Learning Algorithms for Cervical Cancer Detection

Marco Antonio Acevedo

, María Elena Acevedo

and Sandra Dinora Orantes

Instituto Politécnico Nacional, ESIME Zacatenco, Av. Lindavista, Mexico City, Mexico

Instituto Politécnico Nacional, CIC, Mexico City , Mexico

Keywords: Artificial Intelligence, Machine Learning, K-NN Algorithm, Artificial Neural Networks, Diagnosis, Cervical

Cancer.

Abstract: Cervical cancer begins in the cervix, the lower part of the uterus (womb) that opens into the upper part of the

vagina. Worldwide, cervical cancer is the third most common type of cancer in women. This type of cancer

can be detected visually with pap-smear images. A secondary alternative is to evaluate the relevant risk factors

to see the possible formation of cervical cancer; these factors are recorded in a questionary. In this paper, the

dataset from the questionary is analysed with two machine learning algorithms: K-NN and Multi-Layer

Perceptron. We proposed the architectures and the parameters which achieve the best results. Two validation

algorithms were applied: K-Fold Cross-Validation and Hold Out (80-20). The results from the machine

learning algorithms were: 100% with 1-NN and Multi-Layer Perceptron together with K-Fold Cross-

Validation and 97% with 1-NN and 98% when Multi-Layer Perceptron was applied, and the validation

algorithm was Hold-Out.

1 INTRODUCTION

Cervical cancer (INM, 2021) (CC) is the growth,

development, and disorderly and uncontrolled

multiplication of cervical cells, which is the lower

part of the uterus (womb) that empties into the upper

part of the vagina.

Worldwide, cervical cancer is the fourth most

frequent cancer in women. In 2020 it had an incidence

of 604 000 new cases (Link 1). In the same year, there

were 342,000 deaths from cervical cancer, and about

90% of these occurred in low- and middle-income

countries. When women present the Human

Immunodeficiency Virus (HIV), they are six times

more likely to develop cervical cancer than women

without HIV, and an estimated 5% of all cervical

cancer cases are attributable to HIV. Moreover,

globally, HIV contributes to cervical cancer falls

disproportionately on younger women. In Mexico,

cervical cancer is the second leading cause of cancer

death in women (GM, 2015). An annual occurrence

of 13,960 cases in women is estimated, with an

incidence of 23.3 cases per 100,000 women. For

https://orcid.org/0000-0002-3535-1164

https://orcid.org/0000-0001-8179-0398

https://orcid.org/ 0000-0002-8626-5936

2019, the CC mortality rate in women aged 25 years

and over was 10,410 deaths per 100 thousand women.

Among the primary malignant tumours from which

women aged 25 and over die, CC is in second place,

with 13.2% of deaths from malignant tumours. By

age group, the mortality rate of cervical-uterine

cancer goes from 10.7 deaths per 100 thousand

women aged 40 to 49 years to 18.0 in women aged 50

to 59 years and 33.8 in women 60 and older. The

exams for the diagnosis of cervical cancer are (Cohen,

2019): pelvic examination, visualization of the cervix

and vaginal mucosa, cervical cytology, HPV test, and

colposcopy. All of these exams are invasive, and they

are intimately annoying for women. In this paper, we

propose Machine Learning algorithms to diagnose

cervical cancer based on data from questions to

women about some events throughout their life.

Literature Review

Table 1 shows the related work with our proposal.

The description column indicates the dataset used and

the algorithms applied to achieve the detection.

782

Acevedo, M., Acevedo, M. and Orantes, S.

Learning Algorithms for Cervical Cancer Detection.

DOI: 10.5220/0012047400003612

In Proceedings of the 3rd International Symposium on Automation, Information and Computing (ISAIC 2022), pages 782-787

ISBN: 978-989-758-622-4; ISSN: 2975-9463

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

Table 1. Related work to cervical cancer detection.

Year Title Author Description Results

2020

An Automatic Mass

Screening System for

Cervical Cancer

Detection Based on

Convolutional Neural

Network (Rehman, 2020)

Aziz-ur -Rehman,

Nabeel Ali, Imtiaz.A.

Taj, Muhammad Sajid ,

and Khasan S.

Karimov

Dataset: Cervical Cells images.

