will apply the knowledge of module categories over the complex representation category of a finite group to find which based modules over r(S4) can be categorified by module categories over the representation category Rep(S4) of S4. For the details about module categories over tensor categories, see, e.g., [32, Section 7].

    First, we recall the required result for the upcoming discussion. For any finite group G, the second cohomology group H2(G,C) is known to be a finite abelian group called the Schur multiplier and classifies central extensions of G. The notion of a universal central extension of a finite group was first investigated by Schur in [34].

    Let Rep(G,α) denote the semisimple abelian category of projective representations of G with the multiplier αZ2(G,C). Equivalently, Rep(G,α) is the representation category of the twisted group algebra CGα of G with multiplication

    gαh=α(g,h)gh,g,hG.

    In particular,

    Rep(G,α)=Rep(G),

    when taking α=1.

    Let αZ2(G,C) represent an element of order d in H2(G,C). Define

    Repα(G)=d1j=0Rep(G,αj).

    According to the result in [35], we know that Repα(G) becomes a fusion category with the tensor product of two projective representations in Rep(G,αi) and Rep(G,αj) respectively lying in Rep(G,αi+j), and the dual object in Rep(G,αi) lying in Rep(G,αdi). Correspondingly, we have the fusion ring

    rα(G)=d1j=0r(G,αj). (4.1)

    Now let H be a subgroup of G and αZ2(H,C). The category Rep(H,α) is a module category over Rep(G) by applying the restriction functor ResGH: Rep(G)Rep(H).

    Theorem 4.1. [17, Theorem 3.2] The indecomposable exact module categories over the representation category Rep(G) are of the form Rep(H,α) and are classified by conjugacy classes of pairs (H,[α]).

    Consequently, by [32, Proposition 7.7.2], we know the following:

    Proposition 4.1. The Grothendieck group

    r(H,α)=Gr(Rep(H,α))

    is an irreducible Z+-module over r(G).

    Next, we show that any Z+-module over the complex representation ring r(G) of a finite group G categorified in this way is a based module.

    Theorem 4.2. Let G be a finite group, H a subgroup of G, and αZ2(H,C). The Z+-module r(H,α) over r(G) is a based module.

    Proof. Let {ψi}iI be the Z+-basis of r(G). Take rα(H) defined in Eq (4.1) as a Z+-module over r(G) with the Z-basis {χk}kJ such that

    ψi.χk=lalikχl,alikZ+.

    On the other hand, we write the fusion rule of the fusion ring rα(H) as follows:

    χiχj=sk=1nkijχk,nkijZ+.

    Since the number nkij is invariant under cyclic permutations of i,j,k, we have

    nkij=njki=njik.

    By the restriction rule, we interpret r(G) as a subring of rα(H) and write down

    ψi=jrijχj,rijZ+.

    Then

    ψi.χk=jrijχjχk=j,lrijnljkχl.

    By comparing the coefficients, we see that

    alik=jrijnljk=jrijnkjl=jrijnkjl=jrijnkjl=akil,

    so rα(H) is a based module over r(G), and r(H,α) is clearly a based submodule of rα(H). Equivalently, any Z+-module over r(G) categorified by a module category Rep(H,α) over Rep(G) must be a based module.

    By Theorem 4.2, we only need to focus on those inequivalent irreducible based modules Mi,j over r(S4) collected in Section 3, each of which is possibly categorified by a module category Rep(H,α) for some H<S4 and αZ2(H,C).

    All the non-isomorphic subgroups of the symmetric group S4 are as follows:

    (i) The symmetric group S3;

    (ii) The cyclic groups Zi,1i4;

    (iii) The Klein 4-group K4;

    (iv) The alternating group A4;

    (v) The dihedral group D4;

    (vi) The symmetric group S4 itself.

    Correspondingly, the Schur multipliers we consider here are given as follows (see e.g., [36]):

    H2(Zn,C)H2(S3,C)0, n1, H2(K4,C)H2(D4,C)H2(A4,C)H2(S4,C)Z2.

    As a result, we only need to consider the following two situations:

    (1) Module category Rep(H) for any subgroup H<S4;

    (2) Module category Rep(H,α) for any subgroup H<S4 and nontrivial twist αZ2(H,C).

    (ⅰ) First, we consider the representation category Rep(S3) as a module category over Rep(S4).

    Theorem 4.3. r(S3)=Gr(Rep(S3)) is an irreducible based module over r(S4)=Gr(Rep(S4)) equivalent to the based module M3,2 in Table 3.

