A New Perspective on the Treatment of Ovarian Cancer: Deep

Learning Algorithms-Based Prediction

Yan Yan

Intelligent Medical Engineering, Nankai University, Tianjin, China

Keywords: Deep Learning, Ovarian Cancer, Medical Diagnosis, Precision Oncology.

Abstract: Ovarian cancer is one of the most common gynaecological cancers worldwide, because the initial symptoms

are not obvious, they are often discovered in the late stages, resulting in poor treatment effects and poor

prognosis. This paper explores various applications of deep learning (DL) in early detection and personalized

treatment of ovarian cancer. The construction process of the deep learning model including data collection,

preprocessing, model training and evaluation is introduced in detail. In the application of Convolutional

Neural Network (CNN), this paper discusses how to reduce and learn parameters through image enhancement

and feature mapping, and how to use SoftMax and cross-entropy loss functions in the later stages of data

preprocessing and classification to improve the recognition accuracy of the model. For Long Short Term

Memory (LSTM), this paper analyses its significant advantages in handling irregular time series data,

especially in handling missing value patterns and complex time dependencies. In addition, this study also

explored the clinical challenges of DL models when dealing with ovarian cancer-related issues, such as the

“black box” nature of decision-making, generalization capabilities, and privacy issues of sensitive data. The

comprehensive results indicate that the DL method will become an effective way to advance the development

of the oncology field.

1 INTRODUCTION

Ovarian cancer is one of the most common

gynecological cancers worldwide, its mortality rate

ranks first among the three major malignant tumors

of the reproductive system, so it is called the “silent

killer” (Wang, 2019). It is difficult to detect in the

early stages of the disease, and there is no effective

and reliable screening method for early-stage ovarian

cancer, therefore clinical surgical treatment,

radiotherapy, and chemotherapy have poor efficacy,

has seriously threatened women's health and lives.

Ovarian cancer occurs in the ovarian tissue deep

in the pelvic cavity. Early symptoms are vague and

easily misdiagnosed. More than 75% of ovarian

cancer patients are diagnosed in stages III-IV (late

stage). When diagnosed, patients already have typical

symptoms such as pelvic pain, abdominal distension,

abdominal distension, or loss of appetite. At this time,

the tumor cells have spread, and the five-year relative

survival rate is only 29% (Gao, 2022). Mortality rates

https://orcid.org/0009-0003-9945-1230

from ovarian cancer are particularly concerning, so it

is crucial to find effective treatment strategies and

diagnostic methods, and thus more effective

monitoring indicators and interventions.

In recent years, the application of deep learning

has become more and more widespread and has

demonstrated its unique advantages in various fields

(Liu, 2021; Qiu, 2022; Zhao, 2023), which is

particularly significant in medical applications. Deep

learning, through its complex multi-level computing

structure, ability to deeply analyze image data and

capture its core features (Thompson, 2018). Based on

this, deep learning can learn these features and make

more intelligent, accurate and personalized decisions

for image analysis, early diagnosis, and drug

treatment. As a model that is efficient, accurate, and

does not completely rely on doctors’ subjective

judgment, deep learning shows great potential in the

diagnosis and treatment of ovarian cancer.

The combination of deep learning and ovarian

cancer research is currently showing huge advantages.

Yan, Y.

A New Perspective on the Treatment of Ovarian Cancer: Deep Learning Algorithms-Based Prediction.

DOI: 10.5220/0012939300004508

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence (EMITI 2024), pages 417-421

ISBN: 978-989-758-713-9

417

Based on the advantages of artificial intelligence and

machine learning algorithms, deep learning

algorithms can independently mine key features from

medical images. It then identifies and differentiates

potential pathological variations that are difficult to

detect by clinicians, greatly improving the accuracy

and efficiency of early diagnosis in the process

(Sadeghi, 2024). This facilitates earlier and more

accurate diagnosis. This model is more efficient and

accurate than traditional methods such as ultrasound,

Computed Tomography (CT), and Magnetic

Resonance Imaging (MRI), greatly improving the

efficiency of tumor detection and feature analysis.