Algorithms: Softmax regression (SR), Support

vector machine (SVM), and GentleBoost

ensemble of decision trees (GEDT), and two

protocols: 2-class problem and 7-class problem.

2-class problem

SR 98.8%, SVM 99.5%,

GEDT 99.6%.

7-class problem

SR 97.21%, SVM 98.12%,

GEDT 98.85%.

2021

Automatic model for

cervical cancer screening

based on convolutional

neural network: a

retrospective,

multicohort, multicenter

study (Tan, 2021)

Xiangyu Tan1, Kexin

Li, Jiucheng Zhang,

Wenzhe Wang, Bian

Wu, Jian Wu, Xiaoping

Liand Xiaoyuan Huang

Dataset: 424,106 images from ThinPrep

cytologic test. Algorithm: Faster region

convolutional neural network.

Sensitivity: 99.4%,

Specificity: 34.8%.

Sensitivity for atypical

squamous cells of

undetermined significance:

89.3%; low-grade squamous

intraepithelial lesion: 71.5%,

and high-grade squamous

intraepithelial lesions:

73.9%.

2021

Classification of Cervical

Cancer Detection using

Machine Learning

Algorithms (Arora, 2021)

Aditya Arora, Anurag

Tripathi and Anupama

Bhan

Dataset: Herlev pap-smear image. Algorithms:

Active contour models for segmentation and

three types of Support Vector Machines for

classification: Polynomial order 4, Gaussian

RBF, and Quadratic.

Accuracy:

Polynomial SVM order 4:

95%, Gaussian RBF SVM:

85%, and Quadratic SVM -

85%.

2021

Classification of cervical

cancer using Deep

Learning Algorithm

(Tripathi, 2021)

Anurag Tripathi,

Aditya Arora, and

Anupama Bhan

Dataset: SIPAKMED pap-smear image.

Algorithms: ResNet50, ResNet-152, VGG-16,

VGG-19

Accuracy:

ResNet50: 93.87%, ResNet-

152: 94.89%, VGG-16:

92.85%, and VGG-19:

94.38%.

2021

DeepCervix: A deep

learning-based

framework for the

classification of cervical

cells using hybrid deep

feature fusion techniques

(

Rahaman, 2021

)

Md Mamunur

Rahamana, Chen Li,

Yudong Yao, Frank

Kulwa, Xiangchen Wu,

Xiaoyan Li, and Qian

Wang

Dataset: Herlev. Algorithm: DeepCervix, a

hybrid deep feature fusion technique with two

protocols: 2-class problem and 7-class problem.

Accuracy:

2-class problem: 98:32% and

7-class problem: 90:32%.

2021

Machine Learning

Assisted Cervical Cancer

Detection (Mehmood,

2021)

Mavra Mehmood,

Muhammad Rizwan,

Michal Gregus ml, and

Sidra Abbas

Dataset: Cervical cancer (Risk Factors) [11].

The same dataset was used in our proposal.

Algorithms: Random Forest (RF) for feature

selection, and an RF and shallow neural

networks combination as a

redictor.

Number of instances for:

Training: 70%, Validation:

15, and Test: 15. Accuracy =

93.6%

The papers from the first five rows used images as a

dataset, while the last work analysed the same dataset

used in this paper. In the Results column appears

standard classification metrics but mainly the

accuracy of the algorithms.

2 METHODS AND MATERIALS

In this section, the algorithms for cervical detection

will be described together with the data set. Also, we

will present the metrics to analyse the performance of

the algorithms.

2.1 Dataset

The dataset was collected at Hospital Universitario de

Caracas in Caracas, Venezuela (Link 2). It has 858

instances with 36 attributes. There are missing values

because some patients decided not to answer some

questions. In (Mehmood, 2021), the authors detected

the cases with missing values and decided to remove

them; they worked with 737 rows. In this paper, we

carried out the same task but worked with 672

instances from the original dataset. In Table 2, some

of the attributes of the dataset are shown.

We used the Biopsy (Boolean) attribute as the

target value. All the attributes with Boolean values

were converted to integers (0 and 1)

The data set is unbalanced, with 655 instances

from class 0 and 17 from class 1. Therefore, the

method SMOTE from Python was applied, resulting

in 655 attributes in class 0 and 655 attributes in class

1. So, finally, we have 1310 records.

Table 2: The attributes of the dataset analysed in this work.