    Proof. According to the branching rule of symmetric groups (see e.g., [37, Theorem 2.8.3]), we have the following restriction rules:

    ResS4S3(1)=1,ResS4S3(Vψ)=χ,ResS4S3(Vρ1)=V,ResS4S3(Vρ2)=1+V,ResS4S3(Vρ3)=χ+V,

    where χ and V denote the sign representation and the standard representation in Rep(S3), respectively. Hence, we get the representation matrices of basis elements of r(S4) acting on r(S3) as follows:

    1E3,Vψ(010100001),Vρ1(001001111),Vρ2(101011112),Vρ3(011101112).

    We see that r(S3) is an irreducible based module M3,2 according to Table 3. In other words, the based module M3,2 can be categorified by the module category Rep(S3) over Rep(S4).

    Remark 4.1. Since the roles of the standard representation and its dual in r(S4) are symmetric, we can exchange the notations Vρ2 and Vρ3 for them to get the following restriction rules instead:

    ResS4S3(Vρ2)=χ+V,ResS4S3(Vρ3)=1+V.

    Therefore, we get another action of r(S4) on r(S3) such that r(S3) is an irreducible based module over r(S4) equivalent to the based module M3,3 according to Table 3. In other words, the based module M3,3 can also be categorified by the module category Rep(S3) over Rep(S4).

    (ⅱ) Second, we consider Rep(Z4) as a module category over Rep(S4).

    Theorem 4.4. r(Z4)=Gr(Rep(Z4)) is an irreducible based module over r(S4) equivalent to the based module M4,5 in Table 4.

    Proof. Let

    Z4={1,g,g2,g3}

    be the cyclic group of order 4, with four non-isomorphic 1-dimensional irreducible representations denoted by Ui,i=0,1,2,3. Let U0=1 represent the trivial representation, and

    χU1(g)=1,χU2(g)=1,χU3(g)=1.

    On the other hand, we consider Z4 as the subgroup of S4 generated by g=(1234). Then, by the character table of S4 (Table 1), we have

    χψ(gi)=(1)i,χρ1(gi)=1+(1)i,χρ2(gi)=(1)i+(1)i+(1)i,χρ3(gi)=1+(1)i+(1)i.

    So, the restriction rule of r(S4) on r(Z4) is given as follows:

    ResS4Z4(1)=1,ResS4Z4(Vψ)=U2,ResS4Z4(Vρ1)=1+U2,ResS4Z4(Vρ2)=U1+U2+U3,ResS4Z4(Vρ3)=1+U1+U3.

    Then, we get the representation matrices of basis elements of r(S4) acting on r(Z4) as follows:

    1E4, Vψ(0010000110000100), Vρ1(1010010110100101), Vρ2(0111101111011110), Vρ3(1101111001111011).

    Let {wi}1i4 be the stated Z-basis of M4,5, and define a Z-linear map φ: M4,5r(Z4) by

    φ(w1)=U3,φ(w2)=U1,φ(w3)=U2,φ(w4)=1.

    Then, it is easy to check that φ is an isomorphism of r(S4)-modules, so M4,5 is equivalent to r(Z4) as based modules by Definition 2.4 (ⅰ). In other words, the based module M4,5 can be categorified by the module category Rep(Z4) over Rep(S4).

    Remark 4.2. By the same argument as in Remark 4.1, Vρ2 and Vρ3 can be required to satisfy the following restriction rules instead:

    ResS4Z4(Vρ2)=1+U1+U3,ResS4Z4(Vρ3)=U1+U2+U3.

    Therefore, we get another action of r(S4) on r(Z4) such that r(Z4) is an irreducible based module over r(S4) equivalent to the based module M4,7 according to Table 4. In other words, the based module M4,7 can also be categorified by the module category Rep(Z4) over Rep(S4).

    Also, one can similarly check that the module category Rep(Z2) over Rep(S4) categorifies the based modules M2,2 and M2,3, while Rep(Z3) over Rep(S4) categorifies the based module M3,1.

    (ⅲ) Now we consider Rep(K4) as a module category over Rep(S4).

    Theorem 4.5. r(K4)=Gr(Rep(K4)) is an irreducible based module over r(S4) equivalent to the based module M4,7 in Table 4.

