Postoperative recurrence analysis of breast cancer patients based on clinical serum markers using discriminant methods

Abstract

BACKGROUND:

Breast cancer is a common gynecological malignant tumor and currently its clinical diagnosis mainly depends on methods of iconography and measurement of serum level.

OBJECTIVE:

To analyze correlation between serum index levels and prognosis of patients with breast cancer in one week and six months after operation, and to establish support vector machine (SVM) model to evaluate its effectiveness.

METHODS:

One hundred sixty eight patients diagnosed with breast cancer at Affiliated Cancer Hospital of Zhengzhou University were collected, 46 of which did palindromia while other 122 didn’t six months after operation. Serum CA153, CA125 and CEA levels of different periods in two groups were analyzed from their differences. Through receiver operating characteristic (ROC) curve analysis, their diagnostic threshold values were calculated, at the same time, SVM model was built.

RESULTS:

There was a significant difference between serum index levels of recurrence group and non-recurrence group in one week and six months after operation ( $P<$ 0.05); SVM model was established with an accuracy of 96.67% (29/30), a sensitivity of 90% (9/10) and a specificity of 100% (20/20).

CONCLUSIONS:

Serum CAl53, CEA and CA125 levels after operation have certain instructional significance for prognosis of breast cancer patients, and the established SVM model has high clinical application value.

Keywords

Breast cancer serum marker prognosis diagnostic threshold value SVM

1. Introduction

Breast cancer is a common gynecological malignant tumor and its clinical diagnosis mainly depends on methods of iconography (which is just a specific diagnostic method in our present study) and measurement of serum level currently. According to many studies, serum markers are related to breast cancer, such as carbohydrate antigen (CA153, CA199 and CA125), carcinoembryonic antigen (CEA) and Cytokeratin fragme antigen 21-l (CYFRA21-1), and so on [1, 2] which have been widely used in clinical diagnosis of breast cancer today. Since breast cancer is usually founded at its early phase and there are a few cases having breast cancer metastasis at their first-time diagnosis, generally 3 $\sim$ 10% [3]. Compared with other malignant cancers, breast cancer has a relatively higher cure rate because resection is relatively effective for patients at the early or middle stage [4], but postoperative recurrence and metastasis directly pose a threat to the lives of breast cancer patients [5]. Therefore, it’s clinically significant to prevent and identify postoperative recurrence for patients with breast cancer. Serum markers have been turned out to be effective in differentiating and indicating whether patients have malignant tumors or not, however, whether it is significant to monitor the effect for patients diagnosed with breast cancer and receiving treatment still needs to be studied further.

Starting from three serum markers with high specificity and sensitivity and combining a number of materials of patients with breast cancer and clinical experience, we retrospectively studied patients’ recurrence after mastocarcinoma resection and its correlation to serum marker levels after the operation, aiming to indicate patients’ postoperative effect by the serum marker levels one week after the resection.

2. Materials and methods

2.1 Subjects

One hundred and sixty eight patients diagnosed as breast cancer and receiving resection at Affiliated Cancer Hospital of Zhengzhou University from January 2013 to December 2014 were selected. And 122 of them didn’t relapse within 6 months, with age of 52.1 $\pm$ 8.9 (non-recurrence group), and 46 of them existed recurrence within 6 months, with age of 56.8 $\pm$ 9.7 (recurrence group).

2.2 Collection of specimens

One hundred and sixty eight patients’ serum CA153, CEA and CA125 levels were determined by enzyme-linked immunosorbent assay (ELISA) at three time points, before the surgery, in one week of postoperation and 6 months after the surgery. All subjects were not cases with diseases like other tumors, infections, diabetes and encyesis, etc. and other reasons that might cause a high serum level were excluded as well.

2.3 Methods

2.3.1 Statistical analysis

SPSS 19.0 software was used to analyze the collected materials. One-way ANOVA method was adopted to analyze differences of serum CA153, CEA and CA125 levels at the three points. Then ROC curves of each markers were drawn to compare their sensitivity and specificity and to find the diagnostic threshold value of breast cancer recurrence. Measurement data were processed by $t$ test, and the result of $P<$ 0.05 was considered statistically significant.