Attribute Type

Age Integer

Number of sexual partners Integer

First sexual intercourse (age) Integer

Num of pregnancies Integer

Smokes Boolean

Hormonal Contraceptives Boolean

IUD (years) Integer

STDs (number) Integer

STDs:condylomatosis Boolean

Learning Algorithms for Cervical Cancer Detection

783

2.2 Artificial Neural Networks

An Artificial Neural Network (ANN) is a set of

interconnected neurons that emulate brain function. It

consists of one input layer, one output layer, and

hidden layers. The input layer receives the data of the

problem to resolve, for example, the value of

attributes or values of pixels in an image. The output

layer will yield the values for the solution of the

problem, either a class for classification or diagnosing

or a value for prediction. The hidden layers process

the data; at the end of the evaluation of each neuron,

a transfer function is applied. The training of the

network stops when minimal error is achieved. The

error decreases in each epoch through the gradient

descent algorithm. A fully interconnected, not

recurrent Multi-Layer Perceptron is the architecture

that will be applied in this paper.

2.3 K-Nearest Neighbours Algorithm

The k-nearest neighbor algorithm (Shai, 2014) is a

technique for classifying objects based on closest

training examples in the problem space. The k-nearest

neighbor algorithm is among the simplest of all

machine learning algorithms: similar things exist

nearby and are close to each other. An object is

classified by a majority vote of its neighbors, with the

object being assigned to the class most common

amongst its k nearest neighbors (k is a positive

integer). If k = 1, then the object is simply set to the

class of its nearest neighbor. Now, the algorithm will

be described. It will be assumed that the instance

domain, X, is endowed with a metric function p. This

is p: X x X → ℜ is a function that returns the distance

between any two elements of X. For example, if X =

ℜ

then p can be the Euclidean distance, as it is

shown in Equation (1).

𝑝



𝑥,𝑥





‖

𝑥−𝑥′

‖







𝑥



−𝑥′











(1)

2.4 Metrics

In this sub-section, we will present the metrics to

evaluate the performance of the classification

algorithms used in this paper. The metrics are

numerical and graphical.

2.4.1 Confusion Matrix

The confusion matrix or error matrix (Ariza, 2018)

shows the results of the predictions of the

classification algorithm. It is a table with different

combinations of n (the number of classes). Table 3

presents an example of a Confusion Matrix with two

different classes. In this Table, the number of

correctly classifications and the number of records in

which the algorithm confounded one class with the

other class can be observed. For example, in the

intersection of row 0 with column 1, the confusion

matrix shows how many records of class 0

confounded with class 1; this value is called false

positives. The intersection of row 0 and column 1 are

the true positives.

Table 3: Example of a Confusion Matrix of two classes.

Predicted values

Correct

values

Correctly classified

(

Incorrectly classified

(

)

Incorrectly classified

(

Correctly classified

(

)

From table 3, True Positive (tp) (Hossin, 2015) is the

number of the positive records that are correctly

classified; False negative (fn) is the number of

negative instances that are misclassified; False

positive (fp) is the number of misclassified positive

instances; finally, True Negative (tn) represents the

negative instances correctly classified.

2.4.2 Classification Metrics

There are important metrics to evaluate the

performance of classification algorithms defined in

the following equations. First, accuracy (Bateman,

2015) calculates the correct predictions given the total

number of predictions (Eq. (2)).

𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦=

𝑁𝑢𝑚𝑏𝑒𝑟𝑜𝑓𝑐𝑜𝑟𝑟𝑒𝑐𝑡 𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑖𝑜𝑛𝑠

𝑇𝑜𝑡𝑎𝑙𝑜𝑓𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑖𝑜𝑛𝑠





(2)

Precision is defined id defined in Eq. (3). This

metric indicates the number of positives predicted

positives.

𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛=

𝑡𝑝

𝑡𝑝 +

𝑓

𝑝

(3)

On the other hand, Eq. (4) defines the Recall,

which is the number of actual positives that are

correctly predicted as positive.