    Proof. We consider K4 as the subgroup of S4 generated by (12) and (34), and it has four non-isomorphic 1-dimensional irreducible representations Y0=1 and Y1,Y2,Y3 such that

    χY1((12))=1,χY1((34))=1;χY2((12))=1,χY2((34))=1;χY3((12))=1,χY3((34))=1.

    On the other hand, by the character table of S4 (Table 1), we have

    χψ((12))=χψ((34))=1,χψ((12)(34))=1;χρ1((12))=χρ1((34))=0,χρ1((12)(34))=2;χρ2((12))=χρ2((34))=1,χρ2((12)(34))=1;χρ3((12))=χρ3((34))=1,χρ3((12)(34))=1.

    So, we have the following restriction rules:

    ResS4K4(1)=1,ResS4K4(Vψ)=Y3,ResS4K4(Vρ1)=1+Y3,ResS4K4(Vρ2)=1+Y1+Y2,ResS4K4(Vρ3)=Y1+Y2+Y3.

    Then we get the representation matrices of basis elements of r(S4) acting on r(K4) as follows:

    1E4,Vψ(0001001001001000),Vρ1(1001011001101001),Vρ2(1110110110110111),Vρ3(0111101111011110).

    Let {wi}1i4 be the stated Z-basis of M4,7 listed in Table 4. Then

    w1Y2,w2Y1,w31,w4Y3,

    defines an equivalence of Z+-modules between M4,7 and r(K4). In other words, the irreducible based module M4,7 can be categorified by the module category Rep(K4) over Rep(S4).

    Remark 4.3. In a manner analogous to the argument in Remark 4.1, it follows that the irreducible based module M4,5 can also be categorified by the module category Rep(K4) over Rep(S4).

    (ⅳ) We consider Rep(A4) as a module category over Rep(S4).

    Theorem 4.6. r(A4)=Gr(Rep(A4)) is an irreducible based module over r(S4) equivalent to the based module M4,1 in Table 4.

    Proof. We know that A4 has three non-isomorphic 1-dimensional irreducible representations and one 3-dimensional irreducible representation, denoted by N0,N1,N2, and N3, respectively, where N0=1 represents the trivial representation, and

    χN1((123))=ω,χN1((12)(34))=1;χN2((123))=ω2,χN2((12)(34))=1;χN3((123))=χN3((132))=0, χN3((12)(34))=1, ω=1+32.

    On the other hand, the character table of S4 (Table 1) tells us that

    χψ((123))=1,χψ((12)(34))=1;χρ1((123))=1,χρ1((12)(34))=2;χρ2((123))=0,χρ2((12)(34))=1;χρ3((123))=0,χρ3((12)(34))=1.

    So, we have the following restriction rules:

    ResS4A4(1)=ResS4A4(Vψ)=1,ResS4A4(Vρ1)=N1+N2,ResS4A4(Vρ2)=ResS4A4(Vρ3)=N3.

    Hence, we get the representation matrices of basis elements of r(S4) acting on r(A4) as follows:

    1E4,VψE4,Vρ1(0110101011000002),Vρ2,Vρ3(0001000100011112).

    Then, r(A4) is an irreducible based module over r(S4) equivalent to M4,1 listed in Table 4. In other words, the irreducible based module M4,1 can be categorified by the module category Rep(A4) over Rep(S4).

    (ⅴ) Next, we consider Rep(D4) as a module category over Rep(S4).

    Theorem 4.7. r(D4)=Gr(Rep(D4)) is an irreducible based module over r(S4) equivalent to the based module M5,2 in Table 5.

    Proof. The dihedral group

    D4=r,s|r4=s2=(rs)2=1

    has four 1-dimensional irreducible representations and one 2-dimensional irreducible representation up to isomorphism, denoted by W0,W1,W2,W3, and W4, respectively. Let W0=1 stand for the trivial representation, and

    χW1(r)=1,χW1(s)=1;χW2(r)=1,χW2(s)=1;χW3(r)=1,χW3(s)=1;χW4(r)=χW4(s)=χW4(rs)=0.

    On the other hand, we consider D4 as the subgroup of S4 by taking r=(1234) and s=(12)(34). Then rs=(13). By the character table of S4 (Table 1), we have

    χψ((1234))=1,χψ((12)(34))=1,χψ((13))=1;χρ1((1234))=0,χρ1((12)(34))=2,χρ1((13))=0;χρ2((1234))=1,χρ2((12)(34))=1,χρ2((13))=1;χρ3((1234))=1,χρ3((12)(34))=1,χρ3((13))=1.