Taken together, deep learning, as a transformative

technology, can assist clinicians in making more

informed decisions, thereby promoting progress in

the field of ovarian cancer treatment, and ultimately

improving patients' treatment effects and quality of

life (Sadeghi, 2024).

The core goal of this study is to deepen the

understanding of the application of deep learning in

the field of ovarian cancer and explore new

possibilities for intelligent treatment. It will provide

an in-depth exploration of how deep learning methods

can improve diagnostic efficiency, improve the

efficiency and accuracy of medical imaging, and

provide a basis for identifying new therapeutic targets.

It also provides new perspectives and strategies for

the diagnosis and treatment of ovarian cancer, brings

new hope to patients, and provides experience and

methods for the research of other types of tumors.

This review provides an in-depth exploration of deep

learning methods in the field of ovarian cancer

(Section 2), including key steps such as data

collection, preprocessing, and model training, and

introduces different models used in ovarian cancer

research through examples, including convolutional

neural networks (CNN), long short-term memory

networks (LSTM) and other technologies. Next, in

Section 3, this paper discusses the limitations of

existing research in deep learning research on ovarian

cancer and provides solutions to these problems.

Finally, in Section 4, this paper summarizes the main

contents of this review and provide ideas for future

research. This article highlights the unique potential

of deep learning in the field of ovarian cancer and

explores its application prospects in clinical practice.

2 METHODS

2.1 Framework of Deep

Learning-Based Ovarian Cancer

Prediction

Deep learning processes complex data by simulating

the working mechanism of the human brain. In

ovarian cancer prediction research, it can

autonomously learn from pathological images,

genetic information, and patients’ electronic medical

records to identify key disease indicators.

The effectiveness of deep learning technology

depends largely on the quality and quantity of data.

Ideally, high-quality data should include ovarian

cancer images and corresponding clinical information

from different perspectives and times. At the same

time, in real life, different ethnic groups and different

types of lesions will affect the adaptability of the

model, so the diversity of data is particularly

important. In published studies, multiple research

groups have used different data collection methods to

obtain data for ovarian cancer prediction. For instance,

Gao et al. (Gao, 2022) extracted data from the

ultrasound image database of 10 hospitals in their

study. They classified the data according to whether

women were single or not and used rectal ultrasound

and transvaginal ultrasound images separately;

Thomas et al. (Buddenkotte, 2023) used 451 scan data

from four different institutions and two countries.

Although the sources of these data are very wide,

considering that most of the data (380 samples) come

from the United Kingdom, they chose to use this part

of the data for model training and evaluation.

In the process of establishing and evaluating deep

learning models, data preprocessing, model selection,

training, evaluation, and testing are closely related. In

the data preprocessing stage, the data is processed to

adapt to the requirements of the model, including

standardization, denoising and enhancement of

images, and processing of missing data; for non-

image data, it needs to be converted into a structured

format and normalized. Depending on the

characteristics of the data and the prediction task,

select an appropriate deep learning model, such as a

CNN for image recognition, a LSTM for sequence

data. In the model training stage, a large amount of

labeled data is used to adjust the weights of the

network to minimize the gap between the predicted

output and the actual output, and appropriate loss

functions and optimizers are used. Finally, the

performance of the model is evaluated on an

independent test set, and the generalization ability of

the model is verified through indicators such as

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

418

accuracy, recall, and F1 scores, and is visualized

through qualitative analysis such as Gradient-

weighted Class Activation Mapping (Grad-CAM)

technology for model decision-making process.

Finally, the developed deep learning model can be

deployed in clinical settings to assist doctors in more

accurate diagnosis and treatment planning. When

deployed, ensure that the model can integrate

seamlessly with existing healthcare record systems

and maintain accuracy and efficiency while

protecting patient privacy.

2.2 CNN-Based Ovarian Cancer

Prediction

Among deep learning methods, CNN, is one of the

commonly used architectures. Convolutional neural

network (CNN) is a very broad structure in deep

learning technology. CNN consists of different types

of processing layers, including convolutional layers,

fully connected layers (FC), and pooling layers.