2.3.2 Establishment of SVM model

SVM, first put out by Vapnik [6], has many distinctive advantages in solving small samples, nonlinear and high-dimensional pattern recognition and it could be generalized to the learning of function fitting and other machine learning. In accordance with limited sample data, reaching the optimal compromise between model’s recidivity and its learning ability hoped to obtain the best generalization capacity. In this study, all patients were divided into two groups, recurrence group and non-recurrence group in six months after surgery, the recurrence group was assigned as 1, and non-recurrence group as 0; five indexes, level of CA153 one week after the operation, level of CA153 one year after the operation, level of CA125 one week after the operation, level of CEA one week after the operation and level of CEA one year after the operation, were included. Choosing appropriate kernel function parameter and error penalty factor C is very important to the performance of learning machine. Penalty factor C controls punishment for misclassification. If the C value is infinite, all the constraint conditions must be satisfied, which means that all training samples should be accurately classified. But this will bring complexity to the algorithm. Therefore, the selection of C value should be based on practical application to obtain a relatively simple decision function by taking minimal value while classification accuracy is acceptable, besides, the parameters of the kernel function also need to be determined in advance. When establishing SVM model, a parallel grid search method was used to screen SVM parameters. The grid search algorithm gave M and N values to the penalty parameter C and kernel parameter $\gamma$ , respectively, to obtain $M\times N$ ( $C,\gamma$ ) combinations. In the experiment, the data sets were grouped and tested through five-time cross validation [7]. The data were randomly divided into five groups, and each group was regarded as test set in turn with the rest as training sets. The training sets were used to establish the SVM model, and then the test set samples were input to conduct simulation, and then the model was evaluated.

Table 1
Serum marker CA153, CA125 and CEA levels of collected patients

Group CA153 (U/mL) CA125 (U/mL) CEA ( $\mu$ g/L)

Before the One week after 6 months after Before the One week after 6 months after Before the one week after 6 months after

operation the operation the operation operation the operation the operation operation the operation the operation

Recurrent 88.08 $\pm$ 52.18 26.97 $\pm$ 14.75 13.95 $\pm$ 6.90 61.57 $\pm$ 41.84 28.10 $\pm$ 21.06 14.39 $\pm$ 5.35 35.10 $\pm$ 1 1.12 23.34 $\pm$ 4.33 22.69 $\pm$ 10.98

group (46 cases)

Non-recurrent 71.84 $\pm$ 44.80 10.85 $\pm$ 4.02 10.39 $\pm$ 4.58 64.80 $\pm$ 37.39 16.90 $\pm$ 7.45 14.30 $\pm$ 4.35 33.87 $\pm$ 9.20 16.96 $\pm$ 5.69 17.23 $\pm$ 7.75

group (122 cases)

$P$ 0.048 0.001 0.001 0.631 0.000 0.914 0.468 0.002 0.001

Group	CA153 (U/mL)	CA125 (U/mL)	CEA ( $\mu$ g/L)
	Before the	One week after	6 months after	Before the	One week after	6 months after	Before the	one week after	6 months after
	operation	the operation	the operation	operation	the operation	the operation	operation	the operation	the operation
Recurrent	88.08 $\pm$ 52.18	26.97 $\pm$ 14.75	13.95 $\pm$ 6.90	61.57 $\pm$ 41.84	28.10 $\pm$ 21.06	14.39 $\pm$ 5.35	35.10 $\pm$ 1 1.12	23.34 $\pm$ 4.33	22.69 $\pm$ 10.98
group (46 cases)
Non-recurrent	71.84 $\pm$ 44.80	10.85 $\pm$ 4.02	10.39 $\pm$ 4.58	64.80 $\pm$ 37.39	16.90 $\pm$ 7.45	14.30 $\pm$ 4.35	33.87 $\pm$ 9.20	16.96 $\pm$ 5.69	17.23 $\pm$ 7.75
group (122 cases)
$P$	0.048	0.001	0.001	0.631	0.000	0.914	0.468	0.002	0.001

Note: There’s no apparent differences of the three serum marker levels before the operation between recurrent group and non-recurrent group, while they were significantly different one week after the operation ( $P<$ 0.05), and serum CA153 and CEA levels were dramatically different six months after the operation ( $P<$ 0.05).

Table 2

Area under ROC curves of each single markers

Detection result variable	Area	Standard errora	Approximation Sig.b	Approximation 95% CI
				Lower limit	Upper limit
CA153 one week after surgery	0.825	0.052	0.000	0.724	0.926
CA153 six months after surgery	0.664	0.049	0.001	0.569	0.760
CA125 one week after surgery	0.635	0.056	0.007	0.526	0.745
CEA one week after surgery	0.815	0.034	0.000	0.749	0.881
CEA six months after surgery	0.653	0.052	0.002	0.551	0.756

Note: Serum CA153 and CEA levels one week of postoperation had highest sensitivity and specificity to recurrence.