    So, we have the following restriction rules:

    ResS4D4(1)=1,ResS4D4(Vψ)=W2,ResS4D4(Vρ1)=1+W2,ResS4D4(Vρ2)=W3+W4,ResS4D4(Vρ3)=W1+W4.

    Then we get the representation matrices of basis elements of r(S4) acting on r(D4) as follows:

    1E5,Vψ(0010000010100000100000001),Vρ1(1010001010101000101000002),Vρ2(0001100101010011000111111),Vρ3(0100110001000110010111111).

    Then r(D4) is an irreducible based module over r(S4) equivalent to M5,2 listed in Table 5. In other words, the irreducible based module M5,2 can be categorified by the module category Rep(D4) over Rep(S4).

    (ⅵ) Finally, we consider Rep(S4) as a module category over itself.

    Theorem 4.8. The regular Z+-module r(S4) over itself is equivalent to the irreducible based module M5,1 in Table 5.

    Proof. Let r(S4) be the regular Z+-module over itself with the Z-basis {1,Vψ,Vρ1,Vρ2,Vρ3}, and the action of r(S4) on it is given as follows:

    1E5,Vψ(0100010000001000000100010),Vρ1(0010000100111000001100011),Vρ2(0001000001000111011101111),Vρ3(0000100010000110111110111).

    Then, the regular Z+-module r(S4) over itself is equivalent to M5,1 listed in Table 5. In other words, the irreducible based module M5,1 over r(S4) can be categorified by the module category Rep(S4) over itself.

    Remark 4.4. Following the argument presented in Remark 4.1, if we exchange the notations Vρ2 and Vρ3 with their restriction rules given in the proof of Theorems 4.7 and 4.8, we see that r(D4) and r(S4) are still equivalent to M5,2 and M5,1, respectively.

    Lastly, we consider the module category Rep(H,α) over Rep(S4), where H is a subgroup of S4 with α representing the unique nontrivial cohomological class in H2(H,C). All non-isomorphic irreducible projective representations of H with the multiplier α form a Z-basis of r(H,α), whose cardinality is the number of α-regular conjugacy classes by [38, Theorem 6.1.1].

    First, we consider the twisted group algebra of K4. There is only one irreducible projective representation with respect to α up to isomorphism, see, e.g., [39, Appendix D.1]. Hence, r(K4,α) is a based module of rank 1 over r(S4) equivalent to M1,1 defined in (3.8). Namely, the based module M1,1 can also be categorified by Rep(K4,α).

    Second, we consider the twisted group algebra of D4.

    Theorem 4.9. r(D4,α)=Gr(Rep(D4,α)) is an irreducible based module over r(S4) equivalent to the based module M2,3 in Table 2.

    Proof. Let

    D4=r,s|r4=s2=(rs)2=1.

    Let αZ2(D4,C) be the 2-cocycle defined by

    α(risj,risj)=(1)ji.

    Here, i,i{0,1,2,3},j,j{0,1}. As shown in [40, Section 3.7], this is a unitary 2-cocycle representing the unique non-trivial cohomological class in H2(D4,C). According to [35, Section 3], there exist two (2-dimensional) non-isomorphic irreducible projective representations of D4 with respect to α, which are given by

    πl:D4GL2(C),risjAilBj,

    where

    Al=((1)l00(1)1l),B=(0110),l=1,2.

    Also, for irreducible representations W0W4 of D4 mentioned in the proof of Theorem 4.7, we have

    W0πl=W1πl=πl,W2πl=W3πl=π3l,W4πl=π1+π2.

    Next, using the previous restriction rule of r(S4) on r(D4), we get the representation matrices of basis elements of r(S4) acting on r(D4,α) as follows:

    1E2,Vψ(0110),Vρ1(1111),Vρ2(1221),Vρ3(2112).

    Then, r(D4,α) is an irreducible based module over r(S4) equivalent to M2,3 listed in Table 2. In other words, the irreducible based module M2,3 can be categorified by the module category Rep(D4,α) over Rep(S4).

    Remark 4.5. As discussed in Remark 4.1, it follows that the irreducible based module M2,2 can also be categorified by the module category Rep(D4,α) over Rep(S4).