Among them, the convolution layer generates an

output feature map through a 2D convolution kernel

operation with the input feature map, which is the

core part of CNN. The pooling layer downsamples the

feature map after the convolutional layer and selects

the average or maximum value to summarize the

characteristics of each block. Finally, the fully

connected layer performs the classification task,

which is similar to traditional artificial neural

networks (ANN) (Banerjee, 2019; Ghoniem, 2020).

The rapid development of deep learning models such

as CNN is of great significance in the field of medical

diagnosis and has gradually shown outstanding

potential (Ghoniem, 2021).

In previous studies, many scholars have studied

the application of CNN-based methods in ovarian

cancer prediction. For example, Ziyambe et al.

(Ziyambe, 2023) proposed a novel CNN algorithm

for prediction and diagnosis of ovarian cancer. This

study used a dataset consisting of 200 images of

serum ovarian cancer and non-cancer samples,

extended to 11,040 images by augmentation methods

for use with deep learning architectures. After

preprocessing the data set, it is processed using a

convolution operation. After preprocessing the data,

convolution operation is used for preliminary

processing. The data is then further processed through

the ReLU nonlinear activation function, and on this

basis, the pooling layer is used for data simplification.

Through this process, the size of the feature map and

the number of future learning parameters are reduced.

The data is then flattened and transformed into a one-

dimensional vector in preparation for data input to the

fully connected layer. In the final stage, the author

uses SoftMax as the classifier and cross-entropy as

the loss function. After training, the CNN model

showed an accuracy of up to 94% and was able to

identify cancer samples with an accuracy of 95.12%,

it can also distinguish healthy cells with an accuracy

of 93.02%.

2.3 LSTM-Based Ovarian Cancer

Prediction

Long short-term memory network is an extension of

the basic recurrent neural network (RNN) and aims to

solve the gradient disappearance and gradient

explosion problems of traditional RNN (Ghoniem,

2020; Datta, 2020). The LSTM network introduces

memory units (cells) and gating mechanisms to

effectively process time series data. In this process, it

is necessary to control the memory unit (c) of the long

short-term memory network (LSTM) by using a set

of gate control networks, including the forget gate (f),

the input gate (i) and the output gate (o)

(Ghoniem,2021). This built-in gating mechanism can

also avoid computational problems of vanishing or

exploding gradients. It can be adapted to handle

irregular time series data with missing values,

inconsistent time intervals and complex time

dependencies. Based on these advantages of LSTM

and being very suitable for TM data containing

missing data and irregular intervals between sequence

tests, this method has become an ideal algorithm for

processing actual data sets (Wu,2022).

Researchers such as Wu et al. (Wu,2022) applied

the LSTM model to an incomplete large-scale TM

data set to create a tool for predicting cancer risk,

which is undoubtedly a pioneering act. This tool

adopts simple missing value processing strategies

including directly filling gaps with 0 or KNN, MICE

interpolation technology, etc. The design includes a

hidden layer configured with 100 units. On this basis,

a standard S-shaped activation function was used, and

a 10-round training cycle was carried out. This study

highlights the potential ability of the LSTM model to

handle irregular time series problems including

missing values, varying time intervals, and complex

dependencies. It also avoids the limitations of

gradient disappearance or excessive growth. Through

single-point verification of TM test data, the model

showed higher accuracy than single threshold

analysis, with an AUROC value of 0.831. And in

another model, by analyzing data at up to four time

points, the AUROC value increased to 0.931. This

series of tools greatly improves the efficiency of the

A New Perspective on the Treatment of Ovarian Cancer: Deep Learning Algorithms-Based Prediction

419

further screening process for individuals and helps to

detect potential tumor risks at an early stage.