𝑅𝑒𝑐𝑎𝑙𝑙=

𝑡𝑝

𝑡𝑝 +

𝑓

𝑛

(4)

However, in some problems precision will have

higher priority than recall and vice versa. Therefore,

ISAIC 2022 - International Symposium on Automation, Information and Computing

784

there is a metric that combines recall and precision.

Eq. (4) defines F1-score.

𝐹1 − 𝑠𝑐𝑜𝑟𝑒= 2

𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑅𝑒𝑐𝑎𝑙𝑙

𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛+ 𝑅𝑒𝑐𝑎𝑙𝑙

(4)

This metric would take values between 0 and 1. A

value of zero means that both the precision and the

recall are zero, while if F1-score has a value of one,

both metrics will be equal to 1.

2.4.3 Receiver Operating Characteristic

(ROC) Curve

The ROC curve (Yang, 2017) represents the trade-off

between sensitivity and specificity. These two metrics

are inversely related; if one increases, the other

decreases. We define these metrics in Eq. (5), and Eq.

(6). Sensitivity (Kumar, 2011) or true positive rate

(TPR) is a conditional probability of correctly

identifying a disease.

𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦=

𝑡𝑝

𝑡𝑝 +

𝑓

𝑛

(5)

Specificity or true negative rate (TNR) is a

conditional probability of correctly identifying a

normal condition.

𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦=

𝑡𝑛

𝑡𝑛+

𝑓

𝑝

(6)

The Area Under the Curve (AUC) is an effective

measure that combines sensitivity and specificity to

validate the diagnostic test.

Figure 1: Example of ROC curve

Figure 1 depicts an example of a ROC curve. On

the Y-axis, we have the sensitivity, and the specificity

is on the X-axis. The best performance would be at a

point in the upper left corner, i.e., no false negative or

false positive. The diagonal is called a non-

discrimination line and is a random classification.

The point of a ROC random classifier will shift to the

position (0.5, 0.5). Points above the diagonal

represent good ranking results, and points below the

line for poor outcomes.

3 RESULTS

Two classification algorithms were applied to detect

cervical cancer: K-Nearest Neighbours (KNN) and

Multi-Layer Perceptron (MLP). The validation of the

algorithms was achieved by means of K-Fold Cross-

Validation and Hold-Out (80-20).

First, we present the results of K-NN algorithm.

The value of the K of the classification algorithm was

Knn = 1 (we used Knn to differentiate the K from K-

Fold Cross-Validation named as Kfcv) and Manhattan

distance. The value of Kfcv = 5. In Table 4, the

classification metrics for each iteration are shown.

Table 4: Classification metrics and Confusion matrices

from K-Fold Cross-Validation with Kfcv = 5, KNN

algorithm with Knn = 1, and Manhattan distance.

Table 4 shows that in iterations 1 and 3, the

algorithm confused just one instance, but the

precision and recall are equal to 0.99; that is the

reason F1-score = 1. This result means that the

performance of 1-NN algorithm is correct; also, the

accuracy is 1. Figure 2 shows de ROC curve for 1-

NN classifier. In this figure, it can be observed that

sensitivity and specificity had the value of 1, which

means that both classes were correctly classified.

Figure 2: ROC curve for 1-NN algorithm.

Table 5 shows the classification metrics from 1-

NN algorithm and Hold-Out (80-20) validation. We

can observe that the classifier confused 34 instances

belonging to class 0 as instances from class 1, which

is why a value of accuracy = 0.97. On the other hand,

all the cases from class 1 were correctly classified.

Iteration Precision Recall F1-score Accuracy Specificity Sensitivity

1 0.99 1 1 1 1 1

2 1 1 1 1 1 1

3 1 0.99

1 1 1

4 1 1 1 1 1 1

5 1 1 1 1 1 1

Learning Algorithms for Cervical Cancer Detection

785

Table 5: Results of the classification metrics with 1-NN

algorithm and Hold-Out validation.

The ROC curve from 1-NN algorithm with hold-

out validation is shown in Figure 3.

Figure 3: Results of ROC curve with 1-NN classifier and

the validation with Hold-Out.

In this case, specificity had a value of 1, but

sensitivity showed a value of 0.95. This behaviour is

reflected with some points over the x-axis line; the

line is not flat.