    Next, we consider the twisted group algebras of A4 and S4. By [38, Theorem 6.1.1], A4 has three (2-dimensional) non-isomorphic irreducible projective representations, denoted as Vγ1,Vγ2, and Vγ3, respectively. Similarly, S4 has two (2-dimensional) non-isomorphic irreducible projective representations Vξ1,Vξ2, and one (4-dimensional) irreducible projective representation Vξ3. We give the character table for projective representations of A4 and S4 in Tables 6 and 7, respectively, where primes are used to differentiate between the two classes splitting from a single conjugacy class of A4 in its double cover ˜A4, and the same applies to S4; subscripts distinguish between the two classes splitting from the conjugacy classes (31) and (31) in the double cover of , respectively. For more details, see [41, Section 4].

    Table 6.  The character table for irreducible projective representations of .

     | Show Table
    DownLoad: CSV
    Table 7.  The character table for irreducible projective representations of .

     | Show Table
    DownLoad: CSV

    In Table 6, we denote

    Then we have the following theorems.

    Theorem 4.10. is an irreducible based module over equivalent to the based module in Table 3.

    Proof. For the irreducible representations , and of mentioned in the proof of Theorem 4.6, we obtain the following tensor product rule in by computing the values of products of characters:

    where , . Next, by combining this with the previous restriction rule of on , we obtain

    Then is an irreducible based module over equivalent to listed in Table 3. In other words, the irreducible based module can be categorified by .

    Theorem 4.11. is an irreducible based module over equivalent to the based module in Table 3.

    Proof. Let be a nontrivial 2-cocycle in (see e.g., [42, Section 3.2.4]). By checking products of characters, we get the following tensor product rule in :

    where , . Thus, we get

    Then is an irreducible based module over equivalent to listed in Table 3. In other words, the irreducible based module over can be categorified by .

    In summary, we have the following classification theorem.

    Theorem 4.12. The inequivalent irreducible based modules over are

    among which

    can be categorified by module categories over ; see Table 8.

    Table 8.  Inequivalent irreducible based modules over .
    Categorification
    Rank 1 ,
    Rank 2 No
    ,
    ,
    Rank 3 ,
    ,
    Rank 4
    No
    No
    No
    No
    Rank 5

     | Show Table
    DownLoad: CSV

    The analysis in this paper shows that the classification of the irreducible based modules of rank up to 5 over the complex representation ring . We also showed that any -modules over the representation ring categorified by a module category over the representation category must be a based module. At the end, we present the categorification of based modules over by module categories over the complex representation category of , using projective representations of specific subgroups of . We expect that the studies developed here will be helpful in investigations of the structures of module categories over fusion categories. Our future study will focus on the existence of any irreducible based module of rank over and classifying irreducible -modules over , especially for high-rank cases. Also, some other small finite groups may be interesting to consider, e.g., the dihedral group .

    Wenxia Wu: Writing-original draft and editing, conceptualization, software, methodology; Yunnan Li: Topic selection, writing-review and editing, funding acquisition, methodology, supervision. All authors have read and approved the final version of the manuscript for publication.

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

    We would like to thank Zhiqiang Yu for helpful discussion. This work is supported by Guangdong Basic and Applied Basic Research Foundation (2022A1515010357).

    The authors declare that there are no conflicts of interest.


    Acknowledgments



    We wish to express our sincere gratitude and appreciation to the DSI–CSIR Inter-bursary Support (IBS) Programme for financial support. The data used in this study are publicly available and comes from benchmark data and do not raise any ethical issues. The data that supports the findings of the study is available upon reasonable request from the corresponding author.

    Conflict of interest



    The authors declare no conflict of interest.