3 DISCUSSIONS

3.1 Limitations and Challenges

1) Transparency and Explainability: The potential

of deep learning models to improve the accuracy of

ovarian cancer diagnosis cannot be ignored, however,

their use in clinical decision-making processes has

been severely limited, in part due to the so-called

“black box” problem. This means that the model’s

decision-making process is opaque, although the

model can make predictions by learning patterns in

large amounts of data, its internal logic is invisible,

which is a fatal problem in cancer diagnosis. Medical

decision-making requires a high degree of accuracy

and interpretability, if doctors and patients have

difficulty understanding how the model derives a

specific diagnosis from the input data, trust in the

model will be reduced, this limits the application of

deep learning tools in clinical practice. 2)

Limitations on Generalization Ability: Ideally, a

good model can not only perform well on the training

set, but also show good prediction capabilities on

unseen data sets, which is the generalization ability of

the model. In practical applications, deep learning

models often have difficulty achieving this. For

ovarian cancer, the data comes from a wide range of

sources, including different populations, geographic

locations, and devices, so the data may vary

significantly, the diversity of data, differences in data

quality between different devices, and the complexity

of the model itself make it challenging for the model

to maintain performance on new data. Such problems

may cause the model to overfit the distribution of

training data and fail to fully cover all potential

situations. 3) Private issues: Deep learning models

require large amounts of data to train and optimize. In

the medical field, this means collecting and analyzing

sensitive information such as patients ’ personal

records, pathology images, and genetic information.

As a gynecological disease, ovarian cancer will be

more serious. Therefore，if not handled properly, it

may result in unauthorized privacy disclosure. In

addition, existing regulations and laws may not be

able to keep up with the development of science and

technology, making privacy protection measures

insufficient and leading to more serious

consequences. For healthcare organizations, data

breaches can result in reputational damage, financial

losses, and legal liability. For individuals, it may lead

to social discrimination, insurance fraud and even

security risks. Such leaks may also erode public trust

in the field of smart healthcare and affect the further

development of future research projects.

3.2 Solution Strategy

For interpretability, by utilizing techniques such as

Shapely Additive Explanations (SHAP) and Local

Interpretable Model-agnostic Explanations (LIME) to

quantify the contribution of each input feature to

model prediction, it can provide more intuitive use of

the model. In addition, visualizing networks in key

regions using activation maps and Class Activation

Mapping (CAM) techniques is particularly useful for

medical image analysis; For limitations on

generalization capabilities, data enhancement and

transfer learning can be used to increase the diversity

and breadth of model training. Furthermore,

regularization techniques can reduce overfitting,

thereby improving model performance on unseen

data; As a distributed machine learning method,

federated learning allows models to be trained on

local devices, which means that only the weights and

biases of the model are shared during the training

process, rather than the training data itself. This is an

effective method to reduce data transmission and

protect data privacy, especially suitable for handling

sensitive medical information in diseases such as

ovarian cancer. With the continuous advancement of

deep learning technology and the deepening of

artificial intelligence in the medical field, in the

future, the application of deep learning in early

diagnosis of ovarian cancer, treatment plan

formulation, and patient prognosis assessment will

become more and more extensive and accurate.

4 CONCLUSIONS

This review highlights the challenges posed by

ovarian cancer as a common gynecological

malignancy and the revolutionary role of deep

learning in early diagnosis and treatment. In it, this

paper elaborates on the architecture of deep learning

in ovarian cancer diagnosis and discuss the process of

training and validating models on different datasets.

Additionally, the application of deep learning

techniques was explored, specifically CNN and

LSTM, to ovarian cancer prediction and diagnosis,

and how these models can learn from medical images

and genetic data and assist clinical decision-making.

Although deep learning models achieve remarkable

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

420

accuracy and performance in experiments, this paper

also identify and discuss the limitations of these

methods in real-world applications. At the same time,

it can be foreseen that with the maturity and

advancement of explainable artificial intelligence

technology and the strengthening of privacy

protection measures, the application of deep learning

in the field of ovarian cancer will become more

extensive and efficient. Future work will need to

focus on more reliable model evaluation tools and

algorithms for handling diverse and irregular data sets.