We present the results from the Multi-Layer

Perceptron with the same validation algorithms: Kfcv

= 5 and Hold-Out (80-20). The architecture of the

MLP is shown in Table 6.

Table 6: The architecture of the Multi-Layer Perceptron

proposed in this work.

Type of layer Number of neurons Activation function

Input 36 ReLu

Hidden 12 ReLu

Hidden 8 ReLu

Output 1 Sigmoidal

Table 7 shows the results from the Multi-Layer

Perceptron with K-Fold Cross-Validation. The same

results were obtained over the five iterations.

Table 7. Metrics from Multi-Layer Perceptron and the

validation algorithm K-Fold Cross-Validation.

Precision and recall are equal to 1, and sensitivity

and specificity had the same value, which can be

observed in Figure 4, which shows ROC curve.

Both lines, specificity and sensitivity, are flat

because the classification was correct for the two

classes.

Both lines, specificity and sensitivity, are flat

because the classification was correct for the two

classes.

Figure 4: ROC curve for MLP and K-Fold Cross-Validation

algorithm with Kfcv = 5.

In Table 8, we can observe the results from Multi-

Layer Perceptron. The algorithm confused 154

instances because it classified them as class 1 when

they belonged to class 0. That is the reason the

precision is less than 1. Therefore, sensitivity is equal

to 1, which is illustrated in Figure 5. However,

specificity is less than one, which is why the line on

the x-axis presents some blue points; namely, the line

is not flat.

Table 8: The classification metrics from Multi-Layer

Perceptron with the hold-out algorithm validation.

Figure 5: The ROC curve of MLP with Hold-Out algorithm

for validation.

The best results were obtained by the 1-NN

algorithm with the 100% of classification. Now, we

compare the results of our proposal with the results

from literature review.

From Table 9, we can observe that the sixth paper

[10] used the same dataset we used in this proposal.

The authors obtained the 93.6% of accuracy, and our

proposal showed accuracy in the range of 97% and

100%. Therefore, the parameters and the architecture

of the algorithms proposed in this paper overcome the

results from the work of Mehmood et al. The best

results come from the K-Fold Cross-Validation

algorithm.

4 CONCLUSIONS

A secondary alternative for malignant cervical

formation is to analyse some relevant risk factors

which are recorded in a questionary. The analysis of

Precision Recall F1-score Accurac

ecificit

Sensitivit

0.95 1 0.97 0.97 1 0.95

Precision Recall F1-score Accurac

ecificit

Sensitivit

1 1 1 1 1 1

Precision Recall F1-score Accurac

ecificit

Sensitivit

0.76 1 0.86 0.98 1 0.80

ISAIC 2022 - International Symposium on Automation, Information and Computing

786

these records was performed by means of two

algorithms of Machine Learning: K-NN and Multi-

Layer Perceptron. The results from both methods

reached 100% accuracy when they were validated

with K-Fold Cross-Validation algorithm, and they

accomplished 97 and 98% accuracy with Hold Out

Validation algorithm. The obtained results show that

these Machine Learning algorithms are suitable for

the analysis of risk factors because of their high

accuracy. Furthermore, even our lowest results are

greater than the results from [10] that used the same

dataset.

Table 9: Comparisons of the results from the related work.

Ref. Results

Rehman, 2020

2-class problem

SR 98.8%, SVM 99.5%, GEDT 99.6%.

7-class problem

SR 97.21%, SVM 98.12%, GEDT 98.85%.

Tan, 2021

Sensitivity: 99.4%, Specificity: 34.8%.

Sensitivity for atypical squamous cells of

undetermined significance: 89.3%; low-grade

squamous intraepithelial lesion: 71.5%, and high-

rade s

uamous intrae

ithelial lesions: 73.9%.

Arora, 2021

Accuracy:

Polynomial SVM order 4: 95%, Gaussian RBF

SVM: 85%, and Quadratic SVM -85%.

Tripathi, 2021

Accuracy:

ResNet50: 93.87%, ResNet-152: 94.89%, VGG-16:

92.85%, and VGG-19: 94.38%.

Rahaman, 2021

Accuracy:

2-class problem: 98:32% and 7-class problem:

90:32%.