    [1] Cruz CSD, Tanoue LT, Matthay RA (2011) Lung cancer: epidemiology, etiology, and prevention. Clin Chest Med 32: 605-644. https://doi.org/10.1016/j.ccm.2011.09.001
    [2] Bray F, Ferlay J, Soerjomataram I, et al. (2018) Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 68: 394-424. https://doi.org/10.3322/caac.21492
    [3] Schabath MB, Cote ML (2019) Cancer progress and priorities: lung cancer. Cancer Epidemiol Biomarkers Prev 28: 1563-1579. https://doi.org/10.1158/1055-9965.EPI-19-0221
    [4] Yoon SM, Shaikh T, Hallman M (2017) Therapeutic management options for stage III non-small cell lung cancer. World J Clin Oncol 8: 1-20. https://doi.org/10.5306/wjco.v8.i1.1
    [5] Hoffman RM, Atallah RP, Struble RD, et al. (2020) Lung cancer screening with low-dose CT: a meta-analysis. J Gen Intern Med 35: 3015-3025. https://doi.org/10.1007/s11606-020-05951-7
    [6] Silvestri GA, Goldman L, Tanner NT, et al. (2023) Outcomes from more than 1 million people screened for lung cancer with low-dose CT imaging. Chest 164: 241-251. https://doi.org/10.1016/j.chest.2023.02.003
    [7] Shin HJ, Kim MS, Kho BG, et al. (2020) Delayed diagnosis of lung cancer due to misdiagnosis as worsening of sarcoidosis: a case report. BMC Pulm Med 20: 1-4. https://doi.org/10.1186/s12890-020-1105-2
    [8] Del Ciello A, Franchi P, Contegiacomo A, et al. (2017) Missed lung cancer: when, where, and why?. Diagn Interv Radiol 23: 118-126. https://doi.org/10.5152/dir.2016.16187
    [9] Friedland B (2009) Medicolegal issues related to cone beam CT. Semin Orthod 15: 77-84. https://doi.org/10.1053/j.sodo.2008.09.010
    [10] Krupinski EA, Berbaum KS, Caldwell RT, et al. (2010) Long radiology workdays reduce detection and accommodation accuracy. J Am Coll Radiol 7: 698-704. https://doi.org/10.1016/j.jacr.2010.03.004
    [11] Abujudeh HH, Boland GW, Kaewlai R, et al. (2010) Abdominal and pelvic computed tomography (CT) interpretation: discrepancy rates among experienced radiologists. Eur Radiol 20: 1952-1957. https://doi.org/10.1007/s00330-010-1763-1
    [12] Jacobsen MM, Silverstein SC, Quinn M, et al. (2017) Timeliness of access to lung cancer diagnosis and treatment: a scoping literature review. Lung cancer 112: 156-164. https://doi.org/10.1016/j.lungcan.2017.08.011
    [13] Sathyakumar K, Munoz M, Singh J, et al. (2020) Automated lung cancer detection using artificial intelligence (AI) deep convolutional neural networks: a narrative literature review. Cureus 12: e10017. https://doi.org/10.7759/cureus.10017
    [14] Huang S, Yang J, Shen N, et al. (2023) Artificial intelligence in lung cancer diagnosis and prognosis: current application and future perspective. Semin Cancer Biol 89: 30-37. https://doi.org/10.1016/j.semcancer.2023.01.006
    [15] Ardila D, Kiraly AP, Bharadwaj S, et al. (2019) End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nat Med 25: 954-961. https://doi.org/10.1038/s41591-019-0447-x
    [16] Alrahhal MS, Alqhtani E (2021) Deep learning-based system for detection of lung cancer using fusion of features. Int J Comput Sci Mobile Comput 10: 57-67. https://doi.org/10.47760/ijcsmc.2021.v10i02.009
    [17] Huang X, Shan J, Vaidya V (2017) Lung nodule detection in CT using 3D convolutional neural networks. 2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017) : 379-383. https://doi.org/10.1109/ISBI.2017.7950542
    [18] Guo Y, Song Q, Jiang M, et al. (2021) Histological subtypes classification of lung cancers on CT images using 3D deep learning and radiomics. Acad Radiol 28: e258-e266. https://doi.org/10.1016/j.acra.2020.06.010
    [19] Cui X, Zheng S, Heuvelmans MA, et al. (2022) Performance of a deep learning-based lung nodule detection system as an alternative reader in a Chinese lung cancer screening program. Eur J Radiol 146: 110068. https://doi.org/10.1016/j.ejrad.2021.110068
    [20] Dunn B, Pierobon M, Wei Q (2023) Automated classification of lung cancer subtypes using deep learning and CT-scan based radiomic analysis. Bioengineering 10: 690. https://doi.org/10.3390/bioengineering10060690