2.4 Evaluation of the model

The SVM model was established and then the two models were assessed from the following aspects.

$\displaystyle\text{Accuracy}=\text{(TP $+$ TN)/(TP $+$ TN $+$ FP $+$ FN)}$ $\displaystyle\quad∼{}\text{$\times$ 100{\%}}$ (1) $\displaystyle\text{Sensitivity}=\text{TP/(TP $+$ FN) $\times$ 100{\%}}$ (2) $\displaystyle\text{Specificity}=\text{TN/(TN $+$ FP) $\times$ 100{\%}}$ (3)

In those equations, true positive number (TP) represents the number of positive samples that were judged correctly in the test set; true negative number (TN) represents the number of negative samples that were judged correctly in the test set; false positive number (FP) represents the number of positive samples that were judged mistakenly in the test set and false negative number (FN) represents the number of negative samples that were judged mistakenly in the test set.

3. Results

Figure 1.

ROC curves of each single markers. Note: AUCs of CA153 one week and six months after the operation were 0.825 and 0.664 respectively ( $P<$ 0.01); the AUC of CA125 one week after the operation was 0.635 ( $P<$ 0.01); AUCs of CEA one week and six months after the operation were 0.815 and 0.653 respectively ( $P<$ 0.01).

3.1 Levels of every serum marker

Levels of the three serum markers (CA153, CEA and CA125) at the three time points (before the operation, one week after the operation and 6 months after the operation) between recurrent group and non-recurrent group were listed in Table 1. According to the table, there’s no apparent differences of the three serum marker levels before the operation in the two groups, while they were significantly different one week after the operation ( $P<$ 0.05), and serum CA153 and CEA levels were dramatically different six months after the operation ( $P<$ 0.05).

3.2 ROC curves of every marker

The diagnostic value of recurrence in breast cancer was evaluated based on AUC which is between 0.5 $\sim$ 1.0 in this study. When AUC $<$ 0.5, it does not conform to the actual situation (rarely exists); when AUC $=$ 0.5, is showed that the diagnostic method has no diagnostic efficacy so the method can not be used to distinguish benign and malignant patients; when AUC is in 0.5 $\sim$ 0.7, it showed that the diagnostic method has relatively poor accuracy; when AUC is within 0.7 $\sim$ 0.9, it meant that the diagnostic method has certain accuracy; and when AUC is over 0.9 indicated that the diagnostic method has relatively high accuracy, the more close to 1, the better diagnosis effect is. Serum CA153, CEA and CA125 levels at the time points of one week and six months after the operation were selected to establish ROC curves. According to Fig. 1 and Table 2, AUCs of CA153 one week and six months after the operation were 0.825 and 0.664 respectively ( $P<$ 0.01); the AUC of CA125 one week after the operation was 0.635 ( $P<$ 0.01); AUCs of CEA one week and six months after the operation were 0.815 and 0.653 respectively ( $P<$ 0.01). All these results showed that serum CA153 and CEA levels one week of postoperation had highest sensitivity and specificity to recurrence. According to ROC curve coordinates in Fig. 1, Youden’s index of every marker was calculated so as to select the maximum one [8]. Thus we could gain the diagnostic threshold value of serum CA153, CA125 and CEA which respectively was 18.0 U/mL, 19.29 U/mL and 18.09 $\mu$ g/L, providing reference for later research.

Figure 2.

SVM training process. Note: The SVM model in this study included five indexes (the levels of CA153 one week and six months after the operation, CA125 level in one week after the operation, and the levels of CA153 one week and six months after the operation).

3.3 The SVM model

The SVM model in this study included five indexes (the levels of CA153 one week and six months after the operation, CA125 level in one week after the operation, and the levels of CA153 one week and six months after the operation). After verification, the ultimately determined optimal parameter ( $C,\gamma$ ) was (12, 1), and the results were showed in Figs 2 and 3. The training process of SVM is in Fig. 2. In Fig. 3, “ $\circ$ ” represents target output (actual test set classification) and “*” suggests the actual stimulation output (predicted test set classification). In accordance with evaluation, the accuracy of the SVM model is 96.67% (29/30), the sensitivity is 90% (9/10), and the specificity is 100% (20/20).

Figure 3.