Mehmood,

2021

Number of instances for: Training: 70%, Validation:

15, and Test: 15. Accurac

= 93.6%

This proposal

Validation algorithms: 5-Fold Cross-Validation and

Hold Out (80-20) Accuracy:

1-NN: 5-FCV = 100%, HO = 97%; MLP: 5-FCV =

100%, HO = 98%

ACKNOWLEDGEMENTS

The authors would like to thank the Instituto

Politécnico Nacional (COFAA, EDI, and SIP), the

CONACyT, and SNI for their support to develop this

work

REFERENCES

Ariza-Lopez, F., Rodríguez-Avi, J., Alba-Fernández, V.,

2018. Control Estricto de Matrices de Confusión por

Medio de Distribuciones Multinomiales. GeoFocus.

Revista Internacional de Ciencia y Tecnología de la

Información Geográfica. N. 21, p. 215-226.

Arora, A., Tripathi, A., Bhan, A., 2021. Classification of

Cervical Cancer Detection using Machine Learning

Algorithms. 2021 6th International Conference on

Inventive Computation Technologies (ICICT)

Bateman, B., Jha, R., Johnston, B., Mathur, I., 2020. The

The Supervised Learning Workshop: A New,

Interactive Approach to Understanding Supervised

Learning Algorithms. Packt Publishing Ltd.

Cohen, P., Jhingran, A., Oaknin, A., Denny, L., 2019.

Cervical Cancer. Lancet, 393: 169–82.

Gobierno de México, Secretaría de Salud, Estadísticas de

Cáncer de Mama y Cáncer Cérvico Uterino, 2015.

https://www.gob.mx/salud/acciones-y-

programas/informacion-

estadistica#:~:text=En%20M%C3%A9xico%20tambi

%C3%A9n%20a%20partir,35.4%20casos%20por%20

100%2C000%20mujeres.

Hossin, M., and Sulaiman, M., 2015. A Review on

Evaluation Metrics for Data Classification Evaluations.

International Journal of Data Mining & Knowledge

Management Process (IJDKP), Vol.5 (2).

INSTITUTO NACIONAL DE LAS MUJERES, SISTEMA

DE INDICADORES DE GÉNERO, CÁNCER DE

MAMA Y CERVICO-UTERINO, 2021,

http://estadistica.inmujeres.gob.mx/formas/tarjetas/ca

ma_cacu.pdf

Kumar, R., Indrayan, A., 2011. Receiver Operating

Characteristic (ROC) Curve for Medical Researchers.

Indian Pediatrics, Vol. 48.

Link 1 https://www.who.int/es/news-room/fact-

sheets/detail/cervical-cancer

Link 2

https://archive.ics.uci.edu/ml/datasets/Cervical+cancer

+%28Risk+Factors%29

Mehmood, M., Rizwan, M., Gregus, M., Abbas, S., 2021.

Machine Learning Assisted Cervical Cancer Detection.

Frontiers in Public Health.

Rahaman, M., Li, C., Yao, Y., Kulwa, F., Wu, X., Li, X.,

Wang, Q., 2021. DeepCervix: A Deep Learning-based

Framework for the Classification of Cervical Cells

Using Hybrid Deep Feature Fusion Techniques.

Computers in Biology and Medicine, Vol. 136.

Rehman, A., Ali. N., Taj, I., Sajid, M., Karimov, K., 2020.

An Automatic Mass Screening System for Cervical

Cancer Detection Based on Convolutional Neural

Network. Mathematical Problems in Engineering, Vol.

2020.

Shai, S., and Shai, B., 2014. Understanding Machine

Learning: From Theory to Algorithms. Cambridge

University Press.

Tan, X., Li, K., Zhang, J., Wang, W., Wu, B., Wu, J., Li,

X., Huang, X., 2021. Automatic model for cervical

cancer screening based on convolutional neural

network: a retrospective, multicohort, multicenter

study. Cancer Cell Int, 21(1):35.

Tripathi, A., Arora, A., Bhan, A., 2021. Classification of

cervical cancer using Deep Learning Algorithm. 2021

5th International Conference on Intelligent Computing

and Control Systems (ICICCS)

Yang, S., Berdine, G., 2017. The receiver operating

characteristic (ROC) curve. The Southwest Respiratory

and Critical Care Chronicles, Vol 5(19).

Learning Algorithms for Cervical Cancer Detection

787