    [21] de Margerie-Mellon C, Chassagnon G (2023) Artificial intelligence: a critical review of applications for lung nodule and lung cancer. Diagn Interv Imaging 104: 11-17. https://doi.org/10.1016/j.diii.2022.11.007
    [22] Ahmed SF, Alam MSB, Hassan M, et al. (2023) Deep learning modelling techniques: current progress, applications, advantages, and challenges. Artif Intell Rev 56: 13521-13617. https://doi.org/10.1007/s10462-023-10466-8
    [23] Zhang J, Xia Y, Zeng H, et al. (2018) NODULe: combining constrained multi-scale LoG filters with densely dilated 3D deep convolutional neural network for pulmonary nodule detection. Neurocomputing 317: 159-167. https://doi.org/10.1016/j.neucom.2018.08.022
    [24] AL-Huseiny MS, Sajit AS (2021) Transfer learning with GoogLeNet for detection of lung cancer. Indones J Electr Eng Comput Sci 22: 1078-1086. https://doi.org/10.11591/ijeecs.v22.i2.pp1078-1086
    [25] Sakshiwala, Singh MP (2023) A new framework for multi-scale CNN-based malignancy classification of pulmonary lung nodules. J Ambient Intell Human Comput 14: 4675-4683. https://doi.org/10.1007/s12652-022-04368-w
    [26] Mamun M, Mahmud MI, Meherin M, et al. (2023) LCDctCNN: lung cancer diagnosis of CT scan images using CNN based model. 2023 10th International Conference on Signal Processing and Integrated Networks (SPIN) : 205-212. https://doi.org/10.1109/SPIN57001.2023.10116075
    [27] Krizhevsky A, Sutskever I, Hinton GE (2012) Imagenet classification with deep convolutional neural networks. Adv Neural Inf Process Syst 25.
    [28] Hua KL, Hsu CH, Hidayati SC, et al. (2015) Computer-aided classification of lung nodules on computed tomography images via deep learning technique. Onco Targets Ther 8: 2015-2022. https://doi.org/10.2147/OTT.S80733
    [29] Ronneberger O, Fischer P, Brox T (2015) U-net: Convolutional networks for biomedical image segmentation. In: Navab, N., Hornegger, J., Wells, W., Frangi, A. Eds. Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015. Springer International Publishing, 234-241. https://doi.org/10.1007/978-3-319-24574-4_28
    [30] Gulshan V, Peng L, Coram M, et al. (2016) Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 316: 2402-2410. https://doi.org/10.1001/jama.2016.17216
    [31] Kaul V, Enslin S, Gross SA (2020) History of artificial intelligence in medicine. Gastrointest Endosc 92: 807-812. https://doi.org/10.1016/j.gie.2020.06.040
    [32] Esteva A, Kuprel B, Novoa RA, et al. (2017) Dermatologist-level classification of skin cancer with deep neural networks. Nature 542: 115-118. https://doi.org/10.1038/nature21056
    [33] Saouli R, Akil M, Kachouri R (2018) Fully automatic brain tumor segmentation using end-to-end incremental deep neural networks in MRI images. Comput Methods Programs Biomed 166: 39-49. https://doi.org/10.1016/j.cmpb.2018.09.007
    [34] Lorenzo PR, Nalepa J, Bobek-Billewicz B, et al. (2019) Segmenting brain tumors from FLAIR MRI using fully convolutional neural networks. Comput Methods Programs Biomed 176: 135-148. https://doi.org/10.1016/j.cmpb.2019.05.006
    [35] Moitra D, Mandal RK (2019) Automated AJCC staging of non-small cell lung cancer (NSCLC) using deep convolutional neural network (CNN) and recurrent neural network (RNN). Health Inf Sci Syst 7: 1-12. https://doi.org/10.1007/s13755-019-0077-1
    [36] Chaunzwa TL, Hosny A, Xu Y, et al. (2021) Deep learning classification of lung cancer histology using CT images. Sci Rep 11: 1-12. https://doi.org/10.1038/s41598-021-84630-x
    [37] Anari S, Tataei Sarshar N, Mahjoori N, et al. (2022) Review of deep learning approaches for thyroid cancer diagnosis. Math Probl Eng 2022: 5052435. https://doi.org/10.1155/2022/5052435
    [38] Painuli D, Bhardwaj S, Köse U (2022) Recent advancement in cancer diagnosis using machine learning and deep learning techniques: a comprehensive review. Comput Biol Med 146: 105580. https://doi.org/10.1016/j.compbiomed.2022.105580
    [39] Hosseini SH, Monsefi R, Shadroo S (2024) Deep learning applications for lung cancer diagnosis: a systematic review. Multimed Tools Appl 83: 14305-14335. https://doi.org/10.1007/s11042-023-16046-w