REFERENCES

Banerjee, I., Ling, Y., Chen, M. C., Hasan, S. A., Langlotz,

C. P., Moradzadeh, N., ... & Lungren, M. P. 2019.

Comparative effectiveness of convolutional neural

network (CNN) and recurrent neural network (RNN)

architectures for radiology text report classification.

Artificial intelligence in medicine, 97, 79-88.

Buddenkotte, T., Rundo, L., Woitek, R., Escudero Sanchez,

L., Beer, L., Crispin-Ortuzar, M., ... & Schönlieb, C. B.

2023. Deep learning-based segmentation of multisite

disease in ovarian cancer. European radiology

experimental, 7(1), 77.

Datta, S., Si, Y., Rodriguez, L., Shooshan, S. E., Demner-

Fushman, D., & Roberts, K. 2020. Understanding

spatial language in radiology: Representation

framework, annotation, and spatial relation extraction

from chest X-ray reports using deep learning. Journal

of biomedical informatics, 108, 103473.

Gao, Y., Zeng, S., Xu, X., Li, H., Yao, S., Song, K., ... &

Gao, Q. 2022. Deep learning-enabled pelvic ultrasound

images for accurate diagnosis of ovarian cancer in

China: a retrospective, multicentre, diagnostic study.

The Lancet Digital Health, 4(3), e179-e187.

Ghoniem, R. M. 2020. A novel bio-inspired deep learning

approach for liver cancer diagnosis. Information, 11(2),

80.

Ghoniem, R. M., Algarni, A. D., Refky, B., & Ewees, A. A.

2021. Multi-modal evolutionary deep learning model

for ovarian cancer diagnosis. Symmetry, 13(4), 643.

Liu, Y., Liu, L., Yang, L., Hao, L. and Bao, Y., 2021.

Measuring distance using ultra-wideband radio

technology enhanced by extreme gradient boosting

decision tree (XGBoost). Automation in

Construction, 126, p.103678.

Qiu, Y., Wang, J., Jin, Z., Chen, H., Zhang, M., & Guo, L.

2022. Pose-guided matching based on deep learning for

assessing quality of action on rehabilitation

training. Biomedical Signal Processing and Control, 72,

103323.

Sadeghi, M. H., Sina, S., Omidi, H., Farshchitabrizi, A. H.,

& Alavi, M. 2024. Deep learning in ovarian cancer

diagnosis: a comprehensive review of various imaging

modalities. Polish Journal of Radiology, 89, e30.

Thompson, R. F., Valdes, G., Fuller, C. D., Carpenter, C.

M., Morin, O., Aneja, S., ... & Thomas Jr, C. R. 2018.

Artificial intelligence in radiation oncology: a

specialty-wide disruptive transformation?.

Radiotherapy and Oncology, 129(3), 421-426.

Wang, S., Liu, Z., Rong, Y., Zhou, B., Bai, Y., Wei, W., ...

& Tian, J. 2019. Deep learning provides a new

computed tomography-based prognostic biomarker for

recurrence prediction in high-grade serous ovarian

cancer. Radiotherapy and Oncology, 132, 171-177.

Wu, X., Wang, H. Y., Shi, P., Sun, R., Wang, X., Luo, Z., ...

& Wang, Y. 2022. Long short-term memory model–a

deep learning approach for medical data with

irregularity in cancer predication with tumor markers.

Computers in Biology and Medicine, 144, 105362.

Zhao, F., Yu, F., Trull, T., & Shang, Y. 2023. A new

method using LLMs for keypoints generation in

qualitative data analysis. In 2023 IEEE Conference on

Artificial Intelligence (CAI) (pp. 333-334). IEEE.

Ziyambe, B., Yahya, A., Mushiri, T., Tariq, M. U., Abbas,

Q., Babar, M., ... & Jabbar, S. 2023. A deep learning

framework for the prediction and diagnosis of ovarian

cancer in pre-and post-menopausal women.

Diagnostics, 13(10), 1703.

A New Perspective on the Treatment of Ovarian Cancer: Deep Learning Algorithms-Based Prediction

421