SVM simulation results. Note: “ $\circ$ ” represents target output (actual test set classification) and “*” suggests the actual stimulation output (predicted test set classification). In accordance with evaluation, the accuracy of the SVM model is 96.67% (29/30), the sensitivity is 90% (9/10), and the specificity is 100% (20/20).

4. Discussion

It has been clinically proved that CA153 and CA125 play role in breast cancer patients’ early diagnosis and postoperative follow-up, and they are sensitive tumor markers too [9, 10]. CA153 is a soluble-form MUC1 mucin, usually located in normal secretory epithelium, and currently it’s also regarded as a breast cancer marker with optimal diagnostic capability [11]. There are studies suggesting that serum CA153 content is related to the metastasis of breast cancer III stage because the level of CA153 in patients with metastatic liver metastasis is higher than that in the control group [12]. CA125 is widely applied to ovarian cancer diagnosis and has high value in breast cancer diagnosis as well [13]. It was reported that the positive rate of CA125 in breast cancer is 20% and is correlated to breast cancer recurrence [14]. But it is still controversial to use CA125 to detect breast cancer [15]. CEA is the first tumor marker applied in clinic and it has been broadly used in tumor detection. And it has been reported that increasing serum CEA level can be considered as a sign for poor prognosis of primary breast cancer [16] and is related to the stage of breast cancer [17]. Meanwhile, serum CEA level is positively correlated to patient’s condition [18]. CA153, CA125 and CEA are generally used in clinical diagnosis of breast cancer and other tumors, and are often taken as prognostic indicators of some cancers [19, 20, 21, 22]. It had been shown that mastocarcinoma resection may have a dynamic role in breast cancer focuses [23], which may change the levels of serum markers through the promotion of tissue growth or other activities [24, 25, 26], providing a basis for further research.

This study retrospectively analyzed whether breast cancer patient’s tumor marker levels before and after surgery has indicating significance for patient’s therapeutic effect in postoperative six months. Study found that serum CA153, CA125 and CEA levels before the operation were higher than the normal level in both recurrent group and non-recurrent group but there’s no difference between the two groups. While there were apparent differences in one week of postoperation, which indicated that the three serum levels in one week of postoperation were significant for patient’s prognosis. Six months after the operation, CA153 and CEA levels existed dramatic difference between the two groups, while CA125 levels didn’t. And according to the ROC curves of serum markers one week after operation, the diagnostic thresholds of CA153, CA125 and CEA for the two groups were 18.0 U/mL, 19.29 U/mL and 18.09 $\mu$ g/L, respectively. Namely, if these three serum levels of patients in one week of postoperation are higher than the threshold value, patient will have the risk of recurrence in six months after the operation.

Although the ROC curves of each serum marker showed a certain diagnostic value, the area under the curve were not more than 0.9, and there were limitations. Therefore, this study included five indexes (the levels of CEA and CA153 one week and six months after the operation, and CA125 level one week after the operation) to establish the SVM model, and the results showed that the accuracy of the model is 96.67% (29/30), the sensitivity is 90.0% (9/10), the specificity is 100.0% (20/20), whose prediction efficiency is good. With the development of SVM model, it has been widely used in medical field [27, 28, 29]. Oliveira et al. [30] using the SVM model to analyze whether there are lumps in breast X-ray found that the accuracy rate of SVM is 98.88%. Juneja et al. [31] through selecting 23 patients (8 sparse and 15 non-sparse) to research the sparsity of fibroglandular tissue of breast cancer patients and they used CT data to establish SVM model whose accuracy rate was up to 91%. Belekar et al. [32] discussed the effect of inhibitors and non-inhibitors in the breast cancer resistant protein (BCRP), through SVM, k-nearest neighbor (k-NN) and artificial neural network (ANN) models, and the accuracy of the three models was 90.8%, 88.3%, 85.2% respectively. In addition, Zhao [33] detected the levels of CEA, CA125 and CA15-3 in breast cancer patients after surgery and according to her founding that CEA had the highest accuracy rate which was 72.9% [33]. In summary, the SVM model has high accuracy, and provides a more accurate diagnostic basis for the future clinical diagnosis of breast cancer. SVM is expected to be an effective and practical tool for prediction of breast cancer recurrence and objective evaluation tool for therapeutic efficacy, and through SVM, recurrence of patients with breast cancer can be detected accurately so that patients themselves can timely understand their own conditions, reducing the mortality of patients with breast cancer.

5. Conclusions

In this study, through analysis of serum level changes

in patients with breast cancer before and after mastocarcinoma resection, our experimental work suggests that serum levels of CA153 and CEA one week and six months after operation, and the levels of CA125, CEA one week after operation have significant indication for prediction of patients’ recurrence, which provided theoretical reference and experimental evidence for clinical workers in prognosis of patients with breast cancer. Moreover, the SVM prognostic model established in this study provided a high clinical value for the prognosis of breast cancer patients.

Prospect

This research mainly discussed the establishment of the predictive model of recurrence in breast cancer, and analyzed how to improve the accuracy of the prediction model, providing a new idea and a new method for the prevention and treatment of recurrence in breast cancer, but it is still insufficient in practical application. The predictive model constructed in this study only included five input variables, and it can be further studied with more input variables; besides, other tumor markers should be taken into consideration in practical application. Once an application platform built, based on the gradual accumulation of data, the input variables of the prediction model could be selected so as to improve the accuracy of the prediction model as much as possible. Because of the specificity of medical data analysis, the input variables are not limited to data variables, and it is significant to find more suitable combination with other medical data including medical imaging, electrocardiogram, etc. in elevating the predictive accuracy of breast cancer’s recurrence and reducing its recurrence rate.

Footnotes

Acknowledgments

I thank all authors who have contributed to this paper for advice and comments and thank Affiliated Cancer Hospital of Zhengzhou University for providing research materials and experimental base. Project supported by the Natural Science Foundation of Henan Province, China (Grant No. 132300410076).

Conflict of interest

The authors declare that they have no competing interests.

References

Gioia

D.D.

Blankenburg

Nagel

Heinemann

and Stieber

, Tumor markers in the early detection of tumor recurrence in breast cancer patients: CA125, CYFRA21-1, HER2 shed antigen, LDH and CRP in combination with CEA and CA 15-3, Clin Chim Acta 461 (2016), 1–7.

Chen

Zheng

Y.H.

Lin

Y.Y.

M.H.

and Chen

Y.S.

, Clinical and prognostic significance of preoperative serum CA153, CEA and TPS levels in patients with primary breast cancer, Chin J Oncol 33 (2011), 842–846.

Siegel

Naishadham

and Jemal

, Cancer statics, CA Cancer J Clinic 63 (2013), 11–30.

Abbott

D.E.

Brouquet

Mittendorf

E.A.

Andreou

Bernstam

F.M.

Valero

Green

M.C.

Kuerer

H.M.

Curley

S.A.

Abdalla

E.K.

Hunt

K.K.

and Vauthey

J.N.

, Resection of liver metastases from breast cancer: estrogen receptor status and response to chemotherapy prior to surgery define outcome, Surg 151 (2012), 710–716.

Gao

Zhao

and Liu

, Clinical value of combined detection of serum tumor markers and whole body bone scan for diagnosis of bone metastases from breast cancer, J Radioimmunol (2007).

Vapnik

V.N.

, The nature of statistical learning theory, Spring (1995), 1–5.

Zhang

, Study on the parameter optimization in support vector machine, Northwestern Univ (2008).

Hirshman

and Hostetter

, Using ROC curves to test models of recognition memory: The relationship between presentation duration and slope, Memory Cognition 28 (2000), 161–166.

Xiao

L.S.

Huang

J.L.

and Qiu

S.X.

, The value of CA153 and CEA in the three female breast cancer, Guangdong Med 34 (2013), 2029–2031.

10.

Kimura

Yoshida

Lin

and Yoshinaga

, Immunohistochemical study of Muc1, Muc2 and human gastric mucin in breast carcinoma: Relationship with prognostic factors, Oncol Rep 8 (2001), 1177–1182.

11.

Ouadagni

Ferroni

Carlini

Mariotti

and Spila

, A re-evaluation of carcinoembryonic antigen (CEA) as a serum marker for breast cancer: A prospective iongitudinal study, Clin Cancer Res 7 (2001), 2357–2362.

12.

Cao

and Wang

L.P.

, Serological diagnosis of liver metastasis in patients with breast cancer, Cancer Biol Med 9 (2012), 57–62.

13.

Said

S.M.

Visscher

D.W.

Nassar

M.D.A.

Ryan

B.A.

and Robert

M.S.

, Flat epithelial atypia and risk of breast cancer: A mayo cohort study, Cancer 121 (2015), 1548–1555.

14.

Y.J.

Rosenzweig

Sereika

S.M.

Linkov

and Brufsky

, Factors associated with mortality after breast cancer metastasis, Cancer Cause Control Ccc (2012), 103–112.

15.

Dong

Hou

Wang

and Yao

, The diagnostic value of 18F-FDG PET/CT in association with serum tumor marker assays in breast cancer recurrence and metastasis, Biomed Res Int (2015), 489021.

16.

Lee

J.S.

Park

J.M.

Cho

J.H.

and Kim

S.I.

, Elevated levels of preoperative CA153 and CEA serum levels have independently poor prognostic significance in breast cancer, Ann Oncol 24 (2013), 1225–1231.

17.

Geng

Liang

M.M.

X.B.

and Zhao

W.Y.

, Association of CA15-3 and CEA with clinicopathological parameters in patients with metastatic breast cancer, Mol Clin Oncol 3 (2015), 232–236.

18.

Wang

Qin

Zhang

Zhao

and Liang

, Nipple discharge of CA153, CA125, CEA and TSGF as a new biomarker panel for breast cancer, Int J Mol Sci 15 (2014), 9546–9565.

19.

Fersching

D.M.

Nagel

Siegele

Salat

and Heinemann

, Apoptosis-related biomarkers sFAS, MIF, ICAM-1 and PAI-1 in serum of breast cancer patients undergoing neoadjuvant chemotherapy, Anticancer Res 32 (2012), 2047–2058.

20.

Kucera

Cerna

Narsanska

Svobodova

and Strakova

, Growth factors and breast tumors, comparison of selected growth factors with traditional tumor markers, Anticancer Res 31 (2011), 4653–4656.

21.

Kong

Wang

Liu

Chen

and Yang

, Cytokeratin19-2g2, a novel fragment of cytokeratin19 in serum, indicating a more invasive behavior and worse prognosis in breast cancer patients, PLoS One 8 (2013), e57092.

22.

Zhang

S.J.

Qian

H.L.

Jiao

S.C.

and Liu

Z.F.

, Expression and significance of ER, PR, VEGF, CA15-3, CA125 and CEA in judging the prognosis of breast cancer, Asian Pac J Cancer Prev 14 (2013), 3937–3940.

23.

Badwe

Parmar

Hawaldar

Nair

and Kaushik

, Surgical removal of primary tumor and axillary lymph nodes in women with metastatic breast cancer at first presentation: A randomized controlled trial, Cancer Res 2 (2013), s2–02.

24.

Pedersen

A.C.

Sorensen

P.D.

and Jacobsen

E.H.

, Sensitivity of CA15-3, CEA and serum HER2 in the early detection of recurrence of breast cancer, Chem Lab Med 13 (2013), 1–9.

25.

Wang

J.P.

Hong

X.T.

and Sheng

J.H.

, Combined analysis of CA153, CA125 and CEA in diagnosis of breast cancer, Henan Med Res (2007).

26.

Zhao

P.Y.

, Application of peripheral blood CEA, CA153 and CA125 in the clinical screening of breast cancer, Chin Matern Child Health Care 16 (2015), 2520–2522.

27.

Polat

and Güneş

, Detection of ECG arrhythmia using a differential expert system approach based on principal component analysis and least square support vector machine, Appl Math Comput 186 (2007), 898–906.

28.

Dreiseitl

Ohno-Machado

Kittler

Vinterbo

and Billhardt

, A comparison of machine learning methods for the diagnosis of pigmented skin lesions, J Biomed Inform 34 (2001), 28–36.

29.

Fonseca

E.S.

Guido

R.C.

Scalassara

P.R.

Maciel

C.D.

and Pereira

J.C.

, Wavelet time-frequency analysis and least squares support vector machines for the identification of voice disorders, Computer Biol Med 37 (2007), 571–578.

30.

Oliveira

F.S.S.D.

Filho

A.O.D.C.

Silva

A.C.

Paiva

A.C.D.

and Gattass

, Classification of breast regions as mass and non-mass based on digital mammograms using taxonomic indexes and SVM, Computer Biol Med 57C (2014), 42–53.

31.

Juneja

Evans

Windridge

and Harris

, Classification of fibroglandular tissue distribution in the breast based on radiotherapy planning CT, Bmc Med Imag 16 (2016), 1–8.

32.

Belekar

Lingineni

and Garg

, Classification of breast cancer resistant protein (BCRP) inhibitors and non-inhibitors using machine learning approaches, Comb Chem High T Scr (2015), 42–53.

33.

Zhao

, Value of CEA, CA125 and CA153 in monitoring the recurrence and metastasis of breast cancer after operation, Dalian Med Univ (2014).