    [40] Aamir M, Rahman Z, Abro WA, et al. (2023) Brain tumor classification utilizing deep features derived from high-quality regions in MRI images. Biomed Signal Proces 85: 104988. https://doi.org/10.1016/j.bspc.2023.104988
    [41] Sun M, Song Z, Jiang X, et al. (2017) Learning pooling for convolutional neural network. Neurocomputing 224: 96-104. https://doi.org/10.1016/j.neucom.2016.10.049
    [42] Kumar RL, Kakarla J, Isunuri BV, et al. (2021) Multi-class brain tumor classification using residual network and global average pooling. Multimed Tools Appl 80: 13429-13438. https://doi.org/10.1007/s11042-020-10335-4
    [43] Zafar A, Aamir M, Mohd Nawi N, et al. (2022) A comparison of pooling methods for convolutional neural networks. Appl Sci 12: 8643. https://doi.org/10.3390/app12178643
    [44] Nirthika R, Manivannan S, Ramanan A, et al. (2022) Pooling in convolutional neural networks for medical image analysis: a survey and an empirical study. Neural Comput Appl 34: 5321-5347. https://doi.org/10.1007/s00521-022-06953-8
    [45] Xiong W, Zhang L, Du B, et al. (2017) Combining local and global: rich and robust feature pooling for visual recognition. Pattern Recogn 62: 225-235. https://doi.org/10.1016/j.patcog.2016.08.006
    [46] Dogan Y (2023) A new global pooling method for deep neural networks: Global average of top-k max-pooling. Trait Signal 40: 577-587. https://doi.org/10.18280/ts.400216
    [47] Nirthika R, Manivannan S, Ramanan A, et al. (2022) Pooling in convolutional neural networks for medical image analysis: a survey and an empirical study. Neural Comput Applic 34: 5321-5347. https://doi.org/10.1007/s00521-022-06953-8
    [48] Abuqaddom I, Mahafzah BA, Faris H (2021) Oriented stochastic loss descent algorithm to train very deep multi-layer neural networks without vanishing gradients. Knowledge-Based Systems 230: 107391. https://doi.org/10.1016/j.knosys.2021.107391
    [49] LeCun Y, Bottou L, Bengio Y, et al. (1998) Gradient-based learning applied to document recognition. Proc IEEE 86: 2278-2324. https://doi.org/10.1109/5.726791
    [50] Sabri N, Hamed HNA, Ibrahim Z, et al. (2020) A comparison between average and max-pooling in convolutional neural network for scoliosis classification. Int J 9. https://doi.org/10.30534/ijatcse/2020/9791.42020
    [51] Hyun J, Seong H, Kim E (2021) Universal pooling–a new pooling method for convolutional neural networks. Expert Syst Appl 180: 115084. https://doi.org/10.1016/j.eswa.2021.115084
    [52] Özdemir C (2023) Avg-topk: a new pooling method for convolutional neural networks. Expert Syst Appl 223: 119892. https://doi.org/10.1016/j.eswa.2023.119892
    [53] Zhou Q, Qu Z, Cao C (2021) Mixed pooling and richer attention feature fusion for crack detection. Pattern Recogn Lett 145: 96-102. https://doi.org/10.1016/j.patrec.2021.02.005
    [54] Hou Q, Zhang L, Cheng MM, et al. (2020) Strip pooling: Rethinking spatial pooling for scene parsing. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition : 4003-4012.
    [55] Kareem HF, AL-Husieny MS, Mohsen FY, et al. (2021) Evaluation of SVM performance in the detection of lung cancer in marked CT scan dataset. Indones J Electr Eng Comput Sci 21: 1731-1738. https://doi.org/10.11591/ijeecs.v21.i3.pp1731-1738
    [56] Alyasriy H (2020) The IQ-OTHNCCD lung cancer dataset. Mendeley Data . Available from: https://data.mendeley.com/datasets/bhmdr45bh2/1. Retrieved June 15, 2024
    [57] Hany M  (2020) Chest CT-scan images dataset. Available from: https://www.kaggle.com/datasets/mohamedhanyyy/chest-ctscan-images. Retrieved June 15, 2024
    [58] He K, Zhang X, Ren S, et al. (2016) Deep residual learning for image recognition. Proceedings of the IEEE conference on computer vision and pattern recognition : 770-778.
    [59] Szegedy C, Liu W, Jia Y, et al. (2015) Going deeper with convolutions. Proceedings of the IEEE conference on computer vision and pattern recognition : 1-9.
    [60] Paszke A, Gross S, Massa F, et al. (2019) Pytorch: an imperative style, high-performance deep learning library. Adv Neural Inf Process Syst 32.
  • Reader Comments
  • © 2025 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(563) PDF downloads(50) Cited by